AI in your product: where do you start?
Your customers are asking for it, your competitor already claims it and your board wants a plan. But between the pressure to do something with AI and a feature that delivers real value sit a number of choices. A guide for whoever has to decide.
Don't start with AI, start with the friction
The worst reason to build AI into your product is that you can. Features that start from the technology end up as a demo on a sales slide: impressive in the presentation, unused in practice. The right question is the same as for any product decision: where do your users lose time, where do they drop off, which work do they keep postponing because it is tedious.
In those places AI is often gold, precisely because the work is dull, repetitive or language-heavy: summarising a file, preparing a reply, filling in what is already known, pointing to the right next thing. In SARA, the care assistant we built the software for, the value is not in how clever the system is, but in the calm it gives care workers at their busiest moments. That is the right order: first the problem, then the technology.
Small proof beats a big plan
AI features have a property classic software does not have: you only know whether they are good enough once you try them on real cases. Paper is patient, a language model is not. So the most important first step is not a requirements document but a working prototype, built in days, tested with real users and real data. That no longer takes a big budget: the Kompas on our own site, an assistant that points visitors to the right route, was live in days.
Such a prototype gives you exactly the information you are short of as a decision maker: is the quality good enough, will users actually use it, and what does it cost per use. With those three answers, the investment decision suddenly stops being a matter of faith. Without them, every big AI plan is a gamble with a nice cover.
Starts from the technology
- impressive in the demo
- unused in practice
- a big plan without evidence
Starts from the friction
- where users lose time
- a prototype on real cases
- a feature with an owner
Make the business case honest, including the risks
An AI feature that gets it right 95 percent of the time sounds impressive until you realise what the other 5 percent means for your product. A wrong movie suggestion costs nothing, a wrong medical recommendation costs everything. The real business case is not only about what the feature delivers, but also about the cost of the mistake: how often it goes wrong, how bad that is, and who catches it. From that follows the most important design choice: where the human stays ultimately responsible.
Count the ongoing side too. An AI feature is not something you deliver and forget: models get replaced, behaviour drifts, quality needs monitoring. Whoever budgets only the build cost gets a surprise in year two. It rarely is a showstopper, the gains are usually far bigger, but it belongs in the sum.
From pilot to production: where most journeys strand
The world is full of successful AI pilots that never reached production. The cause is almost never the technology. It is ownership: the pilot belonged to the innovation team, but nobody in the line felt responsible for the feature once it had to run for real. Or it is data: the pilot ran on a hand-made export, and nobody had budgeted what it costs to make that flow structural.
If you want your AI feature to make it, treat it as a product from day one: an owner who still owns it next year, a place on your roadmap, agreements about quality and a team that maintains it. That sounds heavier than a pilot, and that is exactly the point. A pilot is allowed to fail. A product feature has to work.
What it comes down to
AI in your product does not start with technology but with a sharp choice: which problem, which proof, which responsibility. Guarding that order is exactly what we do, from first prototype to feature in production. Curious what is possible in your product?
Schedule a meeting