Where AI actually earns its place in a product
Mykola Melnyk
Head of Design Alty
June 10, 2026
Say you've decided AI belongs in your product. The harder question starts there: where in the experience does it actually help, and where does it just get in the way?
I design products for a living, and I've watched AI land in both places. Used well, it removes work the user never wanted to do. Used badly, it takes away the control they did want, in the name of being helpful. The difference usually isn't the model, it's the part of the product you aimed it at.
Here's how I think about it.
Put it where the work is invisible and nobody wants it
The clearest wins are the tasks users do because they have to, not because they want to. Sorting a transaction feed into categories. Pulling the three relevant numbers out of a statement. Flagging the one payment in a thousand that looks wrong. People are glad to hand this off, because doing it themselves was never the point.
AI fits cleanly here because the user's goal sits on the other side of the chore. They want the categorised feed, not the act of categorising. When the boring middle just happens and the result comes back, the product feels faster, not stranger.
Keep it away from the moments people want to own
Then there's the other kind of task, where doing it is the point. Deciding how to split a payment. Choosing what to invest in. Setting a limit on their own spending. The user wants to feel in control here, and an AI that steps in to decide doesn't read as helpful. It reads as being managed.
I've seen teams put their smartest model on exactly these decisions, because that's where it looked most impressive in a demo. In real use it backfired. People don't want their money decisions made for them by something they can't question. The model that won the room in a pitch quietly cost trust once it was live.
The rule I'd give a team: automate the path to the decision. Get the user to the choice faster, with everything they need in front of them. Leave the choice itself to them.
Make it explainable, or don't ship it
In a fintech product, "the system decided" is not an answer a user accepts, and often it's not one a regulator accepts either. If an AI feature touches someone's money, sooner or later they'll ask why it did what it did. If the honest answer is "we're not entirely sure," the feature isn't ready, however good its numbers look.
That filter rules out more than teams expect. I've come to like it for exactly that reason. And showing the reasoning, in a way a worried person can follow at the moment they're worried, turns out to be a design problem as much as a technical one.
The feeling you're aiming for
The best AI in a product is the kind users never clock as AI. There's nothing in the menu called "AI✨". They just notice that the thing which used to take five steps now takes one, and move on without a second thought about how.
So the bar I hold it to isn't "does this show we use AI." It's whether the person using the product notices anything beyond it getting easier. When the answer is yes, it was aimed at the right place. When a product starts feeling clever at the user instead of useful to them, someone aimed it at the wrong one.
I've spent enough years designing financial products to know this shows up in small moments, not in the launch post. The teams whose AI quietly earns its place are usually the ones who argued hardest about where not to put it.
Figuring out where AI belongs in your product's experience? Let's talk.
