Test Before You Scale

The idea tested brilliantly. Heads nodded, the sticky dots clustered on the good option, someone in the corner said game-changer. Six months and a serious budget later, hardly anyone was using the thing. Nobody had lied in that workshop. The workshop had simply measured itself: the enthusiasm of people in a room being asked whether they liked something, which turns out to predict almost nothing about what those same people do on a busy Tuesday when nobody is watching.

This is the central problem a behavioural test exists to solve. Liking is cheap. It costs nothing to nod at a good idea. Doing is expensive, because doing means giving up the old behaviour that was quietly working for you. A test earns its keep by catching that gap while it still costs a prototype to fix, before it costs a rollout.

An idea rarely depends on a single behaviour. It depends on a chain of them, and the chain is only as strong as its weakest link. People have to understand what to do. They have to trust it enough to rely on it. They have to take the first action, then a second one next week, and eventually others have to copy them. Comprehension, trust, first action, repeat action, social uptake. A break at any point stops the whole thing, and scale simply multiplies the break.

Most tests are accidentally kind. They use fresh, motivated users. They run in a quiet fortnight rather than at month-end. There is usually someone helpful hovering, ready to smooth over any confusion. Kind conditions produce kind results, and kind results are a comfort that costs you nothing now and a great deal later. A good test is unkind on purpose.

Being unkind means importing the reality the rollout will actually face. Tired users. A day with five other things competing for attention. A sceptical colleague in the room. And above all, the old way still available, one tab or one habit away, because in real life the incumbent behaviour never politely removes itself to give the new one a fair run. If the behaviour survives all that, it has earned the right to scale. If it only works when the world is gentle, it does not work.

You do not need the whole system built to test the behaviour. You need the moment staged. A prototype exists to fake everything except the one action you care about. One mocked-up screen. One rewritten letter. One changed shelf. One script for the first difficult conversation. If the behaviour breaks at that staged moment, the full build would have broken there too, at a hundred times the cost.

The most useful version of this is a person behind the curtain, quietly doing what the eventual system would do. Users experience it as real, which is the only condition that matters, and you learn how they behave before a single line of the real thing is built. A food truck is a restaurant tested as a prototype: same food, same customers, a fraction of the lease. The menu that survives the truck has earned the building.