Built Your SaaS With AI? Here’s Why It Needs a Human in the Loop Now
Built Your SaaS With AI?
Here’s Why It Needs a Human in the Loop Now
AI can take a SaaS idea to a working product in weeks. We know, because our own product, Stempy, started that way. What AI can’t do on its own is tell you where an AI-built SaaS is quietly losing money. That takes a human in the loop.
The AI-first version of Stempy
Stempy gives independent UK businesses, such as cafés, barbers and salons, loyalty stamp cards that live in Apple Wallet and Google Wallet. Customers scan a QR code at the counter. No app to download.
The first version was built AI-first, and it worked in demos. It was also two separate codebases: a web app with its own database, and a marketing site where even the blog posts were written into the code. Every small change needed a developer, and two systems meant two lots of hosting and maintenance.
What the human in the loop found
We rebuilt Stempy on WordPress. AI still wrote most of the code, but an experienced engineer made every product decision and tested each flow on staging before it went live. That person caught problems nobody had asked the AI to look for:
- A revenue leak. A business account without an active subscription could still use most of the product. Now nothing works without a plan.
- Plans that match the price. What each plan included wasn’t tied to what customers paid. Each plan is now linked to its Stripe price and unlocks exactly what it promises: customer limits, staff and branding.
- A hidden security risk. Business owners upload their logos. Every image is now saved again as a clean PNG, so a disguised file can’t run on the server.
- A quiet bug. Adding a staff member from the admin failed even when the form was filled in correctly. The automated tests passed. A person using it spotted it.
None of these would have crashed the app. All of them would have cost money or trust.
Leaner, and more profitable
Today Stempy runs as one codebase on one server: the marketing site, the merchant portal, Stripe billing and the wallet passes. Website pricing comes straight from Stripe with one click. Pages, blog posts and SEO are edited in WordPress, not in code. Spam protection and security use free, proven tools.
More than 370 automated tests, including 85 browser tests, run before every change is committed. The result is a product that costs less to run and no longer gives away what customers should be paying for. For a small SaaS, that is what profitable looks like.
If you built your SaaS with AI
You were right to use AI. It’s why Stempy can be priced for small businesses. But AI optimises for “it runs”, not “it makes money”. Someone still has to ask the commercial questions: who can use this without paying, what happens when it fails, and what it costs every month.
That’s what we do at 90sType. AI does the heavy lifting, and a human stays in control. If your AI-built SaaS works but you’re not sure it’s ready for paying customers, see how we work.