The give-back test.

Why do adoption metrics mislead?

Because most of them measure activity, not change. Licences issued. Logins recorded. Pilots launched. Training sessions attended. Every one of those numbers can be excellent while nothing in the organisation actually works differently.

We have watched dashboards glow green over programmes that were quietly dead. The gap between reported adoption and real adoption is where most AI investment goes to disappear.

What does the give-back test measure instead?

Dependence, in the good sense. Whether the capability has crossed the line from something people are asked to use into something their work now assumes.

Think about email, or search. Nobody holds adoption workshops for them. Suggest removing them and you would have a mutiny by lunchtime. That is what full adoption looks like: the technology has become ordinary, and ordinary is the destination. The give-back test simply asks whether your new capability has arrived there yet.

How do you actually run it?

Not by ripping tools out of production. You ask the question, seriously, at review: “If we removed this next month, what would happen?” Then you listen carefully.

Fighting sounds specific: “my week stops working”, “we would have to rehire for that task”, “the backlog would double”. A shrug sounds polite: “it’s quite useful”, “some of the team like it”. Specific pain means embedded. Politeness means theatre.

Watch the quieter proxies too. Do people extend the capability without being asked? Do they complain loudly when it breaks? Do they train each other? All three are the sound of something that would not be given back.

What do you do with a shrug?

Diagnose it honestly. Usually the cause is one of four: the tool does not fit the real workflow, people do not trust the output, nobody invested in confidence, or the use case was never real and everyone privately knows it.

Then fix it or kill it. Killing a hollow pilot is not failure; it is judgement, and it frees money and attention for the capabilities people would genuinely fight for. We run our own engagements against this test, and we tell clients to hold us to it.

Common questions

Is the give-back test scientific?

It is a deliberately blunt behavioural signal. Pair it with outcome metrics. Its value is that it is almost impossible to game.

When should you run it?

Around 90 days after rollout, then quarterly. Novelty can carry a weak tool for a month. It cannot carry it for a quarter. The short-range version of the same question is the Next Monday test: ask it a week later, when the enthusiasm has gone and the task has come back.

What if managers value it but their teams do not?

Then adoption is thinner than the dashboard suggests. Ask the people who do the work, not the people who bought the tool.