Metrics that work well inside an analytics team frequently fall apart in a board meeting. Not because the board is unsophisticated, and not because the metric is wrong. They fall apart because a board meeting applies a set of pressures that internal review never does, and metrics are rarely designed with those pressures in mind.
Having sat through a reasonable number of these, the failure modes are predictable enough to design against.
The metric has to survive a hostile question
Somewhere in the room is a person whose instinct, on seeing a favourable number, is to ask what is being excluded. This is a good instinct and it is their job.
A metric survives that question when the exclusions are stated before they are asked about. A metric dies when the presenter discovers the exclusion live, or worse, does not know it exists. We have watched a customer growth figure lose all credibility when it emerged that reactivated accounts were counted as new, a decision made two years earlier by someone no longer present.
The defensive practice is straightforward. For every board metric, write down what is excluded and why, and put the material exclusions on the slide. It looks like weakness and it functions as armour.
Definitions must be stable across the reporting period
Board metrics are read as time series. A number is meaningful relative to last quarter. Which means a definition change mid-series is not an improvement, it is a break, even when the new definition is better.
Teams change definitions for good reasons and then present the series as continuous. The chart shows a step change that reflects methodology rather than the business. If that is discovered in the room, the credibility damage extends well beyond the metric in question.
Restate history when a definition changes. Show both series for at least one period. Annotate the chart at the point of change. This is standard practice in financial reporting and almost absent from operational metrics, which is odd given that the same people read both.
Someone will ask what drove it
The follow-up to any movement is always why. Not the statistical explanation, the causal one. Revenue moved four percent. What happened.
A metric that cannot be decomposed on the spot puts the presenter in the position of promising to come back, which converts a decision meeting into a homework meeting. Boards notice which functions consistently do this.
Design decomposition in from the beginning. Every headline metric should have a standing breakdown along the two or three dimensions people always ask about, prepared before the meeting whether or not it is presented. The preparation is cheap. The alternative is expensive in a way that compounds.
Fewer metrics, held longer
Board packs accumulate. A metric is added because a question was asked once, and it is never removed because removing it invites the question of what is being hidden. Twelve pages becomes forty. Attention is finite, so the effect is to reduce scrutiny of everything.
The functions we see doing this well hold a deliberately small headline set, five to eight measures, and defend it. Everything else is available on request in an appendix. New metrics get added only by displacing something, which forces a real conversation about what matters.
Longevity matters as much as brevity. A metric tracked for three years teaches the room what normal variation looks like. A metric introduced last quarter carries no such intuition, so every movement looks potentially significant and discussion is poorly calibrated.
Show the uncertainty
Operational metrics are almost always presented as point estimates. Many of them should not be, particularly anything derived from sampling, attribution or forecasting.
The usual objection is that boards cannot handle intervals. In our experience boards handle them perfectly well when they are presented plainly, and handle badly the experience of acting on a precise-looking number that turns out to have been approximate.
A range with a stated basis invites better questions than a single figure with hidden error. It also protects the analytics function when the number moves within noise, because the range already said it might.
A short test
- Can you state, without looking it up, what this metric excludes?
- Has the definition changed in the period shown, and if so is the history restated?
- Can you decompose a movement along the obvious dimensions in the room?
- Would a movement of this size, in the other direction, prompt a different decision? If not, why is it in the pack?
- Do you know how uncertain it is, and does the presentation reflect that?
Metrics that pass all five are rarer than they should be. Ones that do tend to stay in the pack for years, which is the clearest signal that they are earning their place.
Targets change what a metric measures
A metric that is merely reported and a metric that someone is held to are different objects, even when the calculation is identical. Attaching a target introduces an incentive, and incentives reshape the behaviour that the metric was supposed to observe.
This is well understood in the abstract and consistently ignored in practice. A support function measured on ticket resolution time closes tickets faster, partly by resolving issues and partly by closing tickets that were not resolved. A sales function measured on pipeline generates pipeline, including pipeline that was never going to convert. Neither team is behaving dishonestly. They are responding to what was asked of them.
The practical defence is to pair any targeted metric with a counterweight that would deteriorate if the target were being met the wrong way. Resolution time paired with reopen rate. Pipeline volume paired with stage conversion. Neither pairing is perfect and both make gaming visibly harder.
It is also worth being deliberate about which metrics carry targets at all. A board pack where every measure has a target reads as rigorous and functions as a set of instructions, several of which will conflict. Targets on a small number of measures, with the rest reported for context, produces better behaviour and better discussion.
The presenter is part of the metric
A final and slightly unfashionable point. The credibility of a number in a board meeting is not independent of who presents it and how they have behaved previously.
A function that has flagged its own bad news, corrected its own errors publicly, and declined to present figures it was not confident in accumulates standing. When it presents a surprising number, the room’s first instinct is to explore it rather than to doubt it. A function that has been consistently upbeat gets the opposite treatment, and no methodological improvement recovers it quickly.
This is worth naming because it is a real asset that is built slowly and spent quickly, and because the moments that build it are usually the moments a team is most tempted to smooth something over.
