Turning an AI prototype into a commercial product means adding the six things a prototype never has: a named buyer, a value metric, a price, an onboarding path to first value, proof that reduces risk, and a repeatable way to reach people. Hardening the code makes the product safe; this work makes it sellable. Both are needed, and most founders do only the first.
Here is the sequence I use, in the order that produces revenue fastest.
1. Name one buyer, not a market
"Small businesses" is not a buyer. "An operations manager at a 20–60 person field services company who currently schedules jobs in a spreadsheet" is a buyer. The narrower version tells you what to build, what to write on the homepage, where to find them, and what number they will accept.
The test: can you list twenty named organisations that match your description, and reach a human at ten of them this week? If not, the definition is still too broad to sell against.
2. Find the value metric
The value metric is the unit that grows as the customer gets more value: seats, jobs scheduled, documents processed, calls handled, revenue collected. Pricing that tracks the value metric feels fair and expands automatically. Pricing that ignores it — flat fees, arbitrary tiers — creates a ceiling and a fight at renewal.
For AI products, be careful: the value metric is what the customer values, not what costs you money. Tokens are your cost, not their value. Price against outcomes and manage cost separately with caps and model routing.
3. Price for a decision, not for perfection
Early pricing is a positioning statement, not a spreadsheet output. Pick a number the buyer can approve without a committee — for UK and US SMB software that usually means under £500 a month — and make the value story obvious against it.
Publish it. Hidden pricing costs early-stage companies more deals than a high number does, because it removes you from consideration before a conversation can happen. Three packages, one recommended, plain numbers.
4. Engineer the path to first value
Measure the number of minutes between signup and the moment the customer sees the thing they came for. Then cut it. Pre-fill data, offer a sample workspace, remove optional fields, defer settings, and delete every screen that sits between arrival and outcome.
For AI products specifically, the first output must be good. A weak first generation is fatal in a way that a slow load never is, because it defines the customer's model of what your product can do. Spend disproportionate effort on the quality of the first result.
5. Add proof that removes risk
A buyer's real question is not "is this good" but "what happens to me if this fails". Reduce that risk visibly: a named case study with a number in it, a security and data-handling page, a clear statement about AI providers and training, a money-back or short-term contract, and a real person's name and face behind the product.
Three specific proof assets outperform a page of adjectives: one quantified outcome, one recognisable logo or credible reference, and one artefact that shows how the work is done.
6. Get the first ten customers by hand
The first ten come from direct outreach, communities, partnerships and events — not from content or ads. Do it manually and personally, because the learning from those conversations is worth more than the revenue.
Track two things per conversation: the words they use to describe the problem, and the specific objection that stops them. After ten conversations, both will have converged, and the homepage rewrite that follows usually doubles conversion on its own.
The order matters
Founders typically do this backwards: build more features, then price, then look for a buyer. The result is a product with broad capability and no obvious reason for any specific person to buy it today.
Buyer first, value metric second, price third, onboarding fourth, proof fifth, outreach sixth. Each step constrains the next and prevents the most expensive mistake in early-stage product work — building things nobody has asked to pay for.
What this looked like in practice
At Wocal the product did not start earning until we picked one venue type, priced against bookings rather than seats, and cut the setup flow from eleven steps to three. None of that was engineering. All of it was commercial product work, and it is what took us to a £2.7M pre-money valuation with 300+ venues live.
Read why most AI products never generate revenue for the failure modes, and what an AI product commercialisation lead actually does for who owns this work.
Frequently asked questions
- How do I turn an AI prototype into a commercial product?
- Define one specific buyer, choose a value metric, set and publish a price they can approve alone, shorten the path from signup to first value, add proof that reduces perceived risk, and win the first ten customers through direct outreach.
- How should I price an AI product?
- Price against the outcome the customer values, not against your token costs. Pick a value metric that grows with their usage, publish plain numbers, and manage model spend separately with caps and routing.
- Why does my AI product get signups but no revenue?
- Usually because the path to first value is too long or the first output is not good enough. Both problems are fixed by engineering the first five minutes, not by adding features.
- How many customers do I need before scaling marketing?
- Ten paying customers won by hand, with a converged description of the problem and a consistent reason for buying. Scale acquisition only once the message and the onboarding are proven manually.