The Metric That Made Everyone Do the Wrong Thing

You introduced a target for time to first response. Within a month the average halved and everybody was pleased.

What actually happened is that people started sending an acknowledgement within four minutes and dealing with the substance three days later. The number improved and the client experience got worse.

People optimise what you count

This is not cynicism about staff. It is the predictable result of telling a team that one number matters, and it happens with good people acting in good faith.

A measured quantity becomes the definition of doing the job well, because that is the only signal anybody receives about whether they are performing. Everything not measured becomes, functionally, optional.

So the question is never whether people will optimise the number. They will. The question is whether optimising it produces the outcome you wanted.

Four that reliably backfire

Time to first response. Produces fast, empty acknowledgements. Measure time to useful reply, and separately the proportion resolved in one exchange.

Jobs completed per day. Produces cherry-picking of easy work and a growing pile of the difficult jobs nobody will touch. Check the age of the oldest open item alongside it.

Utilisation. Produces work being made to fill the time available, and it punishes exactly the efficiency you are trying to encourage. A person who finds a faster way is rewarded with a worse utilisation figure.

Number of calls or activities. Produces activity, reliably, and almost no relationship to revenue. The salespeople gaming this one are frequently the ones doing worst.

Every measure needs a partner

The general fix is that a single number is always gameable and a pair usually is not, because the two pull in opposite directions.

Speed pairs with quality. Volume pairs with the age of what is left. Cost pairs with rework. Utilisation pairs with margin. New business pairs with retention.

State both, together, every time. A team told to be fast will be fast. A team told to be fast without the resolution rate falling has been given a real instruction, and it is not much harder to say.

Where the number came from matters

A measure imposed from above gets treated as a hurdle. The same measure chosen by the people doing the work gets treated as feedback.

That difference is worth more than the choice of metric. If a team picks how their work should be judged, they will pick something honest, because they are the ones who have to live with a number that misrepresents them. Ask them, and expect the answer to be better than yours.

What AI changes about measurement

The reason businesses measure gameable things is that those were the only things countable. Response time, call volume, jobs closed. All easy because they are events with timestamps.

What actually matters is usually a judgement about quality, and judgement meant a person reading everything, which nobody could afford. So the proxy became the measure.

That constraint has gone. An AI step can read the actual exchanges and tell you whether the fast reply answered the question, whether the closed job came back a fortnight later, whether the enquiry was handled well or merely quickly. That is the quality half of every pair above, and it is the half that has always been missing.

Two consequences. You can measure the thing you care about rather than a proxy for it. And gaming becomes much harder, because a system reading the substance is not fooled by a four minute acknowledgement containing nothing.

One caution that matters. Anything measured on individuals rather than on the process will be experienced as surveillance, and a team that feels watched will optimise against the watcher rather than for the client. Keep it at the level of how the work flows, not who did what.

How to introduce one without the damage

• Measure it quietly first. Watch it for a month before telling anybody it is a target, so you know what normal looks like unmanipulated.

• Name the partner measure in the same sentence, always.

• Say what you will not accept. Faster, with the resolution rate holding above where it is now. That closes the obvious workaround before anybody finds it.

• Ask what would make it look good dishonestly. Put that question to the team openly. They will tell you, and they will respect being asked.

• Review it in six months and be willing to retire it. A measure that has done its job should be allowed to stop.

Three numbers, not fifteen

The other failure is the opposite of gaming, and it is more common in businesses that have taken measurement seriously.

A team given fifteen metrics has been given none, because nothing is prioritised and nobody can hold fifteen things in mind while doing the job. What actually happens is that people privately pick the two they think matter, and you have lost control of the choice.

Three is about the limit for any one person: one for volume, one for quality, one for the thing that would otherwise be sacrificed. If a fourth genuinely matters, something else comes off the list.

Never pay against a single number

Everything above becomes considerably more forceful the moment money is attached.

A metric that shapes behaviour mildly when it is reported in a meeting shapes it completely when it determines a bonus, and every weakness in the definition gets found within a quarter. If a measure is going into a compensation plan, assume it will be optimised to its absolute limit and design it on that assumption.

The practical rule: a bonus tied to one number is a bonus tied to whatever that number fails to capture. Pair it, cap it, or keep the money out of it.

The one to watch for

If a number improves sharply and nothing else in the business changes, the number is being managed rather than the work.

Real operational improvement shows up in several places at once: the metric moves, complaints fall, rework drops, somebody mentions the job is easier. A figure that improves alone, cleanly and quickly, is almost always a definitional change rather than a genuine one.

That is not a reason to distrust your team. It is a reason to check what the number is actually counting before you celebrate it.

AI Optimize measures whether the work was done well rather than only how fast it happened, which is the half of the pair most businesses have never been able to count. That work sits under Reporting & Data.