The Requested Metric Fits a Different Business Model. Adapt It or Decline?
The Requested Metric Fits a Different Business Model. Adapt It or Decline?
The request rarely arrives as a demand. It arrives as a column heading in a spreadsheet built before anyone read your model — active customers, net revenue retention, subscriber churn — or as one line on a diligence checklist that the sender has used for every company in the portfolio.
If your business does not contain the thing that heading names, you are not being asked to compute. You are being asked to translate. There are three defensible ways to do that, and one that will eventually cost you more than the number is worth: forcing a flattering figure into the familiar label and hoping nobody asks what the rule was.
Ask what decision the requested number is meant to inform
Labels travel between industries; questions tend to be local. "Active customers" means currently subscribed and paying at one company, logged in during the last thirty days at another, and bought something this year at a third. All three are legitimate. None of them are the same fact.
So the first move is not to check your records. It is to find out which decision the number is feeding. Requests of this kind usually resolve to one of a few concerns:
- Repeat demand. Will the people who bought already buy again?
- Continuing revenue. How much of this period's revenue came from business you already had, rather than new logos?
- Engagement. Are the people paying actually using the product, or holding it open?
- Efficiency. What does it cost to produce a dollar of revenue?
- Concentration. How few customers carry the total?
A useful diagnostic question, when you can ask it without sounding defensive: if this comes back at half of what you expect, what will you conclude? An investor who says "I'd want to know whether usage is broadening or narrowing" has told you what they need. An investor who says "I'd want to see it be bigger" has told you something else.
Avoid the move that feels like an answer and is actually a stall: explaining that your business is unique. Uniqueness is a claim about your company. It does not speak to demand, retention or efficiency, and it moves the conversation from measurement into positioning, where you will lose ground. The stronger version of the same instinct is to restate the question in plain business language and then say what evidence exists that addresses it.
"How much of last quarter's revenue came from customers who were also customers the quarter before?" is a question you can either answer or fail to answer. "What's your net revenue retention?" may presuppose a commitment structure you do not have.
Inspect the metric's parts against the business
Every one of these measures decomposes into four parts. Write them out before you look at a dashboard.
The entity. What is being counted — an account, a legal entity, a seat, a device, a person who signed a purchase order. Whether the same economic relationship can appear twice under two spellings.
The qualifying event. What must happen for an entity to be counted. A completed payment? A login? A status flag called active? Or simply the absence of a cancellation, which is a very different kind of test.
The population. Counted relative to what — all accounts ever created, all accounts that were eligible during the period, all accounts in the sales region.
The observation horizon. Over what stretch of time, and measured as of when. This is the part where most adaptations quietly fail, because a window shorter than the natural cycle of the business measures the cycle and reports it as behaviour.
Then check each part against records that exist. Can you produce the datum at the entity level, under a stable rule, for the whole period you intend to present? Plenty of companies can report a monthly figure for last month and not for the eighteen months before someone changed the tracking.
The part that does the most damage is the qualifying event, because it is where one business fact gets silently swapped for another. A one-time sale is not a subscription start. A renewal is not a repeat purchase. A payment is not engagement. A cancellation is not the end of a customer relationship. Each of those substitutions is where a number stops meaning what its label says.
Consider two fictional companies, used here only to lay out the method. The first bills by usage: customers pay for what they consume, month to month, with no committed term, and an account can consume nothing in March and plenty in April without any contract event in between. It is asked for "active customers." The second sells equipment one unit at a time. It is asked for "subscriber churn," and it has no subscriptions and no cancellation event — not a rare event, no event.
Neither case is drawn from a real company's records, and no figures appear here for a reason worth stating plainly: the method is what transfers, and a fabricated number would only make it harder to see.
Choose a limited adaptation only when its meaning survives
An adaptation is limited when it changes the name or the boundary and leaves the underlying event alone. It is a rescue operation when it manufactures an event that does not occur.
For the metered service, the event that exists is a completed paid transaction. That is real, it is recorded, and it can be counted per account for a stated calendar month. So the qualified answer looks like this: count accounts with a completed paid transaction in the stated month; call them transacting accounts; state on the same line that this is not subscription status and does not indicate whether an account will transact next month.
What survives: the entity and the population are unchanged, the event is one that actually happens, and the horizon is written down instead of assumed. A reader can compare this month to last month under the same rule.
What is lost: that number cannot sit on the same line as a peer's active-subscriber count. It answers "how many accounts paid us in March" and "is that figure moving." It does not answer "how many customers do we have," because an account that transacts in March and not in April is neither retained nor lost — it is an account whose usage changed. Say that in the document, next to the number, not in the conversation that gets forgotten.
Four rules keep a limited adaptation honest:
- Write the definition before computing it. Freeze the wording. Compute afterward, not the other way around.
- Apply one rule to the entire presented period. If the definition changes, the series breaks; restate the whole series under a new name rather than splicing two rules into one line.
- Use owner-verified records. Someone accountable for the data confirms the rule as written, and that person is named in the document.
- Do not select the window after seeing the result. Choosing the twelve months that end on the strongest month is not a measurement choice. It is a conclusion with a chart attached.
One more caution, because it is the most common way this goes wrong: the name has to carry the boundary. If the row says "active customers" and the qualifier lives in a footnote, the qualifier will be gone by the time the number reaches a second spreadsheet. "Transacting accounts (paid, calendar month)" is a clunky heading that survives copying. "Active customers (see note)" does not.
Offer an alternative when the original logic does not transfer
Now the equipment seller, which is the harder case and the more instructive one.
A churn rate implies that customers were in a continuing state and acted to leave it. Where no such state exists, the computation fails in two directions at once. Computed literally, cancellations over customers returns zero, which reads as flawless retention for a business in which no customer carries any ongoing obligation at all — a maximal error wearing the costume of a good result. Computed by proxy — treating every customer who did not repurchase inside some window as "churned" — it fabricates the cancellation event it needs, and converts an ordinary gap between purchases into apparent attrition.
Decline the label. Do not decline the question.
The question underneath is usually continuing demand, and it can be restated as something answerable: of the customers who bought in a given quarter, how many bought again, and how long did the second purchase take? That is a different measure from churn, and the difference is not cosmetic.
A renewal is the absence of an action, supplied by a customer who must do something to stop. A repeat purchase requires an action to happen. In the first case, inertia works for you; in the second, it works against you. Import a churn rate into a repeat-purchase business and you overstate retention in every period, because you have borrowed an assumption in which staying is the default. It isn't.
The replacement cycle then decides whether your alternative means anything. For the equipment seller, the interval between purchases is governed by how long the equipment lasts and when the buyer's need returns. A cohort window shorter than that interval will report "no repurchase" for customers who are simply not due yet. So the horizon has to exceed the cycle — and if the cycle runs five years and the company has existed for four, the honest answer is that the records cannot answer the question yet. That is a different statement from "the question is invalid," and the two should never be merged. The question is fine. The records are short.
Where repeat-purchase evidence does exist, compare what it reveals and what it hides before recommending it. A cohort repurchase rate reveals the depth of demand and roughly how long the cycle runs. It conceals the health of new-customer acquisition, since a company can post a strong repeat rate while its customer base shrinks. It is highly sensitive to the window — "bought at least twice in five years" will look impressive and mean very little unless the window is stated and compared against like windows. And it necessarily excludes recent first-time buyers who may well return, so it should be read alongside the calendar rather than as a verdict.
None of this makes the alternative a superior metric in general. It is the more truthful one here, which is the only comparison that matters.
Make the explanation travel with the number
The definition, the source period, the known limits and any restricted comparability belong beside the result — same cell, same slide, same exported column header. The number will be copied forward; the footnote will not.
Have the person accountable for measurement or finance confirm the rule and the calculation before real figures enter a document that leaves the building. Not because the arithmetic is difficult, but because the person who wrote the rule should be the person whose name is attached to it. Who that is varies: a controller, a data lead, a founder with the ledger open. What matters is that someone with access to the records has confirmed the rule as written, and is identified.
Then be disciplined about what the number is not. It is not endorsed by the investor, it does not bear on whether the company is a suitable investment, and it does not satisfy a financial reporting standard. A management-defined measure placed next to a GAAP figure does not become GAAP by proximity. If the request arrives in a context with contractual or regulatory weight — covenants, fund disclosures, audited statements — that is a different conversation with different rules and a different owner, and this article's method does not substitute for it.
One bounded observation is worth carrying into this section. Stripe's subscription analytics documentation, last reviewed on 8 September 2026 and not re-checked since, exposes configuration choices in how subscription metrics are defined — including when a subscriber becomes active. Its usefulness is narrow, and worth stating precisely: it shows that even inside a single vendor's own reporting, a published number rests on an explicit measurement rule that someone selected. It does not tell you what any investor wants, it is not a financial standard, and it validates neither of the fictional cases above. What it does support is a habit. Decide the rule, then report.
Finally, version the definition alongside the number. When the rule changes, the measure gets a new name and the series restarts. And keep the lost comparability visible rather than hoping a careful reader will infer it: every presentation of a qualified measure should say out loud what it is not.
End with something you can defend
There are three answers that hold up and one that does not.
The first is a qualified measure: here is the number, under this rule, over this period, and here is what it cannot tell you. The second is an alternative: the requested label does not exist in this business, but this evidence addresses the question behind it, and this is what the evidence in turn leaves out. The third is an evidence gap: the question is the right one, the records cannot answer it yet, and here is what would have to exist first.
The one that fails is the flattering figure in the familiar label. It works exactly once, in the room where nobody asks how it was computed — and the money arrives with people who do ask.
The test is short enough to use under pressure. Can you say in one sentence what a customer had to do to be counted, and did they actually do it? If yes, report the result under a name that says so, with its limits attached. If no, say what you have instead.
The label will survive the meeting either way. Your numbers are the part that has to.
Frequently asked questions
What is the first move when a requested metric does not match the business model?
Find out what decision the number is meant to inform. Requests usually resolve to concerns such as repeat demand, continuing revenue, engagement, efficiency, or concentration. A useful diagnostic question is: if this comes back at half of what you expect, what will you conclude? Avoid explaining that your business is unique, because uniqueness is a claim about the company and does not address demand, retention, or efficiency. Restate the question in plain business language and say what evidence exists.
What parts of a metric should be inspected against the business?
The article breaks every such measure into four parts: the entity being counted, the qualifying event, the population counted relative to what, and the observation horizon. Check each part against records that exist, including whether the datum can be produced at the entity level under a stable rule for the whole period. The qualifying event is often most damaging because one business fact gets silently swapped for another: a one-time sale is not a subscription start, a renewal is not a repeat purchase, a payment is not engagement, and a cancellation is not the end of a customer relationship.
When is a limited adaptation honest?
A limited adaptation changes the name or boundary but leaves the underlying event alone. For a metered service, for example, count accounts with a completed paid transaction in a stated month and call them transacting accounts, while stating on the same line that this is not subscription status and does not indicate whether an account will transact next month. Four rules keep it honest: write the definition before computing it; apply one rule to the entire presented period; use owner-verified records and name the accountable person; and do not select the window after seeing the result. The name itself should carry the boundary.
What if the original metric's logic does not transfer, such as asking an equipment seller for subscriber churn?
Decline the label, not the question. A churn rate implies customers were in a continuing state and acted to leave it. Computed literally where no such state exists, cancellations over customers returns zero, which reads as flawless retention but is misleading. Computed by proxy, treating every customer who did not repurchase in a window as churned fabricates the cancellation event and converts an ordinary purchase gap into apparent attrition. The article recommends restating the underlying question as repeat purchase: of customers who bought in a given quarter, how many bought again, and how long did the second purchase take? The horizon must exceed the replacement cycle, and if records are too short, the honest statement is that the records cannot answer the question yet, not that the question is invalid.
What are the three defensible answers, and what is the failing answer?
The three defensible answers are a qualified measure, an alternative measure, and an evidence gap. The failing answer is a flattering figure placed in the familiar label. The test is short: can you say in one sentence what a customer had to do to be counted, and did they actually do it? If yes, report the result under a name that says so, with its limits attached. If no, say what you have instead. The explanation should travel with the number, and an accountable person should confirm the rule before real figures enter a document that leaves the building. A management-defined measure placed next to a GAAP figure does not become GAAP by proximity, and contractual or regulatory contexts have different rules and owners.