Ask what a particular number on a report is for and you often get an answer about visibility: "so we know where we are." That is not a use. A metric is useful when a change in it causes someone to do something different, and when you can name in advance what that something is.

Applied honestly, this test eliminates most of what businesses currently measure.

The four properties

1. It maps to an action. Before adopting a metric, complete this sentence: "If this number moves badly, we will ___." If you cannot fill in the blank with a specific action and a specific person, the metric is decoration. It might still be interesting; it is not useful.

2. Someone can influence it. Metrics nobody can affect produce anxiety rather than action. The overall market rate is not a metric for a business; the proportion of quotes you win is. Look for numbers that are downstream of decisions you actually make.

3. It moves quickly enough to learn from. A metric that responds in eighteen months cannot teach you whether last month's change worked. Slow-moving outcomes are goals; you need faster indicators alongside them that plausibly lead to those outcomes.

4. It is hard to improve dishonestly. Any measured number will be optimised, including in ways you did not intend. Before adopting one, ask how you would improve it if you were being cynical, and whether that route would be visible. "Tickets closed" is improved by closing tickets without resolving them. "Tickets closed without reopening within 14 days" is much harder to game.

Where metrics go wrong

Measuring activity instead of outcome. Calls made, emails sent, posts published, meetings held. These are easy to count and easy to increase without improving anything. They are useful only as diagnostics when an outcome metric is already failing.

Measuring the thing that is easy to measure. Website visits are trivial to count and weakly connected to revenue. Enquiry quality is hard to count and strongly connected. The tendency is to track the first and talk about the second.

Averages hiding the distribution. An average response time of two hours can describe a business where most enquiries are answered in ten minutes and a fifth wait a full day. The fifth is where the complaints come from. Medians and worst cases usually carry more decision-relevant information than means.

Ratios without their components. A conversion rate that improves because volume fell is not good news. Always keep the numerator and denominator visible next to any rate.

Metrics with no target or range. A number with no expectation attached cannot be good or bad, so it produces discussion rather than decisions. Even a rough band — "we expect this between 15 and 25" — makes a number actionable.

Aggregating across things that behave differently. One overall figure covering three product lines with different economics tells you almost nothing and hides the one that is failing.

Building a metric properly

Define it precisely enough that two people would compute it identically. Most disagreements about numbers are definition disputes in disguise. Write down the exact inclusion rules, the time boundary, and what happens to edge cases. "Revenue" alone is ambiguous; "invoiced revenue excluding VAT, by invoice date, excluding intercompany" is not.

Name an owner. Someone should be responsible for the number being correct and for explaining it when it moves. Ownerless metrics decay.

Record where it comes from. Which system, which field, which filter. When it looks wrong later — and it will — this is what allows anyone to check.

Decide the review cadence deliberately. Daily attention on a metric that moves monthly generates noise and false conclusions. Match the cadence to how fast the number actually changes.

Set the threshold before you look. Deciding what counts as bad after seeing the data is how organisations talk themselves out of acting.

Leading and lagging, used properly

Most businesses have plenty of lagging metrics — revenue, margin, retention — and few leading ones. Lagging metrics tell you whether the business worked. Leading metrics tell you soon enough to change something.

A leading metric earns its place only if there is a credible mechanism connecting it to the lagging one. "Time to first response" plausibly leads to conversion rate, because a customer who has moved on cannot buy. "Social media followers" has no such mechanism for most businesses, which is why it makes people feel busy without making them richer.

Test the link periodically. If your leading indicator improves for two quarters and the lagging one does not move at all, the mechanism you assumed does not exist.

A modest set

For most small and mid-sized businesses, a genuinely useful core is small:

Seven numbers, each with an owner, a definition and an expected range, will outperform a dashboard of forty. See why most businesses need fewer metrics, not more.

Metric qualification test

  • The sentence "if this moves badly, we will ___" can be completed specifically
  • Somebody in the business can influence it through their own decisions
  • It responds fast enough to evaluate a change
  • The cynical way to improve it has been identified and is visible
  • Its definition is precise enough for two people to compute it identically
  • The numerator and denominator are shown alongside any rate
  • An expected range or target was set before looking at the data
  • It has a named owner

The point of measurement is not to know how the business is doing. It is to be told, in time, that something needs attention. Any number that cannot do that is costing you the attention it consumes.