Industry
Part of Revenue against income, profit against cash: business vocabulary split into pairs
Business vocabulary metrics that predict anything
Business vocabulary for metrics: measure, ratio, rate and percentage point kept apart, plus stock against flow and what a denominator actually claims.
A metric is a sentence with five parts, and the number is only one of them. What was counted, over what period, per what, as of when, and compared with what. Drop any of the other four and the number can mean almost anything, which is how two people can agree on the figure and disagree about what happened.
This page is about the words used to build and describe metrics. It gives no targets and no benchmarks, because those belong to a particular business in a particular year; what it gives is the vocabulary to read someone else's number and know what it is.
What to take away
- Measure, metric, indicator and KPI are not synonyms. They form a ladder from "a thing counted" to "a thing decisions are made on", and a number moves up it by being chosen, not by being important.
- Percentages hide their base. A change stated as a percentage is not comparable with one stated in percentage points, and mixing them is one of the most common errors in any report.
- Stocks and flows are the deepest distinction. A figure at a point in time and a figure over a period cannot be added, compared, or divided into each other without saying so.
The ladder
A measure is a raw count or amount: units sold, hours worked, tickets closed. A metric is a measure standardized for comparison, usually by dividing by something: units per day, tickets per agent.
The metric ladder
- Measurea raw count or amount
- Metricmeasure divided by a denominator
- Indicatormetric believed to track something
- KPIindicator with a target and consequences
An indicator is a metric chosen because it is believed to move with something you care about but cannot measure directly. A key performance indicator is an indicator someone has decided to manage against, with a target and consequences.
Each step up the ladder adds an assumption. A measure is a fact, while a metric adds a denominator, which is a choice. An indicator adds a belief about what the metric tracks, and a KPI adds a target.
That target risks moving the metric without moving the thing it was meant to track, which is the proxy metric problem in general.
The five parts
| Part | The question | What goes wrong when it is missing |
|---|---|---|
| The count | What exactly is one unit of this? | Two reports count different things under one name |
| The period | Over what stretch of time? | A monthly figure and an annual one are compared |
| The denominator | Per what? | A total is read as a rate, or the reverse |
| The as-of date | When was this true? | A stale figure is read as current |
| The comparison | Against what: last period, a target, a peer? | A number with no context is read as good or bad by mood |
Stock and flow
A stock is an amount at a point in time: cash in the bank on the last day
of the month, number of customers today, headcount. A flow is an amount over a period: cash received during the month, customers gained this quarter, hires this year.
The rule is simple and constantly broken: a stock has an as-of date; a flow has a start and an end. "Customers: 400" is a stock. "New customers: 40" is a flow, and it is meaningless without the period.
A stock divided by a flow gives a duration: cash on hand divided by monthly outflow is months of runway. A flow divided by a stock gives a rate: new customers per existing customer. Either way, the calculation only makes sense if both figures are labeled as the kind they are.
Ratio, rate, percentage, percentage point
A ratio is one quantity divided by another, in any units: revenue per employee. A rate is a ratio where the denominator is time or a count of opportunities: sales per week, conversions per visitor. A percentage is a ratio expressed per hundred.
A percentage point is the difference between two percentages. A conversion rate that went from four percent to six percent rose by two percentage points, or fifty percent. Both are correct.
The first is the absolute change. The second is the relative change, and it sounds impressive. A report saying "up fifty percent" about a rate that moved from four to six is true in a way designed to be misread.
The same trap runs the other way. A fall from six to four is two points, and a third. Which of the two you hear depends on which someone chose to say.
Absolute and relative change
Absolute change is the difference: new minus old. Relative change is the difference divided by the old figure. Small bases make relative change enormous: a second customer is a hundred percent increase. Large bases make it tiny. A relative figure with no base beside it should be read as incomplete. Where the base is small, quote the absolute figure and the base, and let the reader do the arithmetic.
Average, median, distribution
An average (the mean) adds every value and divides by the count, while a median is the middle value once sorted. They agree when values are symmetrically spread and diverge when a few values are much larger than the rest.
In business data that skew is normal: deal sizes, customer lifetimes, response times and order values are nearly always skewed. The average is pulled up by the few big ones, and the median shows a typical case.
Neither is wrong. They answer different questions: the mean tells you the total divided out, useful for forecasting totals; the median tells you what to expect from one case. A report should say which it is using, and when the two differ a lot, that fact is itself the finding.
Leading and lagging
A lagging indicator reports a result after it has happened: revenue, churn, profit. A leading indicator moves before the result and is believed to predict it: pipeline, trial sign-ups, quotes issued. The lagging ones are facts; the leading ones are bets on a relationship.
A dashboard of only lagging indicators tells you what you can no longer change. A plan built only on end-of-plan tests falls into the same trap, as learning plans sets out.
A dashboard of only leading indicators tells you what someone hopes will happen. The pairing is the point, and each leading indicator should name the lagging one it is supposed to move.
Gross and net, once more
Applied to a metric, gross means before an offsetting flow is subtracted and net means after; gross customer loss counts everyone who left, while net loss subtracts those who joined.
A business can report negative net loss (it grew) while gross loss is severe. The two numbers describe different problems: the first is acquisition, the second is product. Ask which whenever a churn or retention figure appears without the word.
Target, threshold, benchmark
A target is the value you are trying to reach. A threshold is the value at which something happens: an alert fires, a bonus is paid, a covenant is breached. A benchmark is a comparison figure drawn from outside: a peer, an industry, a prior period.
All three are comparison values, and a report should say which kind each one is, because the response differs. Missing a target is a management question; crossing a threshold may be a contractual one; falling below a benchmark is a question about whether the benchmark applies to you, which it often does not.
No benchmark figures appear on this page, and that is deliberate. A benchmark without its source, date, definition of the count and population is a number from nowhere. Most that circulate are exactly that.
Where a comparison figure is needed, public guidance on market research and competitive analysis explains how to find one you can cite, not one you have heard.
A target also changes the thing it is attached to. That is the content of Goodhart's law, and it is the reason a threshold and a target should not be the same number.
Cohort and aggregate
An aggregate metric pools everyone: all customers, all orders. A cohort metric follows a group defined by when they started, and reports how that group behaves over time.
Aggregates hide change: new members dilute the old. Cohorts reveal it, because each group is watched separately. When a figure improves in aggregate while every cohort gets worse, the improvement is the mix, not the business.
Reading a metric: the checklist
Reading a metric
- Is it a stock or a flow, and is the date or period stated?
- What is one unit of the count, and who defined it?
- What is the denominator, and was it chosen to flatter?
- Is a change stated in points or in percent, and what is the base?
- Mean or median, and are the values skewed?
- Leading or lagging, and if leading, which lagging figure is it supposed to predict?
- Gross or net?
- What is it being compared with, and is that a target, a threshold, or a benchmark from somewhere?
Eight questions, none about whether the number is good. Whether it is good is the last question, unanswerable until the other eight are answered. The words for the money figures themselves are set out under business vocabulary.
Knowing which metric a given decision actually depends on is what business acumen adds. How to record which numbers would change a decision before it is made is the subject of decision skills.
Common questions
What is the difference between a metric and a KPI?
Choice. A KPI is a metric that someone has decided to manage against. Nothing about the number changes; what changes is that people will now act to move it, which is exactly why it needs a stated relationship to the outcome it stands for.
Should every metric have a target?
No. Many are for watching, not steering. Attaching a target turns a measurement into an incentive, and incentives get gamed. Give targets to the few metrics that are close to the real outcome, and watch the rest.
Is an average ever the right summary?
For totals, yes, because the mean times the count is the total. For "what does a typical one look like", rarely, in business data. Report both when they differ, and say why.
Where do I get benchmarks?
From a source that states its definition, its population and its date. If a figure arrives without those three, it is not a benchmark, whatever it is called.
