Technology & IT Business: AI-era businesses one person can start

🧭 From idea to first paying client

This section is about small technology businesses that one person, or two or three, can start with a laptop and AI tools: setting up automation for a dental office, running Microsoft 365 for a ten-person firm, teaching a team to use AI safely, checking a town’s meeting minutes, editing a podcast. None of them needs investors or an office. All of them need the same few things done in the right order: a clear offer, a handful of real conversations, a first paying client, and the basic legal and money setup.

This guide is the path from idea to first paying client, and the map of the rest of the section. Read it first, then pick a game plan.

People who already work with computers and want their own business. An office manager who became the "AI person" at work. An IT technician tired of working for a large provider. A bookkeeper who automated half her job. You already have the hardest part, which is knowing how a real business runs day to day.

People changing careers. A teacher, nurse, paralegal or tradesperson who knows an industry’s problems from the inside and has learned the tools. Industry knowledge plus AI skills is a stronger combination than AI skills alone, because you can speak the customer’s language.

College students and recent graduates. Several ideas fit around a class schedule. Students have extra rules to know first (visa limits, taxes, campus policies, academic integrity), and a separate guide covers them: For college students and recent grads. If you are on an F-1 visa, read it before you take any paid work.

Who it is not for. If you are looking for passive income, or a business that runs itself while you sleep, this is the wrong section. Every idea here is a service: you show up, solve a problem, and keep solving it. The money comes from being reliable, not from a clever tool.

The timing is reasonable. In the Census Bureau’s survey published October 22, 2026, 24.6% of U.S. employer businesses said they used AI in some business function, up from 17.3% in the first reading of that series in December 2025, and firms with 1 to 4 employees were at 24.3%. But in a Census study of firms using AI, 64% had made no changes to how they work, and among firms with no plans to use it, 22% cited not knowing what AI could do. Owning a tool and putting it to work are different things. The gap between them is the opening for a small service business.

There are more than 50 ideas in this section. Most people waste weeks comparing them. A faster way is to test each candidate against three questions and pick the one where all three answers are yes.

Skills: can you do it well within a month or two? Not "could I learn it someday," but "could I deliver a small paid project by next month." If you have run an office, M365 admin or workflow automation is close. If you have edited video, AI-assisted video is close. If you speak two languages fluently, translation review is close.

Customers: can you reach 50 of them? Can you name a type of business that has this problem and reach at least 50 of them within an hour’s drive or through one association, software community or network? People you already know count most.

Money: do they already pay for this problem? The best sign is that they already spend money on it, on staff time, a vendor or lost revenue. A dental office that loses money on insurance surprises pays. A hobby club that would "love a chatbot" usually does not.

Some matches that tend to work:

Each idea page marks whether demand is growing, steady or crowded. Crowded ideas are real, but AI or built-in software is pushing the basic version toward free, so you need a way to stay above it. If you are unsure, pick a growing idea that matches your background over a crowded one that sounds exciting.

The order matters more than the speed. Each of the seven steps below protects you from a common, expensive mistake. The figure groups steps 3 and 4 because they happen in the same two weeks.

1. Hold ten conversations before you build anything (days 1 to 14). Talk to ten people who own or run the kind of business you want to serve. Not a survey, not a pitch: a 20-minute conversation about how their work runs. Start with people you know, then ask each for one introduction. You are listening for a problem that comes up again and again, that already costs them money, and that they have tried to fix. If three or more of the ten describe the same pain without prompting, you have something.

Questions for the ten conversations. What takes up the most time in your week that you wish someone else did? Walk me through the last time that happened, step by step. What does it cost you, in hours or lost customers? What have you tried to fix it? Why didn’t that work? If someone fixed it, what would that be worth to you per month? Who else do you know who has the same problem?

Write down their exact words. You will use them in your offer, your website and your talks. Do not ask "would you buy this?" People say yes to be polite. Ask about the past, not about a hypothetical future.

2. Write one offer and sell a paid pilot (days 10 to 21). Turn the most common problem into one offer with a name, a scope, a price and a time frame: "Missed-call setup for HVAC companies: connected, tested and running in 10 business days, $750." Offer it as a paid pilot to the two or three people who were most interested, at a fair but lower first-client price, with permission to measure the before and after and to write a case study. Paid pilots get taken seriously; free ones stall. The pricing guide shows how to set the number, and the launch and marketing playbook has the scripts.

3. Do the legal and money basics (days 14 to 30). Decide whether to start as a sole proprietor or form an LLC (Choosing a legal structure). Get a free EIN from the IRS, open a separate business bank account, and check your city or county for a general business license. Get quotes for professional liability (errors and omissions) insurance and cyber liability insurance; if a mistake in your setup costs a client money or exposes their data, these are what respond (Business insurance). Use a simple written contract for every client, even friends; the contracts and AI rules guide lists what it must say. Start a bookkeeping habit on day one (Bookkeeping basics) and set aside part of every payment for taxes (Small business taxes).

4. Buy a small set of tools, not a stack (days 14 to 30). One paid AI assistant, one automation tool or the admin tools your service needs, a business email on your own domain, and a password manager. That can come in well under $100 a month. Clients should pay for their own software in their own names, so their data stays theirs. The AI toolkit guide has prices and a starter stack.

5. Land and deliver the first three clients (days 21 to 60). Deliver the pilots carefully. Measure the before number in week one. Send a short weekly update. Keep a person checking anything AI writes to a customer, any number it quotes and anything that touches money or health data. Meanwhile keep your outreach going: a few warm emails and one partner conversation every week, so a third client is lined up when the first two finish.

6. Turn projects into care plans (days 45 to 75). One-time projects pay once. A monthly care plan covers monitoring, fixes, updates and one small improvement a month, and it is what makes a slow month survivable. Offer it at the end of every pilot, priced in the proposal from the start so it is not a surprise. Software changes, logins expire and AI models get updated, so there is always real work to do.

7. Collect proof and referrals (days 60 to 90). Write a one-page case study for each finished pilot with the measured result, ask for an honest review and ask for one introduction to someone with the same problem. Three case studies in one niche will do more for you than any amount of advertising.

A realistic picture at day 90: two to four paying clients, one or two on care plans, a handful of referral partners, and a clear sense of whether the niche works. Some people get there faster; many take longer, especially part time. Income varies widely and this guide makes no promise about it. If you have had 20 or more real conversations and nobody will pay for a small pilot, the problem is the niche or the offer, not your effort. Change one of them.

Courses that sell an "AI agency" income story. The FTC has sued sellers of AI-powered business opportunities over earnings claims, including the Operation AI Comply cases in 2024 and a 2026 proposed settlement with Air AI that would ban it from selling business opportunities. Be wary of anyone who sells you a course, a template or a software license with screenshots of monthly revenue. Free vendor courses and the help organizations in the free help directory teach the skills without the pitch.

Done-for-you AI stores and passive-income schemes. Several of the FTC’s 2024 AI cases were about exactly these. They are left out of this section on purpose.

Outbound AI cold-calling. The FCC ruled on February 8, 2024 that AI-generated voices count as "artificial or prerecorded voice" under the TCPA, so outbound calls using them need the called person’s prior consent. Selling or running AI cold-calling for clients is a legal trap. Inbound AI answering, where a customer calls the business, is a different situation and is covered in the AI receptionist idea.

Essay mills and doing graded work for pay. Writing essays, doing problem sets or taking tests for students is contract cheating, banned by colleges and illegal in several states, including under California Education Code 66400. Using AI to do it faster does not change that. Honest tutoring is fine; see tutoring that uses AI the honest way.

Fake reviews and review generators. The FTC’s rule on consumer reviews and testimonials, in effect since October 2024, bans fake reviews, including AI-written ones that misrepresent a reviewer’s experience. Do not sell them or use them for yourself.

Promises you cannot back up. "Saves 20 hours a week," "99% accurate," "replaces your receptionist." Measure real results for real clients and quote those, with context.

Client data in the wrong tools. Free consumer AI accounts have different data terms than business plans. Health data needs vendors that will sign a HIPAA business associate agreement, and tax-return data has its own federal rule (IRC 7216). Check before any client data goes into any tool.

The section has three layers. The idea pages tell you what a specific business is and what it pays. The game plans walk through five common business models in full. The running-the-business guides cover the parts every model shares.

A good reading order for a new owner: this guide, then the game plan closest to your idea, then marketing, pricing and contracts, then the idea page itself. You can read the rest as questions come up.

Survey figures (Census Bureau, October 2026) and the FTC and FCC rules mentioned here were checked in November 2026. Business registration fees and license rules vary by state and city; check yours before you file.

🤖 An AI automation studio

🎯 The business: a small AI automation studio, usually one person at first, that does three things for local firms. It connects the apps they already pay for so routine work moves on its own. It sets up assistants that answer from the firm’s own documents. And it trains staff to use AI tools on their real work. You sell a paid review first, then a build, then a monthly care plan that keeps it all running.

This is the fullest of the five game plans and the one to compare the others against. The idea page AI workflow automation for local businesses covers the core skill; this guide turns it into a studio with three related offers and a plan for the first ten clients.

It suits someone who likes untangling how work actually gets done. Former office managers, operations staff, bookkeepers, IT generalists and anyone who builds careful spreadsheets tend to do well. Most of the job is asking good questions, mapping steps and testing. Coding helps later but is not required to start.

It does not suit someone who wants to build alone and never talk to owners, or who wants passive income. Automations break when an app changes. Clients pay you to notice before they do, and to pick up the phone when they don’t.

The timing is good for a plain reason. In the Census Bureau’s October 2026 survey, 24.6% of U.S. employer businesses used AI, and the smallest firms were close to that average. But in the Bureau’s study of AI users (published April 2026), 64% had changed nothing about how they work. A Goldman Sachs survey of its small-business alumni (March 2026) found 76% used AI but only 14% had fully built it into their work, and 73% said they would benefit from more training. Setup, connection and training are the gap. That gap is this studio.

A ladder is a set of offers where each one makes the next one easy to buy. Owners rarely say yes to a $4,000 project from a stranger. They will say yes to a short talk, then to a small paid review, and then to the build the review recommends.

Free talk. A 30 to 45 minute session at a chamber of commerce, a referral group, a library business night or a staff lunch. Show three real before-and-after flows on screen, take questions, and end by offering a paid review. The talk is marketing, so keep it free and keep it short.

Paid review, $300 to $500. You sit with the owner or office manager for about two hours, map one process step by step, and hand back a one-page plan with three fixes and their prices. Credit the fee toward the build. Charging for it filters out people who only want free advice.

First build, $750 to $5,000. A single flow (for example web form to CRM to a drafted reply) at the low end, or a set of three to five connected flows at the high end. A person approves anything AI writes to a customer, quotes a price or touches money.

Care plan, $200 to $500 a month. Monitoring, fixes when an app changes, and one small improvement a month. Tell the client what you did each month in a short note.

Side offers. Two other offers sell to the same buyer at any step. A half-day team workshop (see AI training workshops for small teams) often turns up the processes worth automating. A document assistant that answers from the firm’s own files (see Company chatbots) sells well once a build has earned trust, but only if you test it for wrong answers first (Chatbot accuracy and safety testing). A short AI-use policy (AI readiness and AI-use policy consulting) is a natural add-on for firms with client data.

Why the ladder matters now: in its second-quarter 2026 results, Upwork said AI adoption had cut client activity on contracts of $500 and below, while AI strategy and consulting work grew more than 50%. Small one-off tasks are being absorbed by the tools. Ongoing, tailored work is not.

Clients pay for their own automation and AI subscriptions, in their own name. That keeps their data in their accounts and means the work keeps running if you part ways. Your own subscriptions are for learning and demos. The full tool list and what each plan costs is in Your AI toolkit and what it costs; the legal set-up is in Choosing a legal structure and Business insurance.

Days 1 to 30: learn, set up, choose. Pick one automation tool and build ten flows for yourself: a form that fills a sheet and sends a reply, invoice reminders, meeting notes that become a task list. Register the business, open a bank account, and get insurance quotes. Then choose one kind of customer, such as property managers, roofing companies or physical therapy clinics, and list 40 of them within a short drive. One trade is easier to sell to than "small businesses", and every client teaches you something you reuse.

Days 31 to 60: talk to people. Send the 40 emails, call the ones who open or reply, and book two short talks. Record two-minute screen videos of your own flows with fake data, run your first paid reviews, and write down the problems owners describe in their own words. Those phrases become your website.

Days 61 to 90: build and keep. Deliver the first builds. Test with fake data, then real data, then watch for two weeks. Hand over a one-page guide for each flow: what it does, which accounts it uses, how to switch it off. Offer the care plan on the day you hand over, not a month later.

Where you should be on day 90: eight or more real conversations, three or more paid reviews, one or two builds, and at least one care plan. Fewer than that is a signal, covered under "when to pivot" below.

These are example prices for a U.S. operator selling locally, drawn from agency and freelancer price guides and the research for this section. For contrast, Fiverr’s own cost guide puts simple no-code automations at $40 to $250 and simple rule-based chatbots at $50 to $150. That is where not to compete.

Three habits protect your margin. Take half of a build fee up front. Include the first month of care in every build, so the care plan feels like a continuation, not a new sale. And price by the time the work saves, not your hours: a flow that saves an office manager five hours a week is worth more than the eight hours it took to build. More in Pricing AI and IT services.

Never promise a percentage of savings, an accuracy rate or "fully automated" results. The FTC has acted against AI sellers over claims like these, including a 2025 order against a company that claimed its AI detector was 98% accurate without proof.

A worked month. Here is one example month, about five months in, for a solo operator. Three care plans at $300 bring in $900. One single-flow build brings $1,200. One half-day workshop for a 12-person office brings $900, and one paid process review brings $400. That is $3,400.

Costs are small: $80 for your own software, $60 set aside for insurance, $40 of AI and test credits, and $70 for marketing and printing, $250 in all. That leaves about $3,150 before tax for roughly 25 hours a week, sales included.

Most months in the first half year will be smaller than this. Set aside about a quarter to a third of profit for income and self-employment tax; see Small business taxes.

Who first. Your first three clients should come from people who already trust you: a former employer, a family member’s business, your own dentist, accountant or landlord. Do the first project at a fair discount in exchange for a written case study with real numbers, such as minutes saved a week or how fast new leads now get a reply.

The plan, with numbers. Over about eight weeks: send 40 short personal emails to firms in your chosen trade, follow up by phone with the 15 who opened or replied, give two free talks, and ask three referral partners (a bookkeeper, an IT provider and a web designer) for one introduction each. Expect 8 to 12 conversations, 4 to 6 paid reviews and 2 to 4 first builds. The rest of the ten usually come from those clients’ referrals over the next few months.

Here is an email you can adapt. Change the first line to something you have actually seen in that trade.

Hi Dana, I run a small automation studio here in Springfield. Property managers I talk to get maintenance requests by email, text and phone, and someone retypes each one into the work-order system. I can set up the software you already pay for so every request lands in one place, sorted by urgency, with a draft reply your team approves before it goes out. It usually takes about two weeks. Would a 20-minute call next Tuesday or Wednesday be useful? If it isn’t a fit, I’ll tell you. Thanks, Chris Patel, 555-0164

Channels, best first, for this studio:

Turn each client into proof. Two weeks after a flow goes live, send the owner a short before-and-after note with the numbers and ask whether you may share it without their customers’ details. Record a 60-second screen video with the data blurred. Then ask one question: "Who else do you know who fights the same problem?" The full playbook, with more scripts, is in Launch and marketing playbook.

The routine matters more than the tools. Check every client’s run history for failures on Monday morning, before clients notice. Keep each client’s automations in accounts the client owns, with you added as a user. Test new flows with fake data. Sample AI-written outputs every week, because model updates can change how they behave.

Data rules shape the build. If a client is a clinic or dental office, you will handle protected health information, which makes you a HIPAA business associate: you need a signed business associate agreement (BAA), and every AI or cloud tool that sees that data must offer one too. Business texts need carrier registration (A2P 10DLC), and outbound calls with AI voices need prior consent under the FCC’s 2024 ruling. Inbound answering is a different, lower-risk case. The contract terms you need, including who owns AI output under the Copyright Office’s 2025 report, are in Contracts, client data and AI rules.

Free help first. SCORE mentors and your local Small Business Development Center help with pricing, contracts and a first plan at no cost. Zapier, Make and n8n all run free courses and community forums, and Zapier and Make list certified builders in partner directories, which can bring leads once you qualify. The full list is in Free help directory.

The first person to bring in is usually another builder you can hand overflow to, paid per project. Sign an NDA with them, give them access only to the accounts they need, and tell the client who is working on their systems. The second is often a part-time assistant for scheduling, invoices and the Monday checks.

A rule of thumb for when: bring in help when care plans take more than half your week, or when you turn down a build twice in one month. Whether a helper is a contractor or an employee is decided by IRS and state tests, not by what your agreement calls them, so check before you set up payment.

The main risks. Simple single-step automations are now cheap on marketplaces, so stay on multi-step work tied to a client’s real software. The apps themselves keep adding AI features, which can make one of your builds unnecessary overnight. A flow can fail silently for two weeks and lose leads, which is why monitoring is the product. An AI step can send a customer something wrong, so keep a person approving customer-facing messages until the flow has earned trust. Fixed-price jobs can grow ("while you’re in there..."), so write down what is included. And one large client can become half your income, which makes their slow month your slow month.

When to change course. These are rules of thumb, not data:

If you reach day 90 with fewer than eight real conversations, change who you are calling, not your price. Try a different trade or a warmer channel.

If you have conversations but no paid reviews, your offer is too vague. Name one job and one price, such as "new-lead replies in five minutes, $1,200".

If you sell builds but no care plans, include the first three months of care in the build price and make cancelling the active choice.

If a vendor ships your main flow as a built-in feature, move to configuring that feature and to the work that crosses several apps, or lean harder into training.

If one industry keeps showing up and its rules keep coming up, consider the narrower plan in An industry AI desk. And if after six months you have no repeat client and you dread every sales call, stop. The skills you built are hireable in operations and IT roles, and that is a fine outcome.

Checked in November 2026: Census AI-use figures (October 2026 survey), Upwork and Fiverr demand data, Fiverr price guides, and tool prices on the vendors’ pages (Make and ChatGPT prices were not confirmed there). Re-check tool prices and AI vendors’ data terms before you quote a client.

🖥️ IT and AI help for small offices

🎯 The business: you are the IT person for small offices that don’t have one. You run their Microsoft 365 or Google Workspace accounts, lock down sign-ins and email, set up backups that actually restore, train staff to spot AI-written scams, and roll out Copilot or Gemini without exposing files nobody meant to share. You start with fixed-price checkups and cleanups, move to a monthly per-user plan, and grow toward full managed IT.

AI makes this trade busier, not smaller. Every office now has more accounts, more devices and more AI tools to configure, and phishing emails no longer give themselves away with bad grammar. The idea pages behind this plan are Microsoft 365 and Google Workspace admin and Managed IT for small offices.

It suits someone who already fixes things for family and coworkers, and likes it. Help desk staff, school and nonprofit IT volunteers, and people with CompTIA A+, Network+ or Security+ study behind them have a head start. The core skills are accounts and sign-in security, email security (the SPF, DKIM and DMARC records that stop others from sending as you), sharing permissions, backups and patient explaining.

It does not suit someone who needs predictable hours. Offices call when something breaks, and something breaks on the morning of a deadline. If you cannot be reachable during business hours, sell projects only and say so plainly.

The demand is in the surveys. In Bluevine’s April 2026 survey of 942 small-business owners, data security and privacy was the top worry about AI, at 33%, up from 23% a year earlier. In Goldman Sachs’ March 2026 survey of its small-business alumni, the biggest AI challenges were data privacy and security (50%) and lack of technical expertise (49%). Those owners need a person, not another app.

Checkup, $300 to $750. A 90-minute review of the office’s accounts, sign-in settings, email security records, sharing, devices and backups, followed by a one-page report: what is fine, what is risky, and three fixes in order with prices. Let the client keep the report whether or not they hire you further.

Cleanup, $1,000 to $3,000. Turn on multi-factor sign-in for everyone, set up the email security records, remove old accounts, tidy who can see which folders, fix licenses the office pays for but doesn’t use, and write down how it all works. This is a project with a clear end.

Admin plan, $15 to $40 per user a month. New staff set up and old staff removed the same day, updates checked, settings kept tidy, a monthly note on what you did, and help by email or phone in business hours.

Add-ons sold to the same office: security awareness training with phishing tests (see Security awareness training), tested backups (see Cloud backup and ransomware recovery), and an AI rollout.

The AI rollout job. Copilot in Microsoft 365 and Gemini in Google Workspace work with the files each person can already open. In an office where "everyone can see everything", that means an assistant can surface salary sheets or client files to anyone who asks the right question. So a rollout starts with the sharing cleanup, then licenses and data settings, then a short staff session on using the tools well (AI training workshops for small teams) and a one-page AI-use policy (AI-use policy consulting). For budgeting, Google lists Gemini Enterprise Business from $21 a seat a month; secondary sources put Microsoft’s Copilot Business at about $21 a user a month list. Check Microsoft’s current price before you quote it.

Managed IT, $100 to $250 per user a month. Full support and security for the whole office, usually with monitoring tools on every device. One pricing guide reported an average of about $145 per user a month from a survey of 412 providers and buyers. This is where the plan leads after a year or two, once you have tools, insurance and someone to cover when you are sick.

Insurance is the line not to skip. You will hold admin access to other people’s businesses, and a mistake or a breach can cost them money. Full managed IT later needs more: monitoring and ticketing software, a security stack and higher insurance limits, about $2,000 to $10,000 by the research estimate for this section. Don’t buy that before you have monthly clients to pay for it. See Business insurance.

Days 1 to 30: get ready. Set up your own business tenant and practice on it: multi-factor sign-in, email security records, sharing rules, a backup and a test restore. Register the business, buy insurance, and get a services contract that includes a written authorization to access client systems. Write your checkup as a checklist, so every office gets the same review.

Days 20 to 45: test the checkup. Run it free for two friendly offices, such as a relative’s business or a nonprofit you know. Time it, fix the checklist, and turn the results into a sample report with the names removed.

Days 30 to 80: reach out. Contact 30 offices of 5 to 25 people and four referral partners (details below). Give one talk on AI-written phishing and voice-clone scams. Book paid checkups from about day 45.

Days 55 to 90: deliver and convert. Turn checkups into cleanups, and cleanups into monthly plans. On day 90, a good position is six or more conversations, three or more paid checkups, two cleanups and one or two offices on a plan.

Per-user pricing is how offices already think about software, so it is easy to understand, and it grows with the client. Write down what a plan covers and what it doesn’t. Home devices, the owner’s personal phone and a new office build-out are usually separate. Set a minimum, such as five users, so a two-person office doesn’t take the same time as a ten-person one for a fraction of the fee. More in Pricing AI and IT services.

A worked month. Here is an example month for three offices with 8, 10 and 12 staff, 30 users in all. Admin plans at $35 per user bring $1,050. Phishing training at $3 per user brings $90. Backups for 20 computers at $15 bring $300. One cleanup project for a new office brings $1,800. That is $3,240.

Costs: about $260 for resold tools (backup storage, the training platform and monitoring), $120 set aside for insurance, $60 for your own software and $60 for marketing, $500 in all. That leaves about $2,740 before tax.

The recurring part, $1,440 a month, is small at this stage, but it grows with every office you add and rarely leaves. Offices tend to keep an IT person who answers the phone. Set aside a share of profit for taxes; see Small business taxes.

Who first. Offices whose owners already know you: your accountant, insurance agent, dentist, a former employer, a relative’s business, a place of worship’s office. They will let you in to run a checkup because they trust you, not because of your website.

The plan, with numbers. List 30 offices of 5 to 25 people in two or three kinds of business, such as accounting, law, clinics and nonprofits. Over about seven weeks: send 30 short letters or emails, call the 12 most likely, give one free talk on AI-written phishing and voice-clone scams at a chamber or library, and meet four referral partners. Expect 6 to 10 conversations, 3 to 5 paid checkups, 2 or 3 cleanups and 1 or 2 monthly plans by day 90. The rest of the ten come from referrals and the partners.

Hi Ms. Reyes, I’m an IT consultant here in Lakewood and I work with offices of 5 to 25 people. Fake invoices and login pages are much harder to spot now that AI writes them, and most of the fixes are settings an office already owns: sign-in protection, email security records and who can see which folders. I offer a 90-minute checkup for $450 with a one-page written report, and the report is yours whether or not you hire me for anything else. Would a weekday morning in the next two weeks work? Thanks, Alex Kim, 555-0129

Channels, best first, for small-office IT:

Turn clients into proof. After a cleanup, show the owner the before-and-after list in plain words: "9 of 12 accounts had no second sign-in step; now all 12 do." Ask for a Google review and permission to describe the work without the firm’s name. Ask each office manager which other offices they talk to. The full playbook is in Launch and marketing playbook.

Access. Get written authorization before you touch any client system. Keep every client’s admin logins in a password manager, with a separate vault per client and multi-factor sign-in on your own account. Never keep client passwords in email, texts or notes. Use your own named admin account in each tenant, not a shared one, so every change is traceable.

Records. For each client: a list of users, devices, licenses and key settings, a dated change log, and a short "how this office works" page that someone else could pick up if you were sick. This is also what you hand over if a client leaves, which makes clients more willing to sign.

Routines. A new-hire and leaver checklist, run the same day. A monthly check of sign-in alerts, backups (with a test restore each quarter) and updates. A one-paragraph monthly note to each client. A single inbox or simple ticket form for requests, so nothing lives only in your text messages.

Tools. Start with the admin centers Microsoft and Google already provide, a password manager and a reputable backup service. Add remote monitoring and ticketing software when you have enough monthly users to cover its per-device cost. Both Microsoft and Google run partner and reseller programs; join once you resell licenses. The full tool list is in Your AI toolkit and what it costs, and the contract terms (data, liability, termination) are in Contracts, client data and AI rules. If a client is a clinic, you will likely be a HIPAA business associate and need a signed BAA before you start.

Free help. CISA, the federal cybersecurity agency, publishes free guidance written for small businesses that you can use in checkups and talks. Microsoft and Google both publish free admin training. SCORE mentors and your local Small Business Development Center help with pricing and contracts. CompTIA certifications give clients and insurers something to check. The list is in Free help directory.

An incident partner. Decide before it happens who you call if a client is hit by ransomware or an account takeover that is beyond you. Ask your own insurer, and each client’s cyber insurer, what incident help their policy includes. Write the name and number into each client’s records.

The first hire. Usually a part-time technician for on-site visits and new-hire setups, often a student from a campus help desk program (see Campus IT help desk). Give them their own accounts with only the access they need. A rule of thumb: hire when you have about 60 to 80 users on plans or when you miss a day’s requests because you were on site elsewhere. Check whether they are an employee or a contractor under IRS and state tests before you pay them.

The main risks. A breach on your watch, even one you could not have stopped, lands on your reputation; written scope, insurance and a record of what you recommended protect you. Being the only person who can fix things makes holidays and sick days hard. Scope creeps to the owner’s home printer and their children’s tablets; price it or say no. Tool costs per device can arrive before the revenue does. And Microsoft and Google change their admin settings often, so plan time each month to keep up.

When to change course (rules of thumb):

If offices like the checkups but won’t commit to monthly plans, sell projects plus prepaid blocks of hours. That is a fine business too.

If you dislike being on call, drop the monthly plans and specialize in projects: checkups, cleanups, backups and Copilot or Gemini rollouts.

If your best days are the training sessions, move toward the studio in An AI automation studio, which leans on workshops and automation.

If most of your calls come from older home users, not offices, that is its own business: Tech help for seniors and home users.

And if demand outgrows you, hire a second technician and move toward managed IT pricing, or take an MSP job where someone else carries the on-call phone. The experience is worth the same either way.

Checked in November 2026: the Bluevine and Goldman Sachs surveys, price ranges from provider and training-vendor guides, and Gemini prices on Google’s page. Microsoft Copilot prices came from secondary sources; confirm them with Microsoft before quoting. Re-check AI add-on prices each quarter.

🏥 An industry AI desk

🎯 The business: you pick one industry that lives under a rule, such as dental offices and HIPAA, small CPA firms and tax-return privacy, or town clerks and accessibility deadlines, and you run one painful job for its offices every week. AI does the reading and drafting. You check every item against the source, a licensed professional makes the decisions, and the paperwork the rule requires is built into how you work. The rule is what keeps owners from doing it themselves, and it is a large part of what they pay you for.

This plan is narrower than An AI automation studio and usually pays better per hour once it is running, because you know one industry’s work and rules better than any generalist. It also asks more of you on day one: training, insurance and contracts come before the first client file.

It suits someone who has worked in or near the industry: a former dental front-desk coordinator, a bookkeeper who has lived through tax season, a paralegal, a deputy clerk, an insurance customer-service rep. You already know the words, the software and the bad days. That knowledge is worth more than any AI skill, and it is why owners will trust you.

It does not suit someone who finds rules tedious, or who wants to work on many different kinds of problems. The job is the same task done carefully, hundreds of times, with a written record. It is also not a part-time college job; most of these desks need a steady weekday presence and client data a student should not be holding.

Each row below is an idea page in this section. The "you never" column is the licensing line. Crossing it is how a helpful desk becomes unauthorized practice, or a privacy violation.

Choose with three questions. Do you already know this industry’s work? Is there a deadline or a rule that makes owners act now? And can you reach 30 of these offices within an hour’s drive or through one association? If one industry gets three yeses, stop looking.

How compliance becomes the product. Software vendors sell these industries a portal and leave the rest to the office. You sell the finished job, with the rule built in: the signed BAA before any patient data moves, the 7216 consent step the CPA firm decides it needs before return data reaches any outside tool, the accessibility check before a PDF is posted, the script that hands every coverage question to a licensed agent. Put those steps in writing, show them to the owner, and keep a log. That record is what a careful owner is buying, and it is what a general AI vendor rarely offers.

HIPAA in plain words, because most readers start with a health desk. If you handle protected health information for a practice, you are its business associate. You need a signed BAA with every practice, and a BAA with every AI or cloud tool that sees that data. Encrypting the data does not remove the need for one. Consumer ChatGPT or Claude accounts do not come with a BAA, and Medical Economics reported in July 2026 that pasting patient information into a consumer chatbot without one is an impermissible disclosure. Use only business or API plans from vendors that will sign, and check each vendor’s current terms. Have a healthcare lawyer review your BAA and service agreement once; it is the best few hundred dollars you will spend.

The rest of this plan uses the dental verification desk as the worked example, because it has the clearest pricing. The ladder is the same shape for every industry.

Paid trial, $150 to $300. For one week, you verify the next day’s patients and hand over a clean list of benefits and problems by mid-afternoon. A BAA is signed before the trial, not after.

Desk plan, $300 to $800 a month per location. Every scheduled patient checked ahead of the visit, breakdowns entered into the practice software, and a named person the office can call. For reference, one outsourcing vendor’s published tiers run $250 to $825 a month per location by volume, outsourced checks are priced at about $1 to $8 each on vendor price pages, and competitors’ marketing puts an in-house coordinator at $2,000 or more a month.

Next office. Office managers talk to each other. A referral from a happy practice is your best lead, and the second location of a group practice is often the easiest sale you will make.

Second desk. Once one desk runs smoothly, add the neighbor task, such as prior-authorization packets for specialists, or hire someone to run the first desk while you start another industry.

For a tax-season or town desk, swap the HIPAA lines for the industry’s own requirements, such as a 7216 consent template reviewed by a lawyer, or an accessibility checker. See Business insurance and Choosing a legal structure.

Days 1 to 25: choose and listen. Pick the industry, read its rule in the original wording, and talk to ten owners or office managers. Ask what job they hate, what they have tried, and what they would pay to have it gone. Write down their words.

Days 10 to 50: get ready. Take the training, buy insurance, and have your agreement and BAA (or consent form) reviewed. Build your workflow on test data: the checklist, the checking step, the log and the handoff. Time yourself so you know what a location really takes.

Days 30 to 80: sell. Reach 30 offices and three or four partners (below). Run paid trials from about day 45.

Days 60 to 90: deliver. Convert trials into monthly desks. On day 90, a good position is six or more conversations, two or more paid trials and at least one monthly desk.

Price by the location or the season, not the hour. Owners compare you with a staff member or a software subscription, and a flat monthly fee is easy to set beside either. Raise prices for new clients once you have a waiting list. More in Pricing AI and IT services.

A worked month. Here is an example month for a dental desk with four locations. Three are on a $600 plan, which brings $1,800. One smaller practice pays $450. One new practice runs a $200 paid trial. That is $2,450.

Costs: $80 for AI and document tools on BAA-covered plans, $30 for a phone and secure fax line, $110 set aside for insurance and $30 for marketing, $250 in all. That leaves about $2,200 before tax.

A desk grows one location at a time, and locations rarely leave if the work is accurate. The weak point is concentration: if one group practice is half your income, its decision to bring the work in-house is your bad year. Aim for no single client above a third of revenue. Taxes: see Small business taxes.

Who first. Your own dentist, CPA or agent, and any office you have worked in or with. Then the people who already serve the industry and hear its complaints: for dental, supply reps, practice-management consultants, billing companies and the IT providers who look after dental offices.

The plan, with numbers. List 30 practices within 30 minutes of you. Over about seven weeks: email all 30, drop a one-page sheet at 15 of them in person in the mid-afternoon (never during the morning rush), call 10 to follow up, and meet three partners. Join or attend the local dental society or the industry’s own association meetings where vendors are welcome. Expect 6 to 8 real conversations, 2 to 4 paid trials and 1 to 3 monthly desks by day 90.

Hi Tanya, I run a small insurance verification desk for dental offices in the North County area. For each location I check every scheduled patient’s benefits the afternoon before, enter the breakdown into your practice software, and flag anything that needs a conversation at the front desk. I sign a business associate agreement before I see any patient information, and I only use tools whose vendors sign one too. Would a one-week paid trial be useful, so you can see the work before deciding anything? I’m happy to stop by after your morning rush. Thanks, Renee Alvarez, 555-0173

Channels, best first, for an industry desk:

Turn clients into proof. After a month, show the office manager what changed in their own numbers: patients checked, problems caught before the visit, front-desk hours freed. Ask permission to describe the result without names, and ask whom they know at other practices. Never share any patient detail in a case study. More scripts are in Launch and marketing playbook.

The compliance file. For each client: the signed agreement and BAA (or consent forms), the list of tools that touch their data with each vendor’s BAA or terms, who on your side has access, and a dated log of work and any problems. Re-check vendors’ terms every quarter. Keep a written plan for what you do if data goes to the wrong place, including how fast you tell the client; your BAA will set the deadline.

Daily routine. Work inside the client’s own systems and portals with a login in your name, not a shared one. Use only the minimum information the task needs. Keep patient or client data out of email, texts and any consumer AI account. Check every AI-drafted item against the source before it goes into the client’s system, and record what you checked.

Tools. A business or API plan from an AI vendor that will sign a BAA, if your industry needs one; a password manager; an encrypted laptop; a secure fax or document-sharing service the practice approves. The tool list and data terms are in Your AI toolkit and what it costs, and the contract details (BAAs, 7216, liability) are in Contracts, client data and AI rules.

Getting help. The HHS Office for Civil Rights publishes HIPAA guidance for business associates. A healthcare or industry lawyer should review your templates once. SCORE mentors and Small Business Development Centers help with pricing and planning for free. The industry’s own association is often the best teacher of its rules. See Free help directory.

The first hire. Someone who has done the job inside an office, such as an experienced verification coordinator. Train them on the rule before they get access, have them sign a confidentiality agreement, give them their own logins, and add them to each client’s access list. Hire when you are turning down locations, not before.

The main risks. A privacy incident is the big one; it can end the business, which is why the BAA, the tool list and the log come first. A wrong benefits breakdown can lead to a patient’s surprise bill, so record what the portal or insurer said and when. Software vendors keep bundling the same task (in dental, products such as Weave and NexHealth), so sell running the software, not replacing it. Rules change: HHS has proposed a HIPAA Security Rule update that is not final, a lawsuit challenges the Title II extension, and Colorado’s law takes effect in 2027. Seasonal desks, such as tax season, leave months empty unless you pair them with another desk.

When to change course (rules of thumb):

If offices say "we already bought software", offer to run that software for them. That is often the better business.

If owners don’t care about the rule, you may have the wrong buyer (try the office manager, not the owner) or the wrong industry.

If the core task gets cheap, move to the harder part: the exceptions, denials and follow-ups the tools leave behind.

If you find the rules more stressful than interesting, move to An AI automation studio and serve industries without regulated data. And if after six months no office has signed a monthly desk, take what you learned into a job inside the industry; experienced operators who understand AI tools are in demand.

Checked in November 2026: HIPAA business associate rules as summarized by HHS guidance and trade press, CMS-0057-F decision clocks, IRC 7216 and 6713, the ADA Title II deadlines after the April 2026 interim rule, and dental verification prices from vendor pages. Not legal advice. Re-check the Title II challenge, the HIPAA Security Rule proposal and Colorado’s rules before relying on them.

🎬 An AI-assisted media studio

🎯 The business: a small media studio that makes short video, podcasts and social posts for local businesses and small creators, with AI doing the slow parts: rough cuts, captions, audio cleanup and first-draft show notes. You sell what the tools can’t do on their own: picking the moment worth sharing, filming real people on site, shipping on time every week, and answering the audience in the client’s voice.

This is the most crowded of the five game plans, and this guide says so up front. The idea pages behind it are Short-form video editing, Podcast production and Social media for local shops. Read the next two sections before you spend anything.

It suits someone who already edits for fun, notices why one clip works and another doesn’t, and doesn’t mind deadlines every week. Film, media, communications and marketing students do well, and so do people who ran a club’s or employer’s social accounts. Being comfortable walking into a shop with a phone and a small light matters as much as editing skill.

It does not suit someone who wants to sit alone and cut footage for strangers at any price. That part of the market is where AI and global marketplaces push prices down hardest.

Be clear about what is now cheap or free. Auto-clipping apps turn a long video into dozens of captioned clips in minutes. Editors remove filler words and clean up audio with one click. Chat assistants write captions and show notes. Image and video generators make stock-style visuals. If your offer is one of those steps, you are competing with a button.

The evidence is mixed, not bleak. Upwork’s 2026 skills report found U.S. freelancer earnings for AI video generation and editing up 329% in 2025, and said image and video tools had not reduced demand for video editing. Fiverr reported AI-related searches in its Video and Animation category up 278% (November 2025 to April 2026 against the prior six months). People are buying more video. They are paying less for the mechanical steps and more for judgment.

What still sells, and what to build your offer around:

Story judgment. Choosing the 40 seconds of a one-hour episode that a listener will send to a friend, and cutting the first three seconds so people stop scrolling. Tools pick the loudest moment; you pick the best one.

Being there. Filming the baker at 5 a.m., the mechanic explaining a repair, the regulars at the counter. AI can’t stand in a shop, and local owners trust the person who shows up.

Showing up every week. Most small accounts and shows die from missed weeks, not bad edits. Reliability is a product.

Community and voice. Replying to comments and messages the way the owner would, and knowing when to hand a complaint to them.

Rights and honesty. Licensed music, release forms for people on camera, no fake reviews, clear labels on paid posts. Owners don’t know these rules, and getting them wrong costs them.

Free sample. Three short clips from a client’s own episode or video, or three short videos filmed at a shop you already buy from. Keep it small: two or three hours of work.

Per piece, $75 to $500. One edited and published podcast episode, a set of clips from one long video, or one shoot. Podcast price guides put episodes at about $75 to $500 depending on length and what’s included.

Monthly, $300 to $1,800. A weekly show produced end to end (guides put weekly shows at about $400 to $1,800 a month), or a local shop’s social package with on-site filming and regular posts (freelancer guides: about $300 to $1,500 a month).

Retainer, $500 to $3,000 a month. Planning, filming, posting, replies and a monthly results note. Agency guides put small-business social management at about $500 to $3,000 a month. Few clients reach this rung in the first year; one or two who do will carry the studio.

Buy gear after a client needs it, not before. A phone, a clip-on microphone and window light make most local content. See Choosing a legal structure for the setup.

Days 1 to 30: pick a niche and make samples. Choose one kind of client, such as business podcasts in one field, or restaurants and shops in one neighborhood. Make three samples that show your judgment: a before-and-after edit, a clip with a strong opening, a short shop video filmed on site. Set up the business, a simple contract that covers music and permissions, and a music library.

Days 25 to 80: reach out. Send 50 personal messages (plan below) and visit 10 shops. Sell first pieces from about day 35.

Days 55 to 90: turn pieces into packages. Offer a monthly package to every client whose first piece went well. From about day 70, send each client a short results note: what you made, what got watched or shared, what you will try next.

Day 90 check: ten or more replies, three or more paid pieces and one or two monthly clients is a good start.

For comparison, Upwork’s published rate ranges for video editors run $10 to $60 an hour worldwide, and its beginner range for social media managers starts at $14 an hour. Local clients pay more because you show up, know the area and answer the phone. Price by the month, include a set number of pieces and revisions, and charge for extra rounds. More in Pricing AI and IT services.

A worked month. Here is an example month with four clients. Two weekly podcasts at $600 a month bring $1,200. Two local shops on a $500 social package bring $1,000. One half-day shoot for a new client brings $400. That is $2,600.

Costs: $80 for software and AI tools, $15 for music, $10 for cloud storage, $35 for mileage and $40 for marketing, $180 in all. That leaves about $2,420 before tax for roughly 20 hours a week.

Most studios run below this for the first months. Keep a reserve for the month two clients cancel at once, and set aside money for taxes; see Small business taxes.

Who first. Shops you already buy from, creators you know, a podcast run by someone in your network, your employer’s or your place of worship’s channels. People who know you will let you film, and they forgive the first rough edges.

The plan, with numbers. List 25 podcasts in one field that publish on a steady schedule but have rough audio or no clips, and 25 local shops in one area with weak or stale accounts. Send all 50 a short personal message. Make a free three-clip sample for the 10 best fits, and visit 10 shops in person at a quiet hour. Expect 8 to 12 replies, 4 to 6 paid first pieces and 2 to 4 monthly clients. Referrals from those clients fill the rest of the ten.

Hi Jordan, I listen to Main Street Money, and I cut three short clips from your episode with the bakery owner, captioned, as a sample. They’re in this folder, and you’re welcome to use them either way. I edit and publish weekly shows for small-business podcasts: audio cleanup, show notes, and three clips an episode chosen for the moment people will share. If it would help, I could take next week’s episode as a paid trial for $150. Thanks, Sam Ortiz, 555-0158

Channels, best first, for a small media studio:

Turn clients into proof. Keep a before-and-after for every client, with permission. Use the client’s own numbers in your results notes, never borrowed ones. Ask happy clients for a short written review and an introduction to one other owner. The full playbook is in Launch and marketing playbook.

Workflow. One folder per client with raw footage, finished pieces, the brand kit (colors, fonts, logo), the music licenses used and signed release forms. A posting calendar the client can see. An approval step before anything is published. A written rule for how long you keep raw footage. A monthly results note.

Tools. A desktop editor (several good ones have free versions), an AI tool for captions and rough clips, a scheduler, and a licensed music library. ElevenLabs’ Starter plan ($6 a month) includes a commercial license; its free plan does not. Midjourney allows commercial use on paid plans, but companies with more than $1 million in yearly revenue must be on its Pro or Mega plan. Canva’s prices were not confirmed on its own page in November 2026; check before you budget. The full list is in Your AI toolkit and what it costs.

Rules that come up in media work. Use music licensed for business use; personal-account music often isn’t. Get written permission from anyone you film, and from the owner for staff and customers. Paid or gifted posts need a clear disclosure under FTC rules. Never write or post fake reviews or testimonials; the FTC’s 2024 rule bans them, including AI-written ones. Never clone a real person’s voice without written consent: Tennessee’s ELVIS Act protects voice likeness, and SAG-AFTRA terms require consent for digital replicas. Tell clients that raw AI-generated material may not be protected by copyright; the Copyright Office’s 2025 report says prompts alone are not enough. If you serve clients in the EU, its AI Act has required labels on deepfakes since August 2026. Contract wording is in Contracts, client data and AI rules.

Getting help. SCORE mentors and Small Business Development Centers help with pricing and contracts for free; the list is in Free help directory. Editing communities and the tool makers’ own tutorials are the fastest way to learn new features. The first person to bring in is usually a second editor paid per piece, or a student filmmaker for shoots. Give them access only to the client folders they need, and check contractor rules before you pay them.

The main risks. Prices for basic editing keep falling as tools improve. Social clients leave when they can’t see results. Posting for several clients every day burns people out. A rights mistake, such as unlicensed music or a customer filmed without permission, can get a client’s account flagged or worse. And one client can become most of your income.

When to change course (rules of thumb):

If clip-only work is stuck at marketplace prices, move up to full show production or on-site filming, where you can’t be replaced by a button.

If shop social clients keep cancelling after three months, try work with clearer results: podcasts for businesses, Local search and Google Business Profile care, or Email and text marketing automation.

If you enjoy teaching owners to make their own content more than making it for them, turn that into workshops (see An AI automation studio).

And if after six months you have no monthly client, take the portfolio into an in-house content or editing job. Many studios start as a way to get hired, and that counts as a win.

Checked in November 2026: Upwork and Fiverr demand reports, price ranges from podcast, freelancer and agency guides, and the ElevenLabs and Midjourney plan terms on their own pages. Canva prices were not confirmed. Re-check tool licenses before each new client, because they change.

🎓 A student tech side business

🎯 The business: a part-time tech side business that fits around classes. You tutor with AI the honest way, teach older adults to use AI safely and spot AI scams, build small websites for clubs and youth leagues, or help local shops set up AI tools through a campus clinic. If you are an international student, a paid campus tech job is the version that keeps you inside your visa rules. You start small, keep it legal and honest, and come out with income, references and a portfolio.

This plan is written for college students and recent graduates. The rules that apply to you (visa limits, taxes, financial aid and campus policies) are covered briefly here and in full in For college students and recent grads. Read that page before you take your first dollar.

It suits a student who explains things well, is patient, and can protect their own calendar. Education, computer science, communications, nursing and business students do well, and the work looks good on any resume. You don’t need to be a top coder. Most of these lanes pay for clear teaching and reliability.

It does not suit a semester when your grades are already slipping, or anyone hoping for big money fast. This is modest income with real experience attached. A good month in the middle of a semester might bring in a little over a thousand dollars, and exam weeks will bring in close to nothing.

F-1 visa. International students on F-1 visas may work on campus up to 20 hours a week while school is in session, and full time during breaks (ICE guidance, updated December 2025). On-campus means a job with your school or an on-campus employer, not your own business. Freelancing for outside clients, tutoring for cash, or paid off-campus teaching generally needs work authorization such as CPT or OPT, and many schools say self-employment does not qualify under CPT or before graduation under OPT. Unauthorized work can end your status. DHS published a rule on July 17, 2026 that would end "duration of status" for students; NAFSA and several law firms report that a federal court in Massachusetts blocked it on September 14, 2026 while the case continues (details and caveats in the students guide). Ask your school’s international student office (your DSO) before any paid or unpaid work outside a campus job.

Taxes. If your net self-employment earnings for the year are $400 or more, you must file a return and pay self-employment tax, even if you are a dependent. For payments made in 2026, a client only has to send you a 1099-NEC once they pay you $2,000, and payment apps report on a 1099-K only above $20,000 and 200 transactions. No form does not mean no tax. If you expect to owe $1,000 or more, the IRS generally expects quarterly estimated payments.

Financial aid. Profit from a side business counts as student income on a later FAFSA, which uses tax information from two years earlier. Money you make in 2026 shows up on the 2028-29 FAFSA. Keeping receipts lowers your profit legitimately.

Campus rules. Many schools restrict running a business from a residence hall or selling on campus without permission. Creighton, Monmouth and St. John Fisher are examples; St. John Fisher requires written approval from the Dean of Students. Check your student handbook before you post a flyer in a dorm or use your school email for business.

For a student, the ladder is a set of lanes ordered from safest to most business-like. Pick one or two, not all five.

Campus job. The cleanest start, and the right one for F-1 students. A campus IT help desk job builds a real IT resume. A paid research assistant role in a lab teaches data cleaning, literature searches and coding. Pay follows your school’s student wage bands; the University of Virginia’s undergraduate band, for example, is $13 to $30 an hour. The campus tutoring center is a third option.

Campus small-business clinic. Some business schools and Small Business Development Centers run supervised clinics where students help local owners. A student AI clinic setting up AI tools for shops is a natural fit. Whether a paid clinic role counts as on-campus work depends on who employs you, so F-1 students should ask the DSO.

Tutoring the honest way. Tutoring that uses AI the honest way means teaching students to use AI as a study partner: explaining a step, generating practice problems, checking their own draft for clarity, planning for a test. It never means doing graded work. Put that in a written policy every family sees. Writing essays, doing problem sets or taking quizzes for pay is contract cheating, and several states, including California (Education Code 66400), have laws against selling completed academic work. Mind the tools’ age rules too: Claude’s consumer plans require users to be 18 or older, and ChatGPT allows ages 13 to 17 only with a parent’s permission.

Senior workshops. AI and scam-awareness workshops for seniors at the library or senior center. The need is documented: in the FBI’s 2025 Internet Crime Report, people 60 and older filed 201,266 complaints and reported $7.748 billion in losses. Start with one free pilot talk, then ask for a paid series.

Club and league websites. Websites for student groups, clubs and youth leagues. Start with your own club, then ask the student activities office which groups have dead sites. Follow the school’s brand rules for any official group, and build in accessibility basics.

Clubs and leagues should pay for their own domain and hosting in their own name, so the site survives when you graduate. Don’t use school software licenses or school accounts for paid client work without checking the school’s rules.

Weeks 1 and 2: rules and calendar. Talk to your DSO if you are on a visa, read the student handbook’s business and solicitation rules, and block your classes, labs and exam weeks in one calendar. Only the remaining hours are for sale.

Weeks 1 to 6: choose and ask. Pick two lanes. Write a one-page no-homework policy if you tutor, and a four-session outline if you teach seniors. Contact venues, parents and clubs (plan below). Libraries and senior centers plan programs months ahead, so ask early.

Weeks 4 to 8: start small. Give a free pilot talk, take your first tutoring students, and agree on one club site with a written scope.

Weeks 7 to 12: paid work. Run a paid series, keep weekly tutoring steady, and finish the club site before finals. Midterms in weeks 7 and 8 will squeeze everything; schedule light.

Week 13 onward: finals. No new work. Send invoices and thank-you notes, and record your income and expenses.

Over a full year it looks like this:

Tutoring platforms bring students but take a cut; Wyzant’s commission is reported at about 25%, and its own page says tutors keep all of the fee for students they bring themselves. Libraries and senior centers often have fixed budgets for guest presenters, so ask what they usually pay. More in Pricing AI and IT services.

A worked month. Here is an example of a busy month in the middle of a semester. Twenty-four hours of tutoring at $35 bring $840. One four-session senior series brings $400. One club website brings $400. That is $1,640.

Costs: $20 for an AI plan, $25 for printing, $30 for travel and $25 for flyers and a domain, $100 in all. That leaves about $1,540 before tax for roughly 45 hours in the month, or about 11 a week.

That $1,540 is well above the $400 filing line, so set aside part of it for self-employment tax. See Small business taxes and Bookkeeping basics.

Who first. People who already know you: your old high school’s counselor and teachers, your own club, the library in your hometown, your family’s place of worship, the student activities office. A student is trusted most where people have seen them before.

The plan, with numbers. Over the first six weeks: message 10 parents through your old school network and post once in two parents’ groups where that is allowed; email 15 libraries, senior centers and retirement communities offering a free pilot talk; and contact 8 clubs or youth leagues with dead or missing websites, starting with your own. Expect 3 to 5 tutoring students, 2 or 3 pilot talks leading to 1 or 2 paid series, and 1 or 2 club sites in the semester. That is about ten clients.

Hi, I’m Priya, a junior studying math at State University, and I tutor high school algebra and precalculus. I teach students to use AI tools as a study partner the honest way: explaining steps and making practice problems, never doing graded homework or tests. I put that rule in writing for every family before we start. Sessions are $35 an hour at the public library or online, and the first 30 minutes are free so your student can see whether it helps. I’m happy to send my written policy first. Thanks, Priya (555-0191)

Channels, best first, for a student:

Turn clients into proof. Ask every family, venue and club for one sentence you may quote. Keep a portfolio page with your club sites, your workshop outline and the feedback. Those are what a future employer will ask about. More scripts are in Launch and marketing playbook.

Keep it simple. One calendar with classes and exams entered first. One spreadsheet for income, expenses and hours. A separate bank account or at least a separate card, and payments by traceable methods. A folder with your policy, outlines, handouts and signed scopes for club sites.

Safety and trust. Tutor minors in public places such as a library, or with a parent at home. Never take an older adult’s passwords or log into their bank for them; teach them to do it with a family member present. Keep a short written scope for every website so "one more page" has a price.

Tools. Free AI plans cover most of this work. A free website builder or a low-cost one the club pays for, a slide tool and a printer are enough. The tool list and costs are in Your AI toolkit and what it costs.

Getting help. Your international student office for visa questions. The campus entrepreneurship center, career center and service-learning office for contacts. The local Small Business Development Center and SCORE for free mentoring if this grows into a business. The full list is in Free help directory. If you bring in a friend to help, agree on pay and duties in writing, and remember the same visa and tax rules apply to them.

The main risks. A visa violation is the most serious and the hardest to undo, so check before you start. A family asks you to "just do this one assignment"; say no, and point to your policy. Grades slip around midterms. A tax bill arrives in April because no one sent a 1099. A club’s site breaks after you graduate because it was in your name.

When to change course (rules of thumb):

If your grades drop, cut back to one lane, or to a campus job with fixed hours.

If you are on an F-1 visa and unsure, stay with campus jobs until your DSO says otherwise. They still build the same skills.

If tutoring students don’t come, take a job at the campus tutoring center and build a track record there.

When you graduate, hand your venues and club sites to a younger student, or turn the work into a business: An AI automation studio, An AI-assisted media studio or Tech help for seniors and home users. Or use the portfolio to get hired; that may be the best result of all.

Checked in November 2026: ICE guidance on F-1 work, the IRS $400 rule, the 2026 1099 thresholds, AI tool age rules, the FBI 2025 elder-fraud figures and school pay bands cited. Re-check visa rules with your DSO each term; the 2026 DHS rule was blocked by a court but under appeal when this was written.

🧰 Your AI toolkit and what it costs

A new AI or IT business needs fewer tools than the internet suggests. One good AI assistant, one automation platform or the admin tools your service uses, a business email and a password manager will carry you through your first clients. Everything else should be bought when a paying client needs it, and usually in the client’s name.

This guide lists the tools by job, what each plan costs as of November 2026, the difference between consumer and business plans (which matters more than price), which tools can handle health data, a starter set under $100 a month, and what not to pay for.

Three rules keep tool costs down and protect your clients.

1. Learn one tool in each category well before adding a second. Clients pay for reliable results, and you get those from knowing one platform’s quirks, not from having accounts on six.

2. Clients pay for their own tools, in their own name. Their automation account, their AI business plan, their phone number, their Microsoft 365 tenant. You get admin access. This keeps their data in accounts they control, puts the vendor’s terms between them and the vendor, and means their systems keep running if you part ways. Your own subscriptions are for your business, learning and demos.

3. Match the plan to the data. A free consumer chatbot account is fine for drafting your own blog post. It is the wrong place for a client’s customer list, a patient’s insurance details or anyone’s tax return. The data decides the plan, not the price.

The AI receptionist names come from 2026 comparison blogs, which put those services at about $29 to $300 a month (Rosie about $49, Goodcall about $59 to $79, Smith.ai AI plans about $95 and up). We did not check their own pricing pages.

U.S. list prices, excluding tax, checked November 3, 2026. "Confirmed" means we read the vendor’s own pricing page. "Not confirmed" means the vendor’s page did not load for us or showed prices only in a browser, and the figure comes from secondary sources. Check before you buy or quote a client.

Two notes on the Claude row: the $200 tier is now called Ultra, and Pro is enough for most one-person businesses. Move up only when you keep hitting usage limits on real client work.

AI vendors run two sets of terms. Consumer plans (free and individual paid plans) are written for one person. Business plans (team, enterprise and API access) come with commercial terms, admin controls and, in most cases, stronger commitments about how customer data is used. For client work, that difference matters more than the monthly price.

Training and data use. Anthropic’s privacy center says chats on its consumer plans (Free, Pro, Max) are used to improve its models only in limited cases, such as when you turn on the model-improvement setting, and that its business products are covered by separate commercial terms. Other vendors make similar distinctions, but the details and defaults differ and change. Before any client data goes into a tool, read that plan’s current data terms and turn off training where you can.

Commercial use of outputs. Anthropic’s consumer terms assign its rights in outputs to the user and do not ban commercial use, though they ban reselling the service. OpenAI’s terms say you own the output to the extent the law allows (from secondary sources; its page did not load for us). Midjourney lets any paid subscriber use outputs commercially, but companies with more than $1 million in yearly gross revenue must be on Pro or Mega. ElevenLabs includes a commercial license from the Starter plan up, not on Free.

Copyright of AI output. The U.S. Copyright Office said in January 2025 that output is protected only where a human determined enough of its expressive content, and that prompts alone are not enough. For clients, that means raw AI text or images you deliver may not be protected by copyright. Edit meaningfully, document your human work, and in contracts assign "whatever rights exist" rather than promising full ownership.

Age limits. Anthropic requires consumer users to be 18 or older. OpenAI allows 13 to 17 with a parent’s or guardian’s permission (secondary sources). Google allows 13 and up with a personal account, 18 and up with a work account. If you tutor or train minors, pick tools they are allowed to use.

Admin controls. Business plans let you add and remove users, see who has access and set policies. When a client’s employee leaves, you can remove their access in one place. That alone justifies a business plan for any office with staff.

If a client is a dental office, clinic or any business that handles protected health information, you become a HIPAA business associate when you touch that data. You need a signed business associate agreement (BAA) with the client, and every AI or cloud tool that sees the data must also be covered by a BAA. Consumer chat accounts do not come with one. What we found on the vendors’ own pages in November 2026:

The practical rule: if a vendor will not sign a BAA for the plan you are using, no health data goes into it. Build the flow so the sensitive parts happen inside covered systems, and use the uncovered tool only for steps with no patient information.

Other regulated data has its own rules. Tax-return information is covered by IRC 7216, which makes disclosure or use without the taxpayer’s written consent a crime, and commentators say feeding return data to public AI tools likely needs that consent. School records fall under FERPA. The contracts and AI rules guide covers these; the dental insurance verification and tax-season document desk pages show how they shape a real service.

This is a working set for a one-person automation, admin or training business. Swap any line for an equivalent; the point is the shape, not the brand.

If you run Microsoft 365 offices, swap Google Workspace for a Microsoft 365 business plan so you work in the same tools as your clients (we did not check Microsoft’s current prices). If you build automations for technical clients, n8n’s free Community Edition on a small server can replace Zapier, at the cost of maintaining the server yourself.

Two free things belong in every starter set: multi-factor sign-in on every account, using an authenticator app, and a written list of which client systems you can access. You will need that list for contracts, insurance applications and the day a client asks.

"AI agency" courses and communities with income screenshots. The skills are taught free by the tool vendors and by the organizations in the free help directory. The FTC has sued sellers of AI business opportunities over earnings claims.

Prompt packs. Prompts are easy to write and change with every model update. Your value is knowing the client’s process, not owning a list of prompts.

White-label agency platforms before you have clients. Bundled platforms that agencies resell to clients can cost the agency a few hundred dollars a month (affiliate blogs cite $297 to $497 for one popular platform). That only makes sense once several clients are paying you for what it provides.

Top AI tiers you do not need. The $100 to $200 plans are for heavy daily use. Start on the $20 tier and move up when limits slow real paid work.

Overlapping subscriptions. Three chat assistants and two automation tools double your bills and halve your skill. Cancel the ones you have not used in a month.

Purchased lead lists and bulk-email tools for cold outreach. They break the trust local selling depends on and can run into CAN-SPAM. The launch and marketing playbook explains what works instead.

Tools that promise compliance or certainty. The FTC ordered an accessibility overlay vendor, accessiBe, to pay $1 million (announced January 2025) over claims its AI tool could make any website accessible, and took action against an AI content detector over an unproven accuracy claim. Do not buy, or resell, a tool on the strength of a claim like that. Real accessibility work means manual testing; see website accessibility audits.

Review generators. The FTC’s rule on consumer reviews bans AI-written reviews that misrepresent a reviewer’s experience. A tool that writes reviews for you or your clients is a liability, not a feature.

Prices and vendor terms were checked on November 3, 2026, from the vendors’ own pages where marked "confirmed" and from secondary sources elsewhere. AI plans and data terms change often; check the current page before you buy or put client data into any tool.

💵 Pricing AI and IT services

Most new AI and IT businesses charge too little, by the hour, for open-ended work. That combination leaves them busy and broke. The fix is not a magic number. It is a structure: a paid first step, a fixed-price build, and a monthly plan, each with a written scope. This guide covers that structure, how to set the numbers, what marketplaces pay compared with local clients, and how to get paid on time.

All prices below are examples or ranges from the sources named. They are not market rates for your town. Use them to check that your own numbers are in a sensible range, then set your prices from your costs, your hours and what the work is worth to the client.

Almost every service in this section has two parts. There is a setup: you learn the client’s process, configure tools, build automations, train staff. Then there is upkeep: apps change, logins expire, staff leave, AI models get updated, and someone has to notice when things break. Price them separately.

The setup is a fixed price for a defined result: "intake form to CRM to approved reply, tested and documented, live in two weeks." Clients like fixed prices because there is no meter running. You benefit too: as you repeat similar builds, you get faster and the price stays the same.

The monthly plan (often called a care plan or a retainer) covers monitoring, fixes, small changes, and a set amount of improvement work each month. It is the part that makes income predictable. A business with ten clients on $300 plans starts each month with $3,000 already committed before any new project.

Agency and freelancer price guides put small automation builds at about $1,000 to $5,000 and monthly plans at about $150 to $1,500. Those figures come from vendor and agency blogs, not surveys, so treat them as a rough range. Fiverr’s own cost guide lists maintenance retainers at $50 to $500 a month, which shows where the low end of the global market sits.

Hourly pricing feels fair and is easy to explain. Its problem is that it punishes you for getting good. An automation that took you 20 hours the first time takes 6 hours the fifth time, and an hourly price falls by 70% just as your work gets better. It also makes clients nervous, because every question costs money.

Value pricing sets the price by what the result is worth to the client, then checks it against your costs. If a flow saves an office manager 5 hours a week, and her time costs the business about $30 an hour including payroll taxes, that is about $150 a week or roughly $7,500 over a 50-week year. A $2,500 build pays for itself in under five months, and the client can see why the price is reasonable.

Two cautions. First, the value number is the client’s to confirm, not yours to invent. Ask in the discovery session: "How many hours does this take now? What does that person cost?" Write their answer down. Second, do not put a savings figure in your marketing or contract as a promise. The FTC has brought cases against AI sellers for savings and accuracy claims they could not support. Say what you will build and what you will measure; let the client’s own numbers make the case.

Hourly still has a place: for work outside the agreed scope, for open-ended troubleshooting, and as a fallback rate written into every contract. A typical structure is a fixed price for defined work and an hourly rate for everything else.

The first meeting with a serious prospect should usually be free and short: 20 to 30 minutes to see whether there is a fit. The second step should be paid. A paid discovery session (or process review, or assessment) is two to four hours where you sit with the owner and staff, map one process step by step, and hand back a short written plan: what to fix, in what order, and what each fix costs.

Charge a modest fixed fee and credit it toward the build if they go ahead. This does three things. It filters out people who want free consulting. It pays you for real work, because the map is useful even if they never hire you. And it makes your build quote accurate, because you have seen the process instead of guessing.

How to offer it. "Before I quote the build, I’d like to spend about three hours with you and Jen walking through how a new tenant inquiry moves today. You’ll get a one-page map and a fixed quote for the fixes. The review is $400, and if you go ahead with the build, the $400 comes off the price."

Packages make buying easy. Three tiers is common: a small entry option, the one you expect most people to choose, and a larger option for clients who want more. Name them by what they do, not "bronze, silver, gold."

These are example prices for a U.S. operator selling locally, matching the sample automation page in this section. For workshops, the research behind this section found no credible source for a typical day rate, so price from your preparation time, the group size and what local trainers charge.

Retainers. A retainer is a monthly fee for a set amount of availability or work. Be specific about what it buys: response time (for example, next business day), a number of included hours or tasks, and what happens to unused hours. Rolling unused hours forward for one month is generous; rolling them forever creates a debt you will eventually have to work off for free. If a client keeps going over, move them to a bigger plan rather than billing overages every month.

Per-user IT pricing. IT support for small offices is usually priced per user per month, sometimes per device. The price covers a defined list: help desk support, account setup and removal, security settings, updates, backup monitoring. Projects (a new office, a migration) and hardware are billed separately. One 2026 vendor guide reported an average of about $145 per user per month for full managed IT, with a range of about $100 to $250, citing a survey of 412 providers and buyers. Lighter work, such as Microsoft 365 or Google Workspace administration without full support, sells for less; the research behind this section estimated $15 to $40 per user per month but found no survey for it. Set a monthly minimum so that a three-person office still covers your fixed costs.

Example: a 12-person office on a light admin plan at $35 per user is $420 a month. If two staff leave and two join, the count stays at 12. If the office grows to 15, the plan grows with it, which is one reason per-user pricing is popular with both sides.

Take a property management office with 12 staff. The office manager spends about 6 hours a week sorting tenant inquiries and maintenance requests by hand. The owner confirms she costs about $30 an hour including payroll taxes. That is about $180 a week, or $9,000 over a 50-week year. You propose:

Your time: about 6 hours for the review, 30 hours for the build, and about 4 hours a month on the care plan, or 48 hours over the year. That is 84 hours in all, so the first year pays about $86 an hour ($7,200 divided by 84). Billed hourly at $75, the same 84 hours would have brought $6,300, and every speed-up would have cut your income further.

Two points to take from it. The client gets a price below the value they confirmed, so the decision is easy. And most of your income from this client is the care plan, so it is worth more to keep a client happy for three years than to win a large one-time build.

This example does not include your costs (software, insurance, marketing) or taxes. See Pricing and margins for how to build those in, and set aside a share of each payment for income and self-employment tax (Small business taxes).

Freelance marketplaces are a real source of work, but their price levels are set by global competition. Local clients who meet you, trust you and need you on site pay differently. The table compares the ranges in our research.

Platform fees come off the top. Upwork charges freelancers a variable service fee of 0% to 15% per contract. Fiverr takes 20% of each order according to third-party summaries of its terms; we did not confirm that on Fiverr’s own pages. More on choosing channels is in the launch and marketing playbook.

The gap between the marketplace rows and the local rows is the price of trust, presence and accountability. A local dental office cannot judge your automation skill, but it can judge whether you showed up, explained things clearly and answered the phone. That is what the higher price buys, and it is why selling locally is worth the slower start.

Your first prices will be too low. That is normal, and fixing it is routine. Raise prices for new clients first; it costs nothing to quote the next prospect 15% higher and see what happens. If most new prospects accept without hesitation, you are still underpriced.

For existing clients, put an annual review in the contract from the start ("prices are reviewed each January"). Give at least 30 to 60 days’ notice in writing, say what the new price is and when it starts, and tie it to something real: more tools covered, faster response, a year of improvements. Some owners keep early clients at the old price for a set time as a thank-you. Do not apologize for raising prices, and do not raise them in the same month a problem happened.

Signs you are underpriced: you are fully booked and still not earning what you planned; nobody ever pushes back on a quote; you dread opening the care-plan inbox. Signs you are overpriced, or your proof is weak: proposals stall after the call, and prospects ask for references you do not have yet.

Get paid at the start. For builds, ask for a deposit, commonly 50%, before work begins, and the balance on delivery. For care plans, bill monthly in advance, with automatic card or bank payment if you can. Paying in advance is normal for subscriptions and owners accept it.

Invoice the same day. State the amount, what it covers, the due date and how to pay. Short terms such as 15 days are reasonable for small clients. Use your accounting software’s reminders so you do not have to chase by hand.

Write the late-payment terms into the contract. Say what happens if a payment is late: a reminder at a set number of days, a pause in non-urgent work at a later point, and a late fee only if your contract says so and your state allows it. Pausing work is usually more effective than a fee. Never hold a client’s systems or data hostage; set things up so their accounts are in their name from the start, which also keeps you on the right side of any dispute.

Scope creep is the steady addition of "small" requests to a fixed price. Every one sounds reasonable. Together they turn a profitable project into unpaid work. Three habits prevent it:

1. A written scope. List what is included and, just as important, what is not. "Includes the three flows listed. Does not include new forms, website changes or staff training beyond one walkthrough."

2. A change request. When a new request comes in, reply the same day: "Happy to do that. It’s outside this project, so it would be $X or Y hours at our hourly rate. Want me to add it?" Most clients respect this. A few will drop the request, which is also fine.

3. A home for small requests. The care plan’s monthly improvement is where small requests go. "Good idea; I’ll put it in next month’s improvement." That turns scope creep into a reason to keep the plan.

More on contract terms is in the contracts and AI rules guide and Leases and contracts.

Marketplace rates and fees (Upwork, Fiverr) and the survey figure for managed IT were checked in November 2026. Most price ranges here come from vendor and agency blogs, not surveys; treat them as rough guides and check what local providers charge before you quote.

📣 Launch and marketing playbook

Most new tech businesses do not fail because the work is bad. They fail because too few of the right people hear about it in the first six months. The owner builds a website, posts twice on LinkedIn, waits, and goes back to a job. This guide is the part that comes before and around the work: who you sell to, what you say, where you say it, how you prove it, and the rules that apply when you email, text or quote a happy client.

It is written for a one- or two-person AI or IT service selling to local and regional businesses. It leans on trust and conversations, not ads, because an owner who hands you their email, customer list or patient records does it for someone a friend vouched for, or someone they watched teach a room.

Positioning is the sentence that tells a stranger whether you are for them. "I help businesses with AI" fails that test, because every owner hears it as "I will cost you money and take your time." A good positioning sentence has three parts.

One customer. Pick a type of business you can name in two or three words: dental offices, HVAC companies, small CPA firms, independent insurance agencies, property managers, town clerks. One customer type means your examples, talks and referral partners all point the same way, and each client teaches you something you reuse.

One problem. Name a job the owner already hates and already loses money on. Missed calls after 5 p.m. Insurance checks the morning of the appointment. Tax documents that arrive in pieces in March. Leads that wait a day for a reply. Owners do not buy "AI." They buy the end of a specific headache.

One promise you can keep. The promise should be something you control and can measure, such as "every after-hours caller gets a text within a minute and a call back the next morning" or "your office sees each patient’s benefits the day before the visit." Do not promise outcomes you do not control, like a revenue increase, a percentage saved or "fully automated." The FTC has brought cases against AI sellers over exactly those kinds of claims (more under the rules below).

Put the three together and test it out loud: "I set up after-hours call handling for HVAC companies in our county, so every missed call gets a text in a minute and a booked slot by morning." If a heating contractor hears that and says "how much," it works. If they say "what does that mean," shorten it.

One test: can you reach 50 businesses of your chosen type within an hour’s drive or through one association? If not, widen the niche one step (from "orthodontists" to "dental offices"), not back to "everyone."

An offer is what someone can say yes to this week. It has a name, a scope, a price and a time frame. "Missed-call setup: we connect your phone line, write the text and the after-hours script, test it with you, and run it for 30 days. $750, live in 10 business days." Compare "let’s talk about how AI could help your business," which asks the owner to do your thinking.

The paid pilot. Your first three to five clients should be paid pilots: a small, fixed-scope version of your main service, at a fair but lower price, in exchange for two things written into the agreement. First, permission to measure the before and after. Second, permission to publish a short case study if the results are good, with names removed if they prefer. People who pay show up to meetings, give you access and take the result seriously. Free work stalls.

Keep the pilot to 30 days or less, one process, and one number you can measure. Credit the pilot price toward the full service if they continue. That turns the pilot into the first step of the sale rather than a discount you have to walk back.

Your proof. A new business has no reviews and no logos to show. Proof is how you close that gap. In order of strength:

1. Before and after numbers you measured. "Reply time to new leads went from next day to under 5 minutes." "The office manager spent 6 hours a week retyping forms; now it is about 40 minutes of checking." Measure the before during the pilot’s first week, not from the owner’s memory. Write down how you measured it.

2. A short case study. One page: the business type, the problem in their words, what you set up, the measured result, and one quote from the owner. Three of these are worth more than any website design.

3. A screen recording. Two minutes of a real flow running, with client details blurred. Windows and macOS both include a basic screen recorder.

4. A live demo on fake data. Build it for a made-up business in your niche, "Northside Heating," with fake customers, so you can show it anywhere without risking anyone’s data.

5. Testimonials. A sentence from a real client, used with written permission, saying what changed. The FTC rules on testimonials apply (see below), so never edit the meaning, never write one for them, and do not imply that one client’s result is typical unless it is.

Ask for the case study at the end of the pilot, while the result is fresh. Here is a message you can send.

Case-study request. Hi Maria, now that the missed-call setup has run for a month, I pulled the numbers: 41 after-hours calls got a text within a minute, and 17 of them booked a visit. Would you be OK with me writing this up as a one-page case study? I’d send you the draft first, and I can leave out the company name if you prefer. A two-sentence quote from you about what changed would help other shop owners understand it. No pressure either way, and thanks again for being an early client.

The first 30 days are about conversations, not content. The target is simple: 40 to 60 direct, personal messages sent, 10 or more real conversations held, and two or three paid pilots agreed. Everything else (the website, the LinkedIn profile, the logo) supports that and should take no more than a few evenings.

After day 30 the plan shifts from finding first clients to delivering for them, turning results into proof, and building the channels that bring steady leads. The figure shows the weekly rhythm through day 90.

Weeks 5 to 8. Deliver the pilots well. Send each client a short weekly note on what changed. Keep two or three outreach sessions a week going so the pipeline does not empty while you work. Give your first talk.

Weeks 9 to 13. Turn each finished pilot into a case study and a monthly care plan. Ask each happy client for a review and one introduction. Give a second talk. If, and only if, the free channels are producing conversations, test one marketplace profile or a capped ad budget for a month.

Your first clients will almost certainly come from people who already know you or know someone who does. That is not a shortcut; it is how most service businesses start. The trick is to ask in a way that is specific and easy to say yes to, and to ask one person at a time.

Send each warm email by hand. Mention something real about the person. Name the niche and the problem so they can think of a specific person. Ask for an introduction or a 20-minute conversation, not a sale.

Warm email. Subject: A quick favor, and a question about dental offices. Hi James, it’s been a while since we worked together at Fairview. I’ve started a small business setting up insurance verification for dental offices, so the front desk sees each patient’s coverage the day before the visit instead of at checkout. I’m looking for two or three offices to run a paid 30-day pilot with. Do you know an office manager or dentist who complains about insurance surprises? A name would help a lot, or if you have 20 minutes I’d like to hear what you think of the idea. Thanks, Priya.

About half the people you write to will not answer the first time. That usually means busy, not no. Follow up once, a week later, short and friendly. After that, stop.

Follow-up. Hi James, bumping this in case it got buried. Even one name of a dental office manager who might talk to me for 15 minutes would help. If now’s not a good time, no problem at all. Priya.

Referral partners. A referral partner is a business that already has your customers’ trust and sees the problem before you do. For AI and IT services, the best partners are:

Bookkeepers and accountants. They see the messy invoices, the retyped spreadsheets and the late reports. They also do not want to do the automation themselves.

IT providers and managed service providers. Many handle hardware, networks and security but do not want to build automations or run AI training. They also need someone they trust for jobs too small for them.

Web designers and marketing agencies. Their clients ask for chatbots, booking flows and CRM setup. A designer who sends you that work keeps the client happy without learning a new trade.

Vendors who sell software to your niche. The local rep for a dental practice software, a field-service app or a tax portal meets dozens of your target customers. Many of their customers buy the software and never set it up fully.

Tell partners exactly what kind of client and problem to send, report back on every referral, and send work their way when you can. If you pay referral fees, put the terms in writing and check that the partner’s profession allows them; some licensed professions restrict referral payments.

Partner pitch. Hi Dana, I run a small business setting up automations for property managers: tenant inquiries answered within minutes, maintenance requests sorted and sent to the right vendor, and rent reminders that go out on their own. Several of your bookkeeping clients are property managers, and you probably see the mess before anyone. I’m not looking to touch the books or anything you do. I’d like to be the person you can send them to when they ask for help with the rest, and I’ll keep you in the loop on every referral. Could I buy you a coffee next week and show you a two-minute demo? Sam.

Small business owners still gather in person, and a newcomer who shows up regularly and helps people becomes known within a few months. Pick two or three of these and go consistently rather than trying all of them once.

Chamber of commerce. Many events are open to guests, and membership puts you in the directory. Dues vary widely, so go to two events as a guest before joining.

Referral groups. BNI is the best-known, and there are independent groups modeled on it. Chapters usually allow one member per profession, meet weekly, charge yearly dues and expect referrals, so visit as a guest before committing. The payoff is an accountant, an insurance agent and a commercial realtor in the same room each week.

Trade associations. Your niche has them: the state dental association and its local component societies, the state society of CPAs, the local home builders association, a property management association, a county clerks’ association. Their meetings and newsletters reach exactly your customer. Many accept vendor or associate members and look for speakers.

Libraries and SBDC events. Public libraries run free programs and often want speakers on practical topics. Small Business Development Centers and SCORE chapters run workshops for owners, and some invite outside presenters.

Teaching as marketing. A free 45-minute talk lets a room watch you explain something clearly and answer questions honestly, which builds more trust than any ad. Choose a practical title tied to your niche: "Three jobs your office can stop doing by hand," "How to spot AI voice-clone scams," or "Using AI with client data without getting in trouble." Teach real things, mention your service once at the end, and hand out a one-page sheet with your contact details and the pilot offer. A lunch-and-learn inside one business works the same way: you bring lunch, teach the staff one useful thing, and the owner sees you work.

Two ideas in this section are built around teaching: AI training workshops for small teams and AI and scam-awareness workshops for seniors. Both make good door openers for other services too.

Workshop invite (to a librarian or association program chair). Hi Ms. Ortiz, I run a small local business that helps offices set up AI tools safely. I’d like to offer a free 45-minute session for your business patrons called "Using AI at work without risking client data." It covers what the main tools do well and badly, what never to paste into a chatbot, and a one-page policy attendees can take home. There’s no sales pitch; I bring the slides and a handout, and I can do a weekday evening or a lunch slot. Would that fit your spring program? I can send an outline and references.

Google Business Profile. If you meet clients in person, at their office or yours, set up a free Google Business Profile. Google’s guidelines are built for businesses with in-person contact with customers; an online-only business does not qualify. If you go to clients rather than having them come to you, set it up as a service-area business, which hides your home address and shows the towns you serve instead. Do not use a P.O. box or virtual office address. Choose the most accurate category, write a plain description that repeats your positioning sentence, add real photos (you at a client site, a slide from your talk), and ask every happy client for a review. For local searches like "IT support near me" or "computer help for seniors," the map listing often matters more than your website.

A one-page website that converts. You do not need ten pages. You need one page that answers a visitor’s three questions (is this for me, can I trust them, what do I do next) in under a minute. The figure shows the order.

Show at least a starting price. Use a real photo of yourself and put your town near the top. Link the button to a calendar with 20-minute slots. Add a privacy note if you collect form data. Skip stock photos of robots and glowing brains.

LinkedIn for business clients. Office owners and managers often check you on LinkedIn after a meeting or referral. Three things matter.

The profile. Your headline is your positioning sentence, not your job title. The About section says who you help, the problem, how you work and how to reach you, in plain sentences. Use the Featured section for one case study and one demo video. Use a clear, friendly photo.

A posting rhythm you can keep. One or two posts a week is enough. Good posts: a problem you saw this week and how you would handle it; a 60-second demo; a lesson from a pilot, with the client’s permission; a short explainer of a rule your niche worries about. Write in your own words.

Direct messages that are not spam. Connect with people you met, people who engaged with your posts, and people a mutual contact introduced. Add a note that says why. Do not pitch in the first message. Do not send the same message to 200 people. LinkedIn’s user agreement bans bots and automation tools that send connection requests or messages for you, and accounts that use them get restricted. A good first message looks like this: "Hi Tom, thanks for the comment on my post about tax-season document chasing. Sounds like your firm has its own system. How do you handle it when a client sends half their 1099s in February and the rest in April?"

Marketplaces. Upwork, Fiverr and Contra can bring paid work, especially for technical tasks like automation builds, chatbots and AI video, where Upwork reported large growth in 2025: U.S. earnings from AI integration work rose 178% and AI chatbot development 71%. They also put you in a global price competition. Upwork’s own public rate ranges for AI developers run about $30 to $50 an hour and chatbot developers about $30 to $61, and those profiles include freelancers in lower-cost countries. Fiverr’s cost guide lists simple no-code automations at $40 to $250.

Fees matter. Upwork charges freelancers a variable service fee of 0% to 15% per contract (Upwork’s rates guide, November 2026) and you pay for Connects to send proposals. Fiverr takes 20% of each order according to its terms as summarized by several third-party sites (we did not confirm this on Fiverr’s own pages). Contra says it charges freelancers no commission and charges clients instead (secondary sources).

When marketplaces make sense: you are building a portfolio and need a few paid projects fast; your service is technical and remote (automation, chatbots, dashboards, video); or you want to test whether a narrow offer sells before building local channels. When they do not: your edge is local trust, in-person training or regulated industries, where a clinic or law firm will not hire a stranger from a marketplace. Treat a marketplace as one channel with a budget of time, not as the business. Move good clients onto a direct contract when the platform’s terms allow it; Upwork and Fiverr both have rules against taking platform clients off the platform, so read them before you do.

Content without drowning. Content helps when it shows your work; it hurts when it eats the hours you need for conversations. A sustainable plan for one person:

That is roughly 8 to 10 hours a month. If you only have time for one thing, make it the monthly newsletter to people who already know you. Use AI to help draft, but read every line and check every fact; Google’s spam policies target scaled, low-value AI content.

Paid ads, last, and capped. Ads amplify something that already works; they do not create it. Wait until you have a clear offer, at least two case studies and a page that turns visitors into booked calls. Then run one test at a time: a single platform, a single offer, a fixed monthly cap, and a stop rule. For example: $150 for the month, a $5 daily budget, search ads on phrases your customer actually uses ("dental insurance verification service," "IT support for small office near me"), sent to your one-page site. Stop the test if it has not produced at least one real conversation by the time half the budget is spent. Track cost per conversation, not clicks.

These rules are easy to follow if you build them in from the start, and costly if you ignore them. This is a summary, not legal advice; the contracts and AI rules guide goes further.

Email (CAN-SPAM). The FTC’s compliance guide says the law covers commercial email, and makes no exception for business-to-business email. The core duties: no false or misleading header information; no deceptive subject lines; identify the message as an ad where it is one; include your valid physical postal address (a street address, a registered P.O. box or a registered private mailbox); tell recipients how to opt out; honor opt-outs within 10 business days and keep the opt-out working for at least 30 days after sending; and you remain responsible if someone else sends email for you. The guide lists penalties of up to $53,088 for each email in violation. Put your address and an opt-out line in anything you send to a list.

Texts and calls (TCPA and A2P 10DLC). Marketing texts and calls sent with automated systems need the person’s prior express written consent. The FCC ruled on February 8, 2024 that AI-generated and cloned voices count as "artificial or prerecorded voice" under the TCPA, so outbound AI calls need prior express consent too, plus identification and a way to opt out. Do not sell or run outbound AI cold-calling. Business texting from ordinary 10-digit numbers must be registered with the carriers (A2P 10DLC); on Twilio, that has meant small one-time brand and campaign vetting fees plus a monthly campaign fee (Twilio’s support page, fees in effect since August 2025). Many owners give out a cell phone, so treat any number as one the Do Not Call rules can cover. Honor "STOP" and other clear opt-out requests right away.

Reviews (FTC rule, 16 CFR Part 465, in effect since October 2024). It bans fake reviews, including AI-written reviews that pretend to be a customer’s experience; buying reviews conditioned on them being positive; undisclosed reviews by owners, staff or relatives; and suppressing negative reviews. Asking all your clients for an honest review is fine. Google’s own review policy goes further: do not offer anything in exchange for reviews, and do not ask only the clients you expect to be happy. Never write a review for a client "to save them time."

Review request. Hi Maria, thanks again for working with me on the call setup. If you have a minute, an honest Google review would help other shop owners decide whether this is right for them. Here’s the link: [link]. Whatever you think, good or not, I’d like to hear it. Priya.

Testimonials and endorsements (FTC Endorsement Guides, revised 2023). Use real quotes, with permission, that reflect the client’s honest view. If you gave the client anything for the testimonial, such as a discount, disclose it. If you show one client’s result, do not imply it is what everyone gets unless you have evidence that it is typical; otherwise say clearly what results people should generally expect.

AI claims (FTC Act, Section 5). The FTC has brought a run of cases over AI marketing: the Operation AI Comply cases in September 2024, including DoNotPay’s "robot lawyer" claims; a final order against Workado in August 2025 over an unproven "98% accurate" claim; a $1 million order against accessiBe, announced in January 2025, over claims its AI tool could make any website accessible; and a 2026 proposed settlement with Air AI over earnings and refund claims made to small businesses. Each turned on a claim the seller could not back up. Do not promise accuracy percentages, income, unmeasured savings, "fully automated" results, or that a bot replaces a licensed professional. Say what the tool does, what a person still checks, and what you measured.

Keep one spreadsheet. Each row is a person or business. Columns: date, name, business type, source (warm, partner, event, talk, LinkedIn, Google, marketplace, ad), stage (contacted, replied, call held, proposal sent, won, lost), value, next step and next-step date. Fill it in the same day. Review it every Friday. In month two you will see which sources produce conversations and which produce nothing, and you can shift your hours.

The funnel below uses example numbers for a first quarter. Your own numbers will differ; the point is to know them.

A budget keeps marketing from turning into random spending. Here are three example monthly budgets. The amounts are allocations, not quoted prices; check the real cost of any tool or membership before you pay.

The $0 plan is a real plan: first clients come from conversations, which cost time. Spend money first where it buys conversations (events, partner lunches) and last on ads.

1. Selling AI instead of a result. Lead with the problem and the measured result; mention tools later, if at all.

2. Serving everyone. "Any business that wants AI" gives nobody a reason to call you. One niche makes every conversation easier.

3. Building the website before talking to customers. A plain one-page site is enough until you have proof to put on it.

4. Working free for too long. A discounted pilot is fine. Months of free work teach clients your time costs nothing.

5. Hiding the price. Show a starting price. It filters out people who cannot pay and saves both sides a call.

6. Mass messaging. Bulk cold email and automated LinkedIn messages burn your reputation and can break CAN-SPAM or platform rules.

7. Overclaiming. "Saves 20 hours a week" or "99% accurate" without measurement is the exact pattern the FTC has pursued. Say what you measured.

8. Not following up. Most yeses come after a second or third contact. Put a next-step date on every row of your sheet.

9. Skipping proof. End every project with a before-and-after number, a case study request and a review request.

10. Stopping outreach once busy. When pilots start, outreach stops, and two months later the pipeline is empty. Keep a few hours a week for it.

Rules and figures here (CAN-SPAM, the FTC review rule and AI cases, the FCC AI-voice ruling, A2P 10DLC fees, Upwork and Fiverr data and fees) were checked in November 2026. Fiverr’s 20% fee comes from third-party summaries, not Fiverr’s own page. Platform fees and policies change often; check them before you rely on them.

📝 Contracts, client data and AI rules

Most new tech businesses lose money on paperwork they skipped, not on bad work. A client says the chatbot was supposed to do more. A dental office learns you pasted patient names into a consumer chat app. A project runs four months past the quote because nobody wrote down what "done" meant. A one-page contract and a few habits about client data prevent nearly all of it.

This chapter covers what your services contract should say, who owns work you make with AI, how to handle client data, the industry rules that come with health, tax, school and hiring clients, the marketing and calling rules the FTC and FCC enforce, the newer AI laws, and the insurance that backs it all up. It is a map, not legal advice. Where a rule could cost you real money, the chapter says so and tells you when to pay a lawyer.

You need a short document both sides read before work starts, not a 30-page agreement. Many people sign a master services agreement once and a short statement of work per project. For a first client, one document of two or three pages is fine.

The Main Street library’s chapter on leases and contracts covers signing basics that apply here too. If you charge a monthly care plan, the contract should also say what the plan covers each month, the response time you promise, and what counts as extra work. The pricing side of that is in Pricing AI and IT services.

Clients often assume they will own a copyright in everything you deliver. With AI output, that assumption can be wrong, and your contract should say so plainly.

The U.S. Copyright Office set out its view in Part 2 of its AI report, on copyrightability, published January 29, 2025. In short: AI output is protected only where a human author decided enough of the expressive elements. Typing prompts alone is not enough. Human work that remains visible in the output can be protected, and so can a human’s creative selection, arrangement or editing of the output. Using AI as a helper does not by itself block copyright.

What that means for a contract:

Assign whatever rights exist. Write that you assign to the client all rights you hold in the deliverables, "to the extent any exist," instead of promising that every piece is copyrighted.

Say where AI was used. One sentence is enough: "Some drafts, images or code in this project were produced with AI tools and edited by us." Clients who plan to register a copyright need to know.

Keep a record of human work. Save drafts and edit history for logos, scripts, course material and other creative work. That record is what shows human authorship if anyone asks.

Check the tool’s terms. Most AI vendors give the user the rights they hold in the output. Some plans restrict commercial use. For example, the ElevenLabs Free plan does not include a commercial license, and Midjourney requires companies with more than $1 million a year in gross revenue to be on its Pro or Mega plan. Use a plan that allows client work. The plan prices are in Your AI toolkit.

Keep what is yours. Your templates, prompts and reusable workflows stay yours; the client gets a license to use them. That is how you sell a similar setup to the next client.

In Bluevine’s survey of 942 owners (April 2026), 33% named data security and privacy as a barrier to AI, up from 23% a year earlier. A clear data plan is part of what you sell.

NDAs. Sign a separate non-disclosure agreement before a discovery call where the client will show you real files, or when a larger client asks. Watch for one-sided terms, no end date, and clauses that would stop you working for anyone else in the industry.

Use business plans for client data. Consumer chat plans and business plans come with different data terms. Business tiers usually say customer data is not used to train models by default and give admin controls. Read the current terms of the exact plan you use, and write which tools you use into the contract.

Take only what you need, in their accounts. An invoice automation needs ten sample invoices, not the whole accounting file; test with masked data where you can. Ask for logins inside the client’s own systems instead of files sent to you, so they can turn your access off at the end.

State privacy laws. About 20 states now have comprehensive consumer privacy laws (IAPP’s count, cited secondhand). Indiana, Kentucky and Rhode Island took effect January 1, 2026. Most apply only above a set number of consumers (Rhode Island’s line is 35,000), so a small client may be below them. But when you work for a larger company, its contract will often pass "processor" duties down to you: follow its instructions, protect the data, help with deletion requests, and report breaches. Read those clauses closely.

The best-paying niches in this section come with federal rules. That is why careful operators get paid there.

HIPAA (clinics, dentists, therapists, billing offices). If you create, receive, keep or send protected health information for a covered practice, you are a business associate under HIPAA (45 CFR 160.103). You need a signed business associate agreement (BAA) with the practice before you touch any patient data. Every tool that sees that data needs a BAA too, which rules out consumer chat accounts. Encrypting the data does not remove the need for a BAA. Many AI and automation vendors sign BAAs only on business or enterprise plans, so check before you design the workflow. The industry AI desk game plan shows how to build a business around this.

IRC 7216 (tax preparers). Section 7216 of the Internal Revenue Code makes it a crime for a tax preparer to disclose or use tax-return information without the taxpayer’s consent. Consent has to be a separate written document for each disclosure or use, following Rev. Proc. 2008-35. The civil penalty under section 6713 is $250 per disclosure, up to $10,000 a year. Commentators say sending return data to a public AI tool likely needs a 7216 consent that names the recipient; the IRS has not issued AI-specific guidance. If you run a tax-season document desk, build the consent step into the workflow and let the CPA own the wording.

FERPA (schools). Schools may share student education records with an outside vendor only under the "school official" exception: the vendor has a legitimate educational interest, works under the school’s direct control, and does not pass the data on (34 CFR 99.31). In practice that means the school’s contract terms, not yours.

COPPA (children under 13). If anything you build collects personal information online from children under 13, COPPA applies, including verifiable parental consent. The FTC’s amended COPPA rule had a general compliance date of April 22, 2026. Tutoring sites, youth league sign-up forms and kids’ apps are where this comes up.

FTC rules on AI claims. The FTC has brought a run of cases over AI claims. DoNotPay paid $193,000 over its "robot lawyer" claims (final order February 2025). Workado’s final order (August 2025) came from claiming its AI content detector was 98% accurate without proof. Air AI’s proposed settlement (March 2026) bans its owners from selling business opportunities after the FTC said it misled small businesses with earnings and refund claims. The safe rule: do not promise income, savings percentages, accuracy rates or "fully automated" results you cannot prove, and never say your bot replaces a lawyer, doctor or licensed agent.

FTC Consumer Reviews and Testimonials Rule (16 CFR Part 465). In effect since October 2024. It bans fake reviews, including AI-written reviews that misrepresent the reviewer’s experience, paying or rewarding reviews conditioned on a positive opinion, undisclosed insider reviews, and suppressing negative reviews. Civil penalties apply per violation. Asking every customer for an honest review is fine. If you sell review-response or reputation services, design them around this rule. The marketing playbook has review-request wording that stays inside it.

TCPA and AI voices. The FCC ruled on February 8, 2024 that AI-generated or cloned voices count as an "artificial or prerecorded voice" under the Telephone Consumer Protection Act. Outbound calls using them need the called person’s prior express consent, and written consent for telemarketing, plus identification and an opt-out. An AI receptionist answering calls customers place is a different, lower risk. Outbound AI sales calling without consent is the danger zone; this section does not recommend it.

Business texting. Texts sent from ordinary 10-digit numbers must be registered with the carriers under A2P 10DLC. Marketing texts also need consent under the TCPA. If you set up missed-call text-back or reminders, the registration is part of your job.

Email. CAN-SPAM covers commercial email: no misleading subject lines, a real postal address, and a working unsubscribe. Mass AI cold email is legal under it but wrecks your domain’s reputation.

Most of what a small AI services firm builds, such as automations, document extraction and staff assistants, is not covered by the AI-specific laws below. They matter when your work helps make decisions about people, especially hiring, or when your clients serve Europe.

NYC Local Law 144. Employers and employment agencies in New York City may not use an automated employment decision tool unless it has had an independent bias audit within the past year, a summary of the results is public, and candidates got notice 10 business days before use. Enforcement began July 5, 2023. If you set up recruiting tools for a NYC client, you do not do the audit; an independent auditor does. Your job is to inventory the tools and set up the notices.

Illinois HB 3773. It amends the Illinois Human Rights Act, effective January 1, 2026. It bars using AI in employment decisions in ways that have a discriminatory effect, bars using zip codes as a stand-in for protected traits, and requires notice to workers. It does not require an audit.

Colorado SB 26-189. Signed May 14, 2026, it repealed and replaced Colorado’s earlier AI law (SB 24-205) with a narrower one on automated decision-making technology used in "consequential decisions" about education, employment, housing, lending, insurance, health care or government services. Developers must give deployers documentation, and deployers must give notice, explain adverse outcomes within 30 days, and offer correction and human review. The legislature’s summary dates the developer duties and the attorney general’s rules to January 1, 2027. The attorney general enforces it, with a 60-day cure period until 2030. The hiring-tool notices idea is built on these three laws.

EU AI Act, if you serve EU clients. It entered into force August 1, 2024. Since August 2, 2026, its Article 50 transparency duties apply: tell people when they are talking to a chatbot, label deepfakes, and mark AI-generated content in a machine-readable way (systems already on the market have until December 2, 2026 for the marking part). The heavier "high-risk" duties were pushed back by the 2026 Digital Omnibus to December 2027 and August 2028. A simple website bot for a European client needs a clear "you are chatting with an automated assistant" line and a way to reach a human.

Many insurers sell E&O and cyber together. Ask the agent: does it cover work done with AI tools, a client’s loss from my mistake, and contractors I hire? The Main Street library’s chapter on business insurance explains the policy types in more depth, and the one on legal structure covers how an LLC fits alongside insurance.

You can write a sound first contract from the checklist above. Pay a business lawyer for an hour or two at these moments:

Before your first health, tax or legal client. Have a lawyer review your BAA template or your 7216 consent workflow.

When a client sends you their contract. Large clients use their own paper, often with unlimited liability and broad indemnity. A lawyer can tell you which two clauses to push back on.

Before hiring-tool or AI-decision work. NYC, Illinois and Colorado rules are new, and your notices are close to legal advice. Work alongside an employment lawyer.

When something goes wrong. A data leak, a threatening letter, or a client refusing to pay a large invoice. Call your insurer and a lawyer before you reply.

Free first stops: your local SBDC and SCORE mentors can review a contract in general terms, and many state bar associations run lawyer referral services with a low-cost first consultation. The free help directory lists where to find them.

Checked November 2026: the Copyright Office report, Colorado SB 26-189 on the legislature’s site, the NYC Local Law 144 page, and the FTC cases listed. Not legal advice. Re-check Colorado’s attorney general rules before January 2027, the FTC’s July 2026 proposed policy statement on AI accuracy, and any vendor’s BAA terms before you rely on them.

🛡️ Staying ahead of AI

Anyone starting a tech business today is selling into a market that AI is reshaping while you watch. Some services that paid well three years ago now sell for a fraction of the price, because a client can get a passable version from a chatbot for free. Other services are growing fast because of AI. This chapter looks at what the evidence actually shows, how to keep moving toward the work that holds its value, which skills to learn next, and how to talk to clients about AI without overselling it or scaring them.

One caution before the numbers. No single study measures the whole market. The figures below come from marketplace data, academic papers and government surveys, each with its own window and blind spots. Read them as direction, not destiny.

The work that shrank first is work where the client wants a text, an image or a translation and does not much care how it was made.

Writing and other automation-prone freelance work. A study of 1.39 million job posts on a large global freelance platform (Demirci, Hannane and Zhu, Management Science, 2025) found that posts for automation-prone jobs, such as writing, software and engineering, fell 21% within eight months of ChatGPT’s release compared with manual-intensive jobs. An earlier study of Upwork (Hui, Reshef and Zhou, Organization Science, 2024) found that writers exposed to ChatGPT saw jobs fall about 2% and monthly earnings fall about 5.2%, and more experienced freelancers were not spared.

Translation. Comparing May to October 2022 with January to October 2023, Qiao, Rui and Xiong found translation and localization job volume down 9.0% and earnings down 29.7% on a freelance platform (arXiv working paper, not peer-reviewed in that version). The Bureau of Labor Statistics projects interpreters and translators to grow only 2% from 2025 to 2035, with machine translation plus human post-editing as the normal workflow. It projects medical transcriptionists to decline about 4%.

Small one-off jobs. Upwork’s management said in its second-quarter 2026 results that AI adoption cut client activity on contracts of $500 and below (reported by Staffing Industry Analysts). Ramp, the corporate card company, reported in May 2026 that among its customers who used freelancers in 2022, 58.5% had stopped entirely by early 2026, while their spending on AI model providers rose. That is Ramp’s own customer data in marketing material, and it shows a pattern, not cause and effect.

Pulling those together, the services commoditizing fastest are generic blog and copywriting, basic translation, transcription, simple logos and graphics, data entry, and basic scripted chatbots. That is our reading of the studies above, not a measurement of the whole market. Several idea pages in this section are marked "crowded" for this reason, and each says how to stay above the free version.

Demand has moved toward people who can make AI work inside a real business, and toward the media AI makes possible.

On Upwork, per its In-Demand Skills 2026 report (February 2026), freelancer earnings from U.S. clients on skills that mention AI grew 109% in 2025, against 23% for other skills. AI video generation and editing grew 329%, AI integration 178%, data annotation 154%, AI image work 95%, and AI chatbot development 71%. Upwork also said the most-demanded skills overall barely changed: full-stack development, virtual assistance, data analytics and graphic design. Its word for the shift is "recomposition," not collapse.

On Fiverr, its Business Trends Index (June 2026) compared global searches from November 2025 to April 2026 with the six months before. Searches for "Claude Code" specialists rose 938%, n8n AI automation 125%, vibe coding 61% and AI voice agents 49%.

Upwork’s own business shows the same split. In the second quarter of 2026, its total gross services volume fell 3.6% year over year while AI-related work rose more than 22%, and "AI strategy and consulting" grew more than 50% (company results as reported by Staffing Industry Analysts and Barchart).

The Census Bureau’s Business Trends and Outlook Survey asks employer businesses every two weeks whether they used AI in any business function. On the current question, the share rose from 17.3% in the first reading (published December 2025) to 24.6% in the cycle published October 22, 2026. Firms with one to four employees were at 24.3%, close to the national figure.

Two other Census findings point straight at the opening for a service business. In the Census AI supplement (late 2025 to early 2026, working paper published April 2026), 64% of firms that used AI had made no organizational changes for it, and training staff and building new workflows were each done by only about 15%. And among firms with no plans to use AI, 65% said it did not apply to their business and 22% said they did not know what it could do.

So most owners either have not started, or have bolted a chatbot onto their day without changing how work gets done. Setting up workflows, training staff and showing a skeptical owner where AI fits are exactly the jobs left undone. Private surveys agree on the worry that slows them down: in Goldman Sachs’s 10,000 Small Businesses Voices survey (March 2026), AI users named data privacy and security (50%), lack of technical expertise (49%) and choosing the right tools (48%) as their main challenges.

The pattern across all of this is simple. The closer your work is to producing a generic piece of content, the more AI competes with you. The closer it is to a client’s specific business, data, staff and rules, the more AI becomes your tool instead of your competitor.

Practical ways to climb:

Sell the system, not the artifact. Instead of writing ten social posts, set up the content calendar, the brand voice file, the AI drafting step and the approval flow, then run it monthly. The social media for local shops page shows how.

Pick an industry and learn its rules. A dental office will not hand patient data to a generalist. An operator who knows benefit verification, a business associate agreement and the practice software gets paid for knowing them. See the industry AI desk game plan and Contracts, client data and AI rules.

Add the human check as a feature. Clients worry AI will be wrong. Make review part of the product: you test the chatbot for wrong answers, check the extracted invoice totals, and sign off the translation. The chatbot testing and translation review ideas are built on this.

Move from projects to care plans. A setup fee plus a monthly plan keeps you inside the client’s business as their tools change, which they will. Pricing AI and IT services covers how to structure it.

Teach. Training was one of the two changes AI-using firms made most, and still only about 15% made it. Workshops build trust that turns into setup work. See AI training workshops.

The free help directory lists vendor training and professional groups that teach these. A useful habit: block two hours a week for learning, and spend one of them building something small with a tool you have not used.

Clients have heard big promises. The FTC has taken action against companies that claimed AI results they could not prove, including income claims aimed at small businesses. Honest talk is both the legal line and the better sales pitch, because owners trust the person who tells them what AI will not do.

Say what it does and what it does not. "It drafts the reply; your staff approves it." "It reads the invoice; we spot-check the totals." Never say a system is fully automated or never wrong.

Use their numbers, not industry claims. Measure the before (hours a week on invoices, missed calls a month) and report the after. Do not quote savings percentages you read online.

Say which tools see their data. Name the tools in the proposal and the plan tier you use.

Disclose your own AI use. If AI drafted the training manual or the website copy, say so and say you edited it.

"Here is what I would set up and what it will not do. The AI drafts answers to the five questions your front desk gets most; a person approves anything about prices or appointments. It will get some answers wrong, so for the first month I check every one and fix the instructions. We will count missed calls before and after, and you will see the numbers. If it does not help, we turn it off and you have lost one month’s fee."

When a client asks, "Will AI replace what you do for me?": "Some of it, yes, and I would rather tell you than have you find out. The tools get cheaper and better every few months. What you pay me for is picking the right ones, connecting them to how your office works, keeping your data safe, and fixing things when they break. If a free tool can do part of my job, I will show you how to use it."

Every six months, look at each service you sell and ask four questions. Could a client now get a good-enough version from a free or built-in tool? Has your price or win rate dropped? Are clients asking for something you do not offer? Is there a rule or deadline coming that makes a service more needed? If the first two answers are yes, move that service down to an add-on and lead with something further up the staircase. The Start here chapter shows how to test a new offer with ten conversations before you build it.

Checked November 2026: the Census BTOS files (October 22, 2026 release), Upwork’s In-Demand Skills 2026 report, and the studies named above. Census changed the wording of its AI question in October 2026, so check that later readings are comparable. Re-check the marketplace figures each spring when new reports come out.

🎓 For college students and recent grads

College is a good time to start a small tech business. You have free time in odd blocks, cheap access to software, professors who know people, and a campus full of organizations that need help. But students run into rules that other beginners never meet: visa limits, financial aid formulas, campus policies on selling, and academic integrity codes. Get these wrong and the cost is not a lost client. It can be your visa status, your aid package or your place at school.

This chapter covers those rules in plain terms, then how to build a portfolio and turn side work into a job. It is written for U.S. college students and recent graduates. It is not legal or tax advice, and on visa questions the only answer that counts is the one from your school’s Designated School Official (DSO).

If you are a U.S. citizen or permanent resident, yes. Skip to the tax section. If you are an international student on an F-1 visa, read this section before you take a single dollar, including from a friend.

ICE’s Student and Exchange Visitor Program (SEVP) page on F-1 employment (updated December 30, 2025) says an F-1 student may work on campus up to 20 hours a week while school is in session, and full time when school is not in session or during the annual vacation. It also tells DSOs to terminate a student’s SEVIS record if they know the student is engaged in unauthorized employment. A terminated record means you are out of status and may have to leave the United States.

Off-campus work needs separate authorization: Curricular Practical Training (CPT) during your program, Optional Practical Training (OPT), or, in rare cases, an economic hardship approval. Freelancing for outside clients is off-campus work. No USCIS page we found addresses freelancing directly, but university international offices and immigration lawyers read the rules the same way: without authorization, it is not allowed. The University of Memphis international office, for example, says on-campus employment does not include self-employment such as performing services for other students or working remotely for an off-campus company.

CPT got tighter in 2026. Law-firm summaries report that SEVP sent schools broadcast messages on August 12 and August 24, 2026 that read "integral to the curriculum" narrowly: the training should be a required part of the program, not an optional internship, and DSOs were warned about improper approvals. That includes limits on "Day 1 CPT" programs. Do not plan a business around CPT.

After graduation, OPT is where a business becomes possible. ICE’s practical training page says OPT must relate to your major. Study in the States says employment on post-completion OPT must be at least 20 hours a week, and that you can be unemployed for a total of 90 days. SEVP’s 2010 OPT policy guidance, still posted on ice.gov, lists "self-employed business owner" as allowed on regular post-completion OPT if you can prove the business has the proper licenses and is related to your degree. You must report it to your DSO. The STEM OPT extension is different: it needs an employer enrolled in E-Verify and a signed Form I-983 training plan, and schools and lawyers disagree on whether you can be your own employer. Ask your DSO and an immigration attorney before relying on it.

On July 17, 2026, the Department of Homeland Security published a final rule that would end "duration of status" for F and J students. Instead of staying as long as you remain enrolled, you would be admitted for a fixed period of up to four years and would have to apply to extend. The grace period after finishing a program or OPT would drop from 60 days to 30. It was due to take effect September 15, 2026.

On September 14, 2026, a federal judge in Massachusetts (Judge F. Dennis Saylor IV, U.S. District Court for the District of Massachusetts, case 1:26-cv-13799) granted a nationwide preliminary injunction blocking DHS and ICE from carrying out the rule while the case continues. The government appealed to the First Circuit on September 30. As of NAFSA’s October 2, 2026 update, the rule remained blocked and briefs were due through December 19.

What we could and could not confirm: the injunction is reported consistently by NAFSA, which is one of the plaintiffs, and by several immigration law firms and university international offices. We did not read the court order itself, and ICE’s own employment and practical training pages, checked November 3, 2026, show a notice about the July rule but say nothing about the court action. An appeal can change this quickly. Before you plan travel, a program extension or your post-graduation timeline, ask your DSO what applies to you that week.

For U.S. citizens and permanent residents, freelance income is self-employment income. Three numbers matter.

$400. The IRS says you must file a return if your net earnings from self-employment are $400 or more. Self-employment tax pays Social Security and Medicare, and the rate is 15.3%. This applies even if you are claimed as a dependent and owe no income tax.

$2,000. The 2025 tax law raised the threshold at which a business must send you a Form 1099-NEC from $600 to $2,000, starting with payments made in 2026. The IRS’s 2026 instructions say to report nonemployee compensation of $2,000 or more. A client who pays you $1,500 in 2026 will not send a form, but that $1,500 is still taxable.

$20,000 and 200 transactions. Under the same law, payment apps and marketplaces send a Form 1099-K only when payments top $20,000 and 200 transactions in a year (according to accounting-firm summaries; we did not check the IRS form instructions).

A worked example. Say you take in $4,000 from clients in 2026 and spend $1,000 on software, a laptop repair and printing. Your net profit is $3,000. On Schedule SE, self-employment tax is figured on 92.35% of net earnings: $3,000 × 0.9235 = $2,770.50, and $2,770.50 × 15.3% = $423.89. Income tax may come on top, depending on your total income for the year.

Quarterly payments. No employer withholds tax from freelance pay. The IRS says self-employed people generally pay estimated tax quarterly using Form 1040-ES. In practice that matters once you expect to owe $1,000 or more for the year. A simple habit: move 25% of every payment into a separate savings account and leave it there.

Open a free business checking account or at least a separate personal account for the business, and log income and expenses in one spreadsheet. The Main Street library chapters on small business taxes and bookkeeping basics go further.

The FAFSA uses tax information from two years before the school year. The 2027-28 FAFSA uses 2025 income, so money you earn in 2026 will show up on the 2028-29 form. Self-employment income counts as student income, and the figure used is net profit from Schedule C, not the gross amount clients paid you.

Student income above the income protection allowance raises your Student Aid Index, which can reduce need-based aid. A few thousand dollars of net profit is unlikely to matter much, but a busy year might. Two practical points: keep every business receipt, because legitimate expenses lower net profit, and if your income jumps, ask your financial aid office how it will affect your package before the year closes. They see this all the time.

Many schools ban selling or running a business from residence halls, or require written permission for any student business on campus. For example, Creighton treats operating a business from a residence hall room as solicitation, Monmouth bars using a residence hall room or address for a commercial enterprise, and St. John Fisher requires prior written approval from the Dean of Students for on-campus student businesses. Read your own student handbook’s solicitation section before you post flyers in dorms or advertise in a class group chat.

Also check the rules on using your .edu email address, the school’s name or its logo for business. Most schools restrict it. Use a personal email and your own name. If you build a site for an official student organization, follow the school’s brand guidelines and get the activities office’s sign-off.

The cleanest route on many campuses is a paid campus role that does the same work: the IT help desk, a professor’s lab, the tutoring center, or a student consulting program run by the business school or the local Small Business Development Center. Those count as on-campus jobs and are open to F-1 students within the 20-hour limit. See campus IT help desk, paid research assistant for professors and student AI clinic for small businesses.

The line is simple. Teaching is fine. Doing someone’s graded work is not.

Schools treat paid work done for another student as contract cheating, and many discipline the seller as well as the buyer. It is also illegal in several states. California Education Code section 66400 bars preparing, selling or distributing term papers, theses and similar work for pay when you know or should know it will be submitted for academic credit. Florida, Nevada, New Jersey, Pennsylvania and Washington are among the other states with similar statutes (we confirmed those through secondary legal sites, not every official code). England and Wales made essay mills a crime under the Skills and Post-16 Education Act 2022. Prosecutions are rare. Expulsion and a ruined reference are not.

Using AI changes nothing here. Having a chatbot write a classmate’s essay for money is the same contract cheating, just faster. If you tutor, put a written no-homework policy on your site and send it to parents, as the tutoring that uses AI the honest way page explains.

If you teach or tutor younger students, check the tools’ age rules before you put a minor in front of one.

So a high-school student cannot be handed a Claude consumer account, and a ten-year-old should not use any of these without a parent-managed setup. Plan sessions with minors so you share your screen instead. If anything you build collects personal information from children under 13, the FTC’s COPPA rule applies; see Contracts, client data and AI rules.

Employers and clients both want proof. Small, real projects beat certificates.

Do three real projects. Your own club’s website, a dashboard for a professor’s lab, an automation for a family business. Real users and real constraints make better stories than tutorials.

Write each one up in half a page. The problem, what you built, the tools, and one number: "volunteers now sign up in two minutes instead of by email," "the lab’s weekly report went from a manual afternoon to a click." Ask permission before naming a client.

Show the work. A short screen recording, screenshots, or code on GitHub with a clear readme. Strip any personal or client data first.

Say how you used AI. Employers now ask. "I used Claude to draft the scripts, then tested and fixed them" is a stronger answer than pretending you did not.

Collect two references. A professor or club adviser who saw your work is worth more to a first employer than a stranger’s star rating.

Side work does three things for a job search: it shows you can deliver, it gives you stories for interviews, and it builds a network of people who will vouch for you. A few ways it turns into a job:

The client hires you. Small businesses that liked your work often become a first employer or a reference. Ask before graduation whether they need part-time help.

Your niche becomes your target. If you built automations for dental offices, apply to the companies that sell software to dental offices. You speak their customers’ language.

Campus roles lead somewhere. Help desk work leads to IT support and systems jobs, lab work to research and data roles, and SBDC clinic work to consulting.

For international graduates, a job related to your degree is what OPT is built for. Report each employer to your DSO within the required time, and track your unemployment days.

The student tech side business game plan puts all of this into a semester calendar, and the AI workshops for seniors and websites for student groups pages are two of the best first ideas.

Checked November 2026: ICE’s F-1 employment and practical training pages, SEVP’s 2010 OPT guidance, NAFSA’s case page on the duration-of-status lawsuit (updated October 2, 2026), the IRS self-employed tax center and 2026 Form 1099-NEC instructions, and California Education Code 66400. The duration-of-status case is on appeal and can change quickly: confirm with your DSO before any visa decision. Re-check the 1099-NEC threshold in 2027, when it starts to rise with inflation.

☎️ Free help directory

You do not have to figure out a new business alone, and most of the help you need is free. The federal government pays for mentors and business advisors in nearly every county. Libraries and universities open their doors and databases. Regulators publish plain-language guides. The companies that make AI and automation tools give away training, because people who know their tools buy them.

This chapter is a directory of that help: who each organization is, what it is good for, and when to call it. Every organization named here is real and was checked in November 2026. We do not list web addresses; search for the name, and make sure you land on the organization’s own site. A warning at the end covers the businesses that dress up as free government help.

These four networks are funded through the U.S. Small Business Administration. They are the best first call for a new business, and few people use them.

Your local SBA district office can point you to all of them, and the SBA’s online learning platform has self-paced courses for starting and growing a business. Before your first meeting, write one page: what you sell, to whom, at what price, and how they will find you. You will get far more from an hour with a mentor. If you will need a loan, the Main Street library’s chapter on funding options covers SBA microloans and how to ask.

A tech business gets a lot out of an SBDC in particular. Advisors meet dozens of local owners and know who is struggling with what. Ask yours whether they run a technology or AI workshop you could help teach, and whether they would refer clients who need setup help. The student AI clinic idea is built on that relationship.

Public libraries. Many larger library systems have a business librarian or a small-business desk, and nearly all give cardholders free access to research databases that would cost hundreds of dollars a month to subscribe to: company directories, industry reports, local business counts and consumer data. Use them to build a list of, say, every dental office or insurance agency within 20 miles. Libraries also host free meeting rooms and adult-education programs, which makes them a venue for the workshops in the marketing playbook.

University entrepreneurship centers. Most universities with a business school run an entrepreneurship center or startup program. Many accept community members for events, and some run pitch competitions, coworking space or student consulting projects for local businesses. Some law schools run small-business or entrepreneurship clinics where supervised students help with formation documents and contracts at no charge. Ask the center what is open to non-students.

Community colleges. Many host the local SBDC and run short, low-cost courses in IT support, networking and cybersecurity.

Security is both a risk to your own business and something you can sell. These are free.

CISA (Cybersecurity and Infrastructure Security Agency). Its Cyber Essentials guide is written for leaders of small businesses and small local governments. It breaks security into six areas: yourself, your staff, your systems, your surroundings, your data and your crisis response. It starts with three first steps: back up data, require multi-factor authentication, and keep software patched. CISA also lists a set of no-cost cyber services. Cyber Essentials makes a good outline for a client security review or a staff training session.

NIST Small Business Cybersecurity Corner. The National Institute of Standards and Technology collects quick-start guides, videos and topic pages on phishing, ransomware and multi-factor authentication, with guidance for health and manufacturing businesses. It also has a finder for cybersecurity support programs.

FTC cybersecurity for small business. Under Privacy and Security on the FTC’s business guidance pages: short guides and videos on data security and secure remote access, written for owners rather than IT staff.

If you offer security awareness training or backup and recovery, build your checklists on these. Clients trust "this follows CISA’s guide for small businesses" more than a checklist you invented.

The companies whose tools you will use want you to know them well. These were checked in November 2026; course lists change often.

A useful order for someone starting the AI automation studio plan: an AI fluency course, then one automation platform’s beginner course, then the admin training for whichever office suite your clients use (Microsoft Learn covers Microsoft 365). The AI toolkit chapter covers what the tools cost once you start paying.

None of these are required to start. A certification helps most when you sell to regulated clients or apply for IT jobs. For a first client, three good case studies matter more.

Your chamber of commerce. Membership costs money, but many chambers let you attend a mixer or two as a guest first. It is where local owners meet the people they hire.

Trade and industry associations. If you pick a niche, join its local or state association, or at least attend its events: dental societies, independent insurance agents’ associations, landlord associations, CPA societies. Speaking slots at their meetings are some of the best marketing a specialist can get.

Referral groups. Weekly networking groups where each member brings referrals for the others. They cost time and fees, and work best for services every business needs, like IT support.

How to turn each of these into clients is in the launch and marketing playbook.

Paid filing sites that look official. An EIN from the IRS is free and takes minutes on the IRS website. Sites that charge a fee to "process" it, or to file your LLC, are selling a form you can fill in yourself.

Fake government notices. New LLCs often get official-looking letters demanding fees for "annual compliance" or certificates. Check any notice against your secretary of state’s own site before paying.

"Free" webinars that end in a pitch. The FTC has acted against AI business-opportunity sellers that misled small businesses about earnings, including Air AI in 2026. Free training from a tool’s own maker is fine; a free session that ends with a $3,000 coaching offer and income claims is the pattern the FTC has gone after.

Mentors with something to sell. SCORE and SBDC advisors are not paid by what you buy. If a "mentor" steers you to their own product, find another.

Checked November 2026: the SBA resource partner page, CISA’s Cyber Essentials page, the NIST Small Business Cybersecurity Corner, FTC Business Guidance, the IRS self-employed tax center, and the vendor and association pages named above. Course lists and free offers change often; ISC2’s free program, for example, closed in May 2026. Re-check any offer before you send a client or student to it.