A metric dictionary: making everyone read the same number

Monday morning, the management meeting. The sales lead reports conversion at roughly a third; the marketing manager opens the same screen in the same system and says it is closer to a tenth. Both are right. One measures how many qualified opportunities were won, the other how many web form submissions ever became customers. Forty minutes go to arguing which figure is correct and none to the question on the agenda, which was whether to raise the ad budget. The decision slides to next week, where the same two numbers collide again. The seasonal campaign window closes in the meantime.

A metric dictionary is a short document that states, in one place and one form, what every number used in the company actually means. Its purpose is not to teach vocabulary but to end the argument. Below we start with the mechanics of how one screen produces two numbers, then work through what a complete definition must carry, how the time window quietly distorts a figure, which decisions the denominator argument changes, why ownership belongs to one person, where the dictionary should live, what happens to history when a definition changes, the cost of overdoing this, a two-week build, and how to tell whether it is working.

DefinitionNumeratorDenominatorTime windowScopeOwner

How does one screen produce two different numbers?

Clashing figures usually come not from bad data but from two different calculations wearing the same name. The places they diverge are few enough to count: numerator, denominator, time window, scope filter, and the entity being counted. Change one of the five and two reports will never agree, while both remain technically correct. The fault lies in the naming, not the reporting.

The entity being counted is the item most often missed. If one company opened three opportunities in a month, then how many customers did we win answers three if you count opportunities and one if you count companies. If the same person filled in two forms, the lead count doubles overnight. How that distinction changes the real yield of the funnel is worked through in lead to customer conversion rate.

The second common source is filters. Are test records, internal requests, forms filled in by your own staff and cancelled orders kept in the set? Two reports that answer that question differently will produce different numbers from an identical query. The filter is the invisible half of a definition, and almost nobody writes it down.

The third source gets discussed less and is more stubborn: units and calendars. When deals closed in different currencies land in one total, using the closing-day rate rather than a monthly average makes the same quarter two different sizes. The same holds for time zones and for where the week begins. Until those details are written into a definition, nobody notices an error, because there is no error — only two different rules.

What must a definition state, at minimum?

Defining a metric does not end with writing its formula. The formula is at most half of it; the other half is which records it runs over and across what stretch of time. The eight headings below are enough to close a definition to argument, and skipping any one of them is not optional.

• Name: The name has to be unique; conversion on its own is not a name, while qualified-opportunity-to-customer conversion is one.
• Numerator: What exactly is being counted — an opportunity, a company, a person, an order line?
• Denominator: What is it divided by, which records enter that set, and which never do?
• Time window: Are records grouped by creation date or close date, and how many days does the window span?
• Scope and filters: Are test records, internal requests, cancellations and your own employees inside the set or outside it?
• Source: Which system and which fields produce the number, and which one wins when two systems disagree?
• Owner and freshness: Who may change the definition, how often does the number refresh, and with what lag?
• Known limit: What does this number not show? Every definition needs a one-line warning underneath it.

All eight take at most six or seven lines per metric. A long definition goes unread, and an unread definition may as well not exist. When one refuses to compress into that space, the problem is rarely the writing; the metric itself has been built too elaborately.

How the time window quietly distorts a figure

Time is the component most easily forgotten and the most expensive to get wrong. A conversion rate can be built two ways: how many of the opportunities that closed this month were won, or how many of the opportunities opened this month were eventually won. The first is a period report, the second a cohort. They converge over a long horizon, but in a growing company they diverge substantially for months.

The direction of that gap is not fixed either. In a fast-growing team the period calculation understates conversion, because the denominator fills with opportunities that have not had time to mature; in a shrinking period it inflates it. Management that does not know this squeezes the sales team precisely in the months it performs best. Where duration metrics start counting is covered in sales cycle length; first contact or opportunity creation can make the same team look strong or slow.

A ratio of what, exactly?

The denominator argument looks like a detail and decides which work you will do. Two reasonable readings of one metric usually point to two different budget lines. The table collects the five that clash most often and what each split changes.

Metric	Two readings that clash	What the split changes
Conversion rate	All form submissions, or qualified opportunities	Ad budget versus sales coaching
Active customer	Transacted in ninety days, or holds a live contract	How large churn appears to be
Sales cycle	From first contact, or from opportunity creation	Which team owns the waiting time
Average deal	All wins, or new business excluding repeat orders	Whether the new-customer target is realistic
Churn	By customer count, or by revenue	Which segment is worth defending

Every row holds two defensible readings. Measure churn by customer count and losing small accounts looks alarming; measure it by revenue and one departing major account fills the picture. Reading both views together is covered in revenue churn versus logo churn, and which costs belong inside acquisition cost in customer acquisition cost.

Both definitions can be right; they cannot share one name

The common reflex is to pick one of two clashing definitions and ban the other. That is usually wrong. Marketing needs the wide denominator to see channel efficiency; sales needs the narrow one to see its own. The fix is not to merge them but to name them separately. Form-submission-to-customer conversion and qualified-opportunity-to-customer conversion are two metrics, and both belong in the dictionary. What produces the argument is not that two calculations exist, but that both were crammed under one word.

A metric with no written definition is as many different metrics as there are people calculating it.

Every metric needs one owner

The owner is not the person who uses the metric most, but the one entitled to change its definition. A definition with no named owner drifts within six months: someone adds a filter, someone widens the denominator, nobody notices, and the number slowly changes meaning. Ownership belongs to one person; a definition assigned to a committee is never updated.

Handover has to be written down as well. When the person who defined a metric leaves or changes role, the definition is orphaned and shifts at the first small edit. Teams that put metric ownership on the handover checklist are rare; they are also the teams whose numbers still mean the same thing a year later.

The owner's second job is defending the fields the definition rests on. An opportunity record with an empty source field quietly breaks channel conversion, and no report will reveal it. That is why dictionary work and field discipline move together; required fields, validation and deduplication are covered in CRM data quality.

Where should the dictionary live?

Most metric dictionaries die inside a well-made document. It gets written, shared, read in the first week, never opened again. The cause is not laziness but distance: the person looking at a number and the one reading its definition are the same person at the same moment on different screens. Two clicks away, a definition goes unread.

The fix is to move the definition next to the number. Under every tile on a dashboard there should be a one-sentence definition, a link to the detail, and a last-updated date. How to fold that into layout is covered in sales dashboard design, and which chart answers which question in CRM report literacy. A standalone dictionary document works only as the source behind those links; it cannot survive on its own.

What happens to history when a definition changes?

Definitions change, and they should: the business changes, channels change, the product changes. The danger is not the change but making it silently. Narrow a denominator without telling anyone and past months improve overnight, with nobody able to say why. That is how trust in the data ends in a single step.

The sound method has three parts. Stamp the change with a date, publish old and new series side by side for at least one period, and if you restate history, announce it and keep the old series. Handling that seam when comparing against your own past is covered in internal benchmarking. An undated change makes every later trend reading unusable.

What does overdoing this cost?

A metric dictionary turns into a project of its own with little encouragement. A sixty-entry dictionary is dead on arrival: nobody reads it, nobody maintains it, and within six months it is a source of wrong answers. The boundary is simple. If a metric does not change a decision, it does not get defined — it gets removed from the dashboard. A metric with no decision attached is noise, however carefully defined.

For a team of eight or ten, eight definitions are usually enough: the numbers that appear on screen in the monthly management meeting. The list lengthens on its own as the company grows, but the trigger for lengthening should be an argument. A new definition is added because two people read the same number two ways, not because it might be needed later.

How do you build the dictionary in two weeks?

The first week goes to inventory. Open the last three months of management decks and standing reports, list every number in them, and flag where one name covers different calculations. That list is usually shorter than expected and the collisions more numerous. For a starting view of which indicators count as core, see the sales KPIs worth tracking.

While taking inventory, settle on a naming convention. Gathering metrics from the same family under a shared prefix makes the list readable in alphabetical order and stops arguments about where a new definition belongs. The rule can be plain: the entity measured, then the transition measured, then the scope. This unremarkable step is the single thing that keeps the dictionary findable six months later.

The second week goes to writing, and the writer should not sit alone. For each definition, bring in the two people who use that number and close it in front of them. The objections raised there rehearse every objection that would otherwise arrive later. Under each closed definition add the line about what the number does not show; that single line becomes the most-read part of the dictionary.

How do you know the dictionary is working?

The test is simple and unforgiving: can a new hire reproduce the number in their first week without asking anyone? If yes, the definition is complete. If not, what is missing is not in the document but in the definition itself. The second test is meeting time; when the minutes spent arguing about whether a figure is right start falling, the dictionary is doing its job.

The third signal is a drop in one-off report requests. Teams stop asking for a number of their own the moment they trust the shared one. How regularly distributed summaries feed that trust is covered in scheduled and automated reports; everyone seeing the same number at the same time matters nearly as much as the definition itself.

A setup where definitions sit beside the numbers, filters are built into the report rather than remembered, and the same indicator reaches everyone in the same form turns the dictionary from a document into a working habit. In Rocketly, custom reports, dashboards and scheduled summaries all run on the same data model — open a free account and set up your own metric definitions.