Find usability problems before your customers do

There is a quiet, unfashionable technique that catches most of the obvious faults in an interface in a single morning. A heuristic evaluation is a structured review against a set of recognised usability principles, carried out by two or three experienced people who examine the product independently and then combine what they found. It is cheap, quick, and can be run at almost any stage of a project, which is exactly why it is so often skipped in favour of something that sounds more impressive.

It works because experience compresses. A reviewer who has watched hundreds of people struggle with the same handful of mistakes can look at a screen and see the trouble coming, the way a builder can look at a wall and know it will not hold. The principles they test against, Nielsen's ten among them, are simply that accumulated experience written down. Is the system's state visible. Does it speak the user's language. Can a person undo a mistake. Is the design consistent with itself and with the world.

What it is not, and must never pretend to be, is a substitute for watching real customers use the thing. Experts find problems of craft. Only real users find the problems that come from being an actual person, with an actual goal, who does not care how the product was built. The two methods answer different questions, and a team that runs one and calls it the other is fooling itself in a way that usually shows up later, in the numbers.

This is the real argument for doing the expert review first. Time with genuine customers is scarce and precious. If you spend it watching people stumble over a broken label or an unlabelled icon that any experienced reviewer would have flagged in an hour, you have burned that scarce resource on problems you already knew how to find. Clear the obvious faults cheaply, and the expensive research is free to surface the things only real use reveals.

The method rewards a little discipline. Have each reviewer work alone before comparing notes, because independent passes find more than a group talking itself towards consensus. Rate every issue by severity, so the team knows what to fix now and what can wait, rather than treating a cosmetic nitpick and a broken checkout as equals. Then carry the questions that remain, the ones about real behaviour, to real people.

Spending a research session watching people trip over faults you could have found in a morning is a waste of the most valuable thing you had.

It is tempting to imagine handing this job to a model. Point an AI at the interface, ask it to list the usability problems, and skip the experts. In practice the machine is a useful pair of eyes and a poor final judge. It can flag inconsistencies, missing states, and obvious violations quickly and tirelessly, which is genuinely helpful for clearing the shallow layer. What it cannot yet do is weigh severity in the context of your particular customer, your business, and what actually matters to them.

So the shape of the work is shifting rather than vanishing. The routine part of a heuristic review, the first sweep for the obvious, is becoming something a machine can accelerate. The valuable part, deciding which faults will cost you and which are noise, and knowing what to test with humans next, is exactly the seasoned judgement a model does not have. Used well, the tools make the expert faster. They do not make the expert unnecessary.

There is a second, quieter benefit that teams discover only once they do this regularly. A structured review builds a shared language for quality. When a team has argued through why a particular screen fails a principle, they stop making the same class of mistake, because they have internalised the standard rather than been handed a list. The review stops being an audit performed on the work and becomes a way of thinking that travels into everything the team builds next.

A structured expert review remains one of the highest-return activities available to a team, precisely because it applies hard-won experience quickly and before anything has cost you. It will not tell you everything, and it is not meant to. What it does is clear the ground, so that the harder, more expensive work of learning from real people can be spent on what only real people can teach you.

That is the real return, and it is why the technique rewards experience so heavily. Anyone can hold a checklist against a screen. The value is in knowing which failures will actually cost the business and which are harmless, a judgement that comes only from having watched the consequences play out many times. Bringing that seasoned eye in early is less an expense than a way of buying back the time you would otherwise spend learning the same lessons the slow and public way.