Your AI feature shipped. Nobody’s using it.

I’ve watched a lot of AI features get built over the last couple of years. Most of them shipped on time. A good number work exactly as they were specced. And a surprising share of them are barely touched three months after launch.

MIT looked at this recently and put a number on it. Ninety-five percent of enterprise AI pilots never move the business. When I first read that I assumed it was a model problem, or a data problem. It usually isn’t. The models are good enough. The data is fine. What’s missing is someone senior deciding what the thing should feel like to use, and holding that line while everyone else races to ship.

The pattern I keep seeing

A company gets excited about AI. Fair enough, so is everyone. The engineers are keen, they wire up a model, and within a few weeks there’s a working feature. The PM writes it up, it goes in the release notes, and the team moves on to the next thing. Then the usage numbers come in flat, and nobody can quite explain why.

The reason is almost never the part everyone was watching. It’s the bit nobody owned.

Think about what you’re actually asking a person to do

An AI feature hands someone an answer they didn’t write, from a process they can’t see, and asks them to trust it enough to act on it. That’s a big ask. If the interface doesn’t help them understand where the answer came from, how sure it is, and what to do when it’s wrong, most people quietly stop using it. They go back to the spreadsheet. They don’t file a complaint. They just don’t come back.

None of that is an engineering problem. It’s a design problem, and a senior one. A junior designer can make a screen look clean. Deciding how much of the model’s reasoning to show, when to let a person overrule it, what happens on a bad answer, how to win back trust on the second use after you lost it on the first, that takes judgement. It comes from having shipped things that failed and working out why.

I learned this the slow way

At Seek I was the only designer on a mobile rebuild, and the numbers only moved once we stopped adding features and started making the ones we had make sense. At Coles I watched platforms that were technically fine go nowhere because nobody owned the experience end to end. The technology has changed a lot since then. The lesson hasn’t. People adopt things they trust, and trust gets built in the interface, one interaction at a time.

The good news is it’s fixable

This rarely needs a big team. It needs one experienced person, in the room, early, asking the questions nobody else is paid to ask. What does this look like when the model is wrong? What’s the first thing a new user sees? Why would anyone come back on day three? Those aren’t polish questions. They decide whether the thing gets used.

It’s a large part of why I work the way I do. Most companies at this stage don’t need a full-time head of design, and can’t really justify one. What they need is someone senior for a day or two a week who owns the design of what’s being built, works next to the team, and is still there next quarter.

And it isn’t only my read on it. McKinsey tracked three hundred companies over five years and found the ones that took design seriously grew revenue about a third faster than everyone else, with much stronger returns for shareholders. Retail banking was one of the industries they looked at, which is to say this holds in the complex, regulated products where people assume design is a nice-to-have. It isn’t.

So if you’ve shipped an AI feature and the numbers are flat, ask one question before you blame the model or the market. Who owned the experience? If the honest answer is nobody, you’ve found your problem. And it’s a good one to have, because it’s the most fixable of the lot.