Article
KPI vs metric: the one-question test
Every KPI is a metric. Almost no metric is a KPI. The line between them is not about the number itself, which is why arguing about definitions never settles it. Ask one question instead: if this figure went the wrong way for three weeks, whose diary would change? A name and an action makes it a KPI. Silence makes it a metric.
By the Key Performance Indicators editorial teamPublished Updated 8 min read
01
Is every KPI a metric?
Yes, and the relationship only runs one way. A metric is any measurement your systems can produce, and most companies have more of them sitting in reports nobody opens than anyone in the room could name. A KPI is a metric somebody has picked out of that pile on purpose and attached three things to: a name, a goal and a day of the week.
Nothing about the underlying number changes when it gets promoted. Support tickets closed is the same figure whether it lives in a monthly export or at the top of your weekly board. What changes is that one of those versions has a person who has to explain a bad run of it.
This is why buying a reporting tool almost never produces KPIs. The tool makes metrics cheaper, and cheap metrics multiply. Choosing which handful to be accountable for is a decision a leadership team has to make in a room, and no dashboard makes it for you.
02
What is the one-question test?
If this number went the wrong way for three weeks running, what would change? A real answer names a person and something they would do differently. If the honest answer is that somebody would mention it and the meeting would move on, you are looking at a metric. Interesting, possibly true, and not worth a weekly slot.
Three weeks is deliberate. One bad week is noise in almost any small business, and demanding a reaction to it teaches your team to explain rather than to act. A run of three is a pattern, and a pattern that produces no action is the clearest evidence a row does not belong on the board.
The test cuts both ways, which is what makes it useful. It kills rows nobody would act on, and it rescues unglamorous ones people forget to include. Invoices sent on time looks like an administrative detail until you notice that three slow weeks of it puts a real hole in next month's cash.
03
How do you spot a vanity metric?
Look for a number that only goes up. Total customers ever, cumulative downloads, followers, hours logged since launch: these read well in a board pack and change no decision, because a figure that cannot fall carries no information about this week. The giveaway is that nobody can describe what a bad reading would look like.
The fix is usually a rate or a ratio rather than a total. Customers gained this month against customers lost is a number that can go the wrong way, which is exactly why it is worth reading. Cumulative signups since founding is a decoration you can safely put on the website instead.
Vanity metrics are not lies. That is what makes them durable. They are true numbers selected because they flatter, and they survive in weekly meetings for years because no one wants to be the person who removes the one row that is always green.
04
What is the difference between an input and an output?
An input is work your team does. An output is what that work produced. Discovery calls held is an input, closed revenue is an output, and the gap between them is however long your sales cycle runs. A weekly list needs both, weighted toward inputs, because only one of the two is still changeable on Monday.
Teams that measure only outputs get a board that reports history accurately and offers nothing to do about it. Teams that measure only inputs get a busy, cheerful board and a bad quarter nobody saw coming, usually because the activity was real and the targeting was wrong.
The useful pairing is one input and one output per outcome you care about, sitting next to each other. When the input is at target and the output is not, you have learned something specific: the work is happening and it is not working, which is a far better problem to bring to a meeting than a red number on its own.
05
How many metrics should a dashboard carry?
As many as are useful to look at when you go looking. The constraint is not on the dashboard, it is on the weekly read: five to fifteen rows at the leadership level, one owner each. Keep the rest available for the week somebody needs to dig, and stop treating a wide report as if it were a management tool.
Separating the two surfaces settles most of the arguments. Nobody has to fight to keep their favorite number when it stays a click away in the detail view. What they cannot do is claim a slot in the weekly read for it, and that is the only scarce thing in the system.
In EOS® the weekly surface has a name, the Scorecard, and EOS Worldwide's own material still puts it at five to fifteen rows, checked in August 2026. You do not need the vocabulary to get the benefit. You need one short list somebody is accountable for and one wide report they can open when the short list raises a question.
06
How do you turn a metric into a KPI?
Write down three things beside it and stop. One owner, chosen for what they can change. One goal, a number rather than a direction, set at what a good week looks like as the team runs today. One cadence, the day it gets read out. A number carrying all three is a KPI. A number missing any of them is not.
Write the definition down too, including what does not count. Half the red weeks in a young board turn out to be arguments about the formula, and each one costs you the meeting time you were trying to save. Proposals sent should say whether a resend counts, and it should say it once, in writing.
Then leave it alone for six to eight weeks. The first fortnight tells you only whether the number can be collected at all, which is a different question from whether it is worth collecting. After two months you can argue about the target from a baseline instead of from whatever somebody guessed in the meeting where the row was added.
Start with a real question
Half your board probably is not earning its slot
Send us the list you read every week. We will mark which rows pass the three-week test, which are totals dressed up as measures, and which outcome you are watching with no early number underneath it. One reply from a person, inside a business day.
A person reads every one.