Five Numbers That Flatter: When Efficiency Metrics Quietly Reward the Wrong Behaviour

A dashboard is an act of compression, and compression is where distortion enters. Five numbers carried to a management meeting cannot describe an operation; they can only describe the parts of it someone chose to count. The trouble begins when the parts chosen are the ones that move easily — because a metric that is easy to improve is usually easy to improve without improving anything that matters.

What follows is not an argument against measurement. It is an argument against unpaired measurement, and against the habit of treating a directional arrow as evidence of health. Each of the five measures below is defensible in isolation and corrosive in practice, for the same underlying reason: it rewards a local adjustment whose cost is paid somewhere the dashboard does not look.

1. Utilisation rate

The proportion of available hours charged to productive work is the most widely relied-upon measure in professional services and, increasingly, in internal service functions that have imported the logic wholesale. It is also the measure most likely to be gamed without anyone intending to game it.

Utilisation rewards visible busyness. Its natural consequences are familiar: the informal query answered slowly because it cannot be recorded, the junior colleague's development conversation deferred, the process improvement nobody starts because there is no code for it. The measure rises precisely as the organisation's capacity to absorb surprise falls. An operation running at 92 per cent utilisation has little slack in which to be resilient, and no line on the report that accounts for what the last eight points bought.

2. Cost per transaction

Efficient in the narrow sense and dangerous in the aggregate. Cost per transaction falls most reliably when demand is deflected, cases are closed early, and the awkward twenty per cent is pushed into a channel that does not count transactions at all.

Any UK operation with a contact-centre dimension knows this pattern intimately. First-contact resolution falls while cost per contact falls faster; repeat contacts migrate to a different reporting line and become somebody else's volume. The measure looks excellent for two quarters and then plateaus at a level that reflects the accumulated cost of having been chased.

3. Spans and layers, or headcount ratio

Fewer layers and wider spans are legitimate ambitions. Used as a headline target, however, they reward the removal of exactly the roles whose value is hardest to evidence: the experienced generalist, the person who knows why the exception exists, the informal co-ordinator across two teams.

An organisation can hit its ratio and lose a meaningful share of its institutional memory in the same restructuring, and the loss will not appear for eighteen months. When it does, it presents as a rise in escalation, slower decisions and a sudden dependence on contractors — none of which will be attributed to the ratio that caused it.

4. Average handling time

A useful diagnostic and a poor target. As a target it invites the closure of the complex case and its reappearance elsewhere: the complaint that returns, the query that becomes a formal dispute, the customer who calls three times because the first call was efficient rather than resolved.

The tell is straightforward. If average handling time is improving while repeat contact, rework or complaint volumes are also rising, the organisation has not become faster. It has moved the work.

5. Adoption rate for change programmes

Adoption is measurable, trackable and almost entirely uncoupled from benefit. A portal can be adopted; a new process can be adopted nominally while the old one continues in parallel; a tool can be adopted by everyone and used by nobody.

Adoption metrics are attractive because they produce a positive number early, when a programme most needs one. The corresponding risk is that the number becomes the objective, and the double-running that should have been switched off in month four is still there in month fourteen, absorbing capacity quietly.

The pairing test

None of these measures is unusable. Each becomes misleading the moment it is reported alone. The remedy is procedural rather than analytical, and it fits on one page.

For every efficiency metric in a board pack, require three things to be reported alongside it:

• A quality guardrail. A measure capable of getting worse when the efficiency metric improves — rework, repeat contact, error rate, complaint volume, re-opening rate.
• A resilience indicator. Something that reflects the organisation's capacity to absorb variation: backlog age, single-points-of-failure in critical processes, cross-training coverage, time to recover a key service.
• A separate owner. The guardrail should not be owned by the same person who owns the efficiency target. Where one function owns both, the pair will drift towards the flattering half within two reporting cycles.

The pairing test is not a guarantee of good judgement, but it makes bad judgement visible. A pair reported side by side on the same page, in the same typeface, is much harder to present selectively than two metrics living in different appendices.

Three signals that a metric has turned

First, implausible improvement. A measure that improves by more than ten per cent in a single quarter, without a corresponding change in the operating model, is more likely to have been redefined than transformed. Ask what changed in the definition before accepting what changed in the number.

Second, improvement without a beneficiary. If nobody outside the reporting line can describe how the improvement was felt — by a customer, a colleague in another team, a supplier — then the gain is internal and possibly circular.

Third, the uncomfortable publication test. Would the leadership team be content to publish this metric next to its guardrail, in a document a customer could read? If the answer is no, the pair is doing its job and the metric is telling you something you would rather not have heard.

Retiring a number is a governance act

Metrics accumulate because removing one requires somebody to admit it was not measuring what it claimed. That admission is unpopular and, in most organisations, unowned. The practical answer is to attach a review date to every measure when it is adopted and to ask, at that review, a single question: has anything we care about got better because of this number?

Where the answer is no, retiring it is not an admission of failure. It is the governance function working as intended — and it is usually the cheapest improvement available to a management team that has grown suspicious of its own reporting.