🎯 The business: you build one small piece of software that does one job for one profession, for example turning a contractor’s site notes into a clean quote, or summarizing a counselor’s case notes into the format their agency requires, and sell it by monthly subscription. AI models do the heavy lifting through an API; your work is choosing the job, building it carefully, and, above all, finding the people who will pay.
At a glance
| Money to start | ≈ $90 to $1,940 |
| Time to the first paying customer | 2 to 6 months; many products never get one |
| Hours a week | 10 to 20 alongside a job or classes |
| Skills that matter most | Knowing one profession’s daily pain, basic coding with an AI assistant, selling to strangers |
| What it can pay | About $1,250 a month before tax at 60 paying users, which can take a year or more (worked below) |
Is this for you?
It suits someone who knows one line of work from the inside, a former dental assistant, an HVAC estimator’s daughter, a case worker, and can code well enough to build with an AI coding assistant. The best ideas come from a job you have done, not from a list of trends.
It is a poor fit if you need income soon, or if you dislike selling. Most small software products fail because nobody hears about them, not because the code is bad. Expect months of unpaid building and outreach, and treat any income in the first year as a bonus.
Who pays, and what
A narrow group of professionals with a repeated, annoying task and a budget: contractors, property managers, therapists and counselors, insurance agents, small law and accounting firms, clinic managers, coaches. They pay $10 to $100 a month per user for something that saves them real time every week (an estimate; prices vary widely by field).
The market is crowded, and say so to yourself early. The big AI assistants add features every month, and a general chatbot already does a rough version of most "AI writing" tools for $20 a month. A product survives only when it fits a workflow the big tools do not: the exact forms of one trade, a connection to the software that trade already uses, and rules the general tools ignore. Upwork reported that freelance demand for AI integration grew 178% in 2025, which tells you businesses want AI built into their own tools, not another chat window.
What it takes to start
| AI coding assistant (for example Cursor Plus at $8 a month, or Claude Pro at $20) | $0 to $240 a year |
| AI model API usage while building and testing (an estimate; pay as you go) | $20 to $300 |
| Hosting and database (free tiers cover most early products) | $0 to $300 a year |
| Domain and business email | $20 to $100 |
| LLC and state fees | $50 to $500 |
| Privacy policy and terms (template plus a lawyer’s review, optional) | $0 to $500 |
| All-in | ≈ $90 to $1,940 |
Card payments cost about 2.9% plus 30 cents a sale through Stripe. A "merchant of record" such as Paddle charges about 5% plus 50 cents but collects and pays sales tax and VAT for you, which matters once you sell in many states and countries.
Start to finish: the steps
1. Pick one profession and one task. Write the task as the customer would say it: "I spend Sunday nights turning my notes into quotes." If you cannot name ten people with that problem, pick another.
2. Talk to 50 of them before you build. Ask how they do it today, what it costs them, and what they already pay for. Do not pitch. The figure above shows what to expect.
3. Fake it by hand first. For the first five users, do the task yourself with an AI assistant and send the result by email. If they will not use it free when you do the work, they will not pay for software.
4. Build the smallest version. One input, one output, one payment page. Use an AI coding assistant, but read and test what it writes. Put a human check where the output reaches the customer’s client.
5. Read the AI provider’s business terms. Build on the API and its commercial terms, not a consumer chat account. Anthropic’s commercial terms let you power products for your own users and assign output rights to you, but forbid reselling the raw service or building a competing model, and its usage policy adds duties for some uses (see Rules to check). OpenAI and Google have their own terms; read the ones you use.
6. Set up the business properly. An LLC, a business bank account, a privacy policy that says which AI provider sees the data, and terms of service. See Choosing a legal structure and Contracts, client data and AI rules.
7. Charge from the first real user. A 14-day free trial, then a monthly price. Free users give polite feedback; paying users tell you what is wrong.
8. Watch your AI costs per user. Every request costs money. Know what one heavy user costs you a month and set limits or a higher plan before a few power users eat your margin.
9. Sell where the profession gathers. Trade associations, their Facebook and Reddit groups, their trade shows and the software they already use. See Getting customers.
10. Decide at six months. If you have fewer than ten paying users after honest effort, change the problem or stop. If you have them, add one connection to software they already use, and raise the price for new users.
Pricing
| Solo plan, one user | $19 to $49 a month |
| Small team plan, up to five users | $79 to $199 a month |
| Annual plan | two months free |
| Done-with-you setup for a firm (optional) | $150 to $500 once |
These are example prices. Price against the time saved for the customer, not your costs, and check what the trade already pays for its main software. Never advertise a percentage of time saved or "fully automated" results unless you can prove them; the FTC has acted against AI companies for claims like that.
The math
Take a product a year in, with 60 paying users at $29 a month: $1,740 a month. Costs are about $150 of AI API usage, $40 of hosting, about $68 of card fees (60 payments at 2.9% plus 30 cents), $30 of other software and $200 of marketing, $488 in all. That leaves about $1,250 before tax for 10 to 15 hours a week of support, fixes and selling. It is real money, but reaching 60 paying users takes most solo founders a year or more, and many never get there.
Software subscriptions may be subject to sales tax in some states; check each state where you have customers, or use a merchant of record. See Small business taxes.
Getting customers
Who to approach first. The 50 people you interviewed in step 2, starting with those who said "I would use that." Then your own past coworkers in the trade. They already told you the problem is real, so the ask is small: try it free for two weeks.
The first ten customers. Offer the free trial to everyone you interviewed: expect 10 to try it and 2 to 4 to pay. Post one useful how-to (not a pitch) in three online groups for the trade, and email 40 more members of a trade association’s directory with a two-line note. Expect 3 to 5 more trials from that and 1 or 2 more paying users a month. Ten paying users in four to six months is a good result.
Hi Marcus, you told me in March that quotes eat your Sunday nights. I built a small tool that turns your site notes and photos into a priced quote in your own format, and you check and send it. It is free for two weeks, then $29 a month, and you can cancel any time. Could I set it up with you on a 20-minute call this week?
Channels, best to worst for this customer.
1. People you interviewed and former coworkers: they know you and the problem.
2. The trade’s own online groups and forums: post answers and how-tos, follow each group’s rules on promotion.
3. Trade associations: newsletters, member discounts, a talk at a chapter meeting.
4. Partners: consultants, bookkeepers or software resellers who serve the trade and can recommend you.
5. Software marketplaces and integration directories for tools the trade already uses.
6. Search: pages answering the exact task ("contractor quote template") bring steady trials over time.
7. Paid ads: last, and only once you know what a paying user is worth over a year.
Case studies and referrals. When a user has used the tool for two months, ask for one specific sentence ("I send quotes the same day now") and their permission to use it with their name and trade. Give a free month to anyone who refers a paying user. More in the launch and marketing playbook.
Running it week to week
Answer support emails within a day; at this size, fast replies are your edge over big companies. Check error logs and AI costs every Monday. Talk to two users a week about what annoys them. Ship one small improvement a week and say so in a short changelog. Once a month, test the AI’s outputs on a fixed set of sample inputs, because model updates can change how it behaves.
Rules to check
No license is needed to sell software, but several rules apply. AI provider terms: Anthropic’s commercial terms let you build products for your own users and assign output rights to you, but ban reselling the service as-is and building competing models; its usage policy requires consumer-facing chatbots to say they are AI, requires a qualified person to review AI recommendations in high-risk areas such as legal, medical, financial, employment and education decisions, and sets extra rules for products used by people under 18. OpenAI and Google have their own business terms; read them for the model you use. Privacy: say in your privacy policy which AI provider processes user data. Health data makes you a HIPAA business associate, which needs a signed agreement with your clients and an AI provider that signs one too. Products used by children under 13 fall under COPPA. Copyright: the U.S. Copyright Office says prompts alone do not create copyright in AI output, so promise users only the rights that exist. Sales tax on software varies by state. And the FTC has fined AI companies for claims about accuracy and earnings they could not support.
What goes wrong
You build for six months and nobody signs up; talk first and sell by hand. A big AI company ships your feature for free; stay close to one trade’s forms and software, which they will not bother with. A few heavy users run up your AI bill; set usage limits. The AI gives a customer’s client a wrong figure; keep a human check and say so in the product. A model update changes the outputs overnight; keep a test set and run it after every change. And you lose interest after the launch; plan the first year of outreach before you write code.
Questions people ask
How much does it cost to start a micro-SaaS?
Often under $2,000 in cash: an AI coding assistant, API credits for testing, hosting on free tiers, a domain and an LLC. The real cost is time, usually months of evenings before the first paying user.
Can I build a product on top of ChatGPT or Claude?
Yes, through their APIs under the business terms, not by reselling a consumer account. Anthropic’s commercial terms allow products for your own users but ban reselling the service as-is and building competing models. Read the provider’s usage policy for rules on disclosure, high-risk uses and minors.
How long does it take to get paying users?
For most solo founders, months. Ten paying users in four to six months of steady outreach is a good result. Many small products never reach that, usually because nobody heard of them.
Do I have to charge sales tax on software?
In some states, yes. Rules differ by state and by how the software is delivered. A merchant of record such as Paddle handles sales tax and VAT for about 5% plus 50 cents a sale; otherwise check each state where you have customers.
Can a college student build a micro-SaaS part time?
Building it in college is good practice and a strong portfolio piece. Selling it is a business, so check the rules first. International students on F-1 visas generally cannot run a business or freelance for outside clients without work authorization; some schools allow self-employment on post-completion OPT if it relates to the degree. Ask your international student office and an immigration attorney.
Free help
SCORE mentors and Small Business Development Centers help with business plans, pricing and contracts at no cost. Many universities run startup incubators and free legal clinics for student founders. The AI providers publish free documentation and cookbooks for building on their APIs. More in The AI toolkit and its costs.
AI tool prices, Anthropic’s commercial terms and usage policy, and payment fees were checked in November 2026. AI providers change their terms and API prices often; re-read them before you launch and every few months after.