Every technology that arrives with this much noise attached deserves two things: a fair hearing, and a sceptical one. AI in digital product development is genuinely useful. It is also being oversold, and the gap between those two statements is where a lot of budget is currently being lost.

So let's take both halves seriously.

Where AI genuinely earns its place

It removes the repetitive work. Quality assurance testing, bug detection, the parts of the development cycle that are essential and dull. Automating those doesn't just save time - it reduces human error in exactly the places human error creeps in, which is the ninth hour of doing something monotonous.

It makes sense of data at a scale people can't. Patterns in user feedback, trends in market data, signals that would take an analyst weeks to surface. That supports genuinely better decisions about what to build, how to price it, and where demand is heading.

It enables products that adapt. AI can analyse behaviour and preferences so a product learns from how it's actually used. Recommendations that improve, interfaces that respond, support that anticipates. Done well, that's a real improvement in the experience rather than a gimmick.

Taken together, that's meaningful. Faster cycles, better-informed decisions, more responsive products.

Now the challenges - and these are the ones that decide whether any of it works

Your data. AI is only as good as what it learns from. Incomplete, inconsistent or biased data doesn't produce cautious output - it produces confident, plausible, wrong output. Which is considerably more dangerous than no output at all, because someone will act on it. If your data foundations are shaky, AI will not paper over them. It will amplify them.

Bias. If the data reflects existing inequities, the model will reproduce them and give them the appearance of objectivity. That's not a theoretical risk in a product that makes decisions about people.

Explainability. When an AI-driven feature makes a call, can you explain why? In regulated industries this isn't optional, and in unregulated ones your customers will still want to know. "The model decided" is not an answer anybody accepts twice.

Skills and integration. The technology is easy to buy and hard to embed. It needs people who understand it, processes that accommodate it, and systems it can actually reach. Most failed AI projects I've seen didn't fail on the model. They failed on everything around it.

Bad data doesn't make AI cautious. It makes AI confidently wrong - which is far more dangerous than no answer at all.

The measurement problem

Here's the one I'd press hardest on, because it's the one most often waved away.

How will you know if it worked? A great many AI initiatives are launched with a benefit that's asserted rather than measured - it will make us more efficient, it will improve the experience - and are therefore impossible to evaluate honestly. Which conveniently means they can never be judged a failure.

Before you build, define what changes if this succeeds, and how you'd see it in the numbers. If you can't answer that, you're not running a project. You're running an experiment, and you should say so out loud, because experiments are allowed to fail and projects aren't.

Where that leaves you

AI in product development is worth doing, and it's worth doing in a specific order: get your data house in order, pick a problem that's real and measurable, prove it small, then scale.

The firms that struggle are the ones doing this backwards - buying the capability first and hunting for a use case afterwards. That approach produces demos, not products.

So the question. What's the first thing you'd build with AI, and how would you know within ninety days whether it had worked? If you have a clear answer to both, you're ready. If you have a clear answer to only the first, that's the gap.

Happy to talk it through - book a short discovery call with the team at Distinction. No pitch, just an honest read on where AI would actually pay for itself.