The signal
The slide says: Active customers.
Nobody objects. The number is large enough to be encouraging and small enough to be credible. Three functions are in the room and all three nod. The meeting moves on to the next slide.
Then someone — usually newly arrived, usually not senior — asks a slightly awkward question.
"Sorry, what counts as active?"
There is a pause of the kind that only happens when a question is both simple and unanswerable.
Do trial users count? Suspended accounts? A customer with no activity for ninety days? Are subsidiaries counted once or separately? Is this month-end, or rolling thirty days?
Five minutes later there are four definitions on the whiteboard. Sales has one. Finance has another. Operations includes something the other two exclude. And the dashboard, it turns out, contains a fourth interpretation that nobody in the room wrote.
Nobody was wrong. Nobody was hiding anything. Everyone had been reasoning carefully — from a different meaning of the same word.
The pattern
Organisations get to naming alignment long before they get to meaning alignment, because naming alignment is cheap and visible.
It is straightforward to agree on a KPI list. It is straightforward to standardise a reporting template, adopt a common dashboard, or publish a one-page metric catalogue. These things look like alignment. They produce artefacts. They can be completed.
What is rarely completed is the harder work underneath the label:
- what precisely is being counted;
- which cases are included;
- which cases are deliberately excluded;
- at what grain the number is produced;
- which time logic applies;
- how the calculation actually runs;
- who is allowed to change it;
- which version this particular screen is showing.
None of that is visible on a slide. All of it changes the number.
The dashboard makes this worse rather than better. A well-designed screen presents a single figure with an air of finality. It does not display the assumptions it inherited. Presented well enough, a contested definition looks like a settled fact.
So the disagreement does not disappear. It simply stops being spoken about, and reappears later as a reconciliation exercise, a competing report, or a quiet decision by someone to keep their own version in a spreadsheet.
What is actually stuck
The instinct is to treat this as a reporting problem: better tooling, tighter dashboard design, a stricter visualisation standard.
It rarely is. The friction sits in the space between five things that are usually owned by different people:
Language — the word used in the meeting. Business rules — the conditions the organisation believes it applies. Data — the tables, joins and filters that actually run. Ownership — who has authority to decide what the word means. Decision context — the choice the number is supposed to support.
When those five agree, a metric is durable. When they drift, the metric still renders perfectly. It just no longer means one thing.
Meaning has to live somewhere and it has to belong to someone. In most organisations it lives implicitly — in a query, in a report author's habits, in the memory of an analyst who left in March. Implicit meaning survives while the people do. It does not survive reorganisation, tooling migration, or scale.
Why this matters more than reporting
If a KPI only appeared in a monthly pack, ambiguity would be an irritation. It rarely stops there.
The same concept is quietly reused: in planning assumptions, in commercial targets, in an operational alert threshold, in an application that shows a customer state to a service agent, in an automated rule that decides who gets contacted, in the context given to an AI system that summarises or recommends.
Each of those reuses inherits whichever interpretation happened to be nearest. Ambiguity does not stay contained; it propagates, and it propagates silently.
A meeting has a defence against this. Humans notice tone, hesitate, ask the awkward question, and resolve the difference in conversation. That defence does not transfer. An automated process cannot tell that two systems using the word active mean different populations. It will simply act consistently on an inconsistent premise — which is a more expensive failure than acting inconsistently, because it is harder to notice.
This is not a futuristic concern. It is the ordinary reason a well-built automation produces results the business does not recognise.
What changes it
The intervention that works is smaller than most organisations expect, and more specific.
Do not begin with a semantic programme. Begin with one consequential concept — one metric that genuinely drives money, capacity or risk — and make it explicit:
- the definition in plain language;
- the calculation;
- the inclusions;
- the exclusions;
- the time logic;
- the grain;
- the named owner;
- where that definition is governed, and how a change to it is approved.
Then do the part that is usually skipped: check whether the meaning holds outside the document. Ask Finance what they are reporting. Ask Operations what they are managing. Ask Analytics what the pipeline computes. Ask whoever owns the application what the interface shows. Ask what context the AI system has been given.
If those five answers converge, the concept is real. If they do not, you have found where the organisation is spending time reconciling rather than deciding — and you have found it deliberately rather than during a board meeting.
One concept done properly is worth more than a catalogue of forty done nominally.
What we learned
Shared language is useful. Shared meaning is operational.
A KPI becomes reliable only when its meaning survives the meeting — when the same word produces the same population in Finance, in Operations, in the pipeline, in the application and in whatever automated system has quietly started acting on it.
Until then, the agreement in the room is real. It is just agreement about a word.