Ask a large organisation how many pilots it is running and you will usually get an estimate rather than a number. Ask how many pilots it has formally concluded in the last two years — adopted, scaled, or stopped — and the number is generally small enough to name individually.
This is not a failure of ambition. These organisations are running more trials than they ever have. What they have not built is any mechanism for a trial to end.
Extension is the only costless decision
Consider the incentives at the point where a pilot reports a positive result. Adopting it means committing budget, changing a process people are used to, and owning the outcome if it disappoints at scale. Killing it means explaining why the last four quarters were spent on something now being abandoned. Extending it — running another phase, adding a second site, gathering more evidence — requires no defence at all, because more evidence is always a reasonable thing to want.
So the pilot is extended. Then extended again. Two years on it is a small permanent operation serving a fraction of the organisation, staffed by people whose job title still describes it as temporary.
The perverse detail is that this happens to successful pilots more reliably than to failing ones. A pilot that clearly fails gets stopped, because stopping it is easy to justify. A pilot that clearly succeeds creates an obligation to act, and acting is the expensive part — so the successful pilot enters a holding pattern that the failed one is spared.
The fix is not more rigour in the trial. Organisations already over-invest there — elaborate success metrics, careful controls, dashboards nobody disputes. The rigour is all at the front. What is missing is at the other end: a date, and a rule stating what happens on that date.
Written before the pilot starts, that rule takes about a paragraph. If the result clears this threshold we adopt it on this timetable and this named person owns the rollout. If it does not, we stop and write down what we learned. If we cannot tell, we run one more defined phase and then choose regardless. The specifics matter far less than fixing them in advance, when nobody yet knows which branch they will be arguing for.
This is the same structural problem I have written about in a different costume: an organisation that cannot agree what a term means cannot act on a measurement of it, and one that never defined what "worked" would trigger has not really measured anything either. It is also why so many companies cannot say whether their AI systems are working — the evaluation was designed to produce a number, not a decision.
There is a management dimension too. Ending things is exactly the sort of judgement that requires someone with enough context to make it and enough standing to carry it, which is the layer a decade of flattening removed and is now being rebuilt. Nobody wrote that on the reorganisation slide.
The organisations that escape this are not braver. They simply wrote the ending down at the beginning, when it was cheap.
Topics opinionmanagementgovernance
