KPI Definition Generator

Two teams can both put "active users" on a dashboard and mean two different numbers. This page writes a metric down properly — a formula, a data source, a plain-language meaning and a list of the ways the number gets inflated — in three versions that differ in how tightly they are bounded.

A Definition Is a Decision, Not a Label

The value of a KPI definition is not the words. It is the argument you no longer have to have every quarter. Writing down that an active user is someone who completed a transaction, rather than someone who logged in, settles a question that otherwise resurfaces whenever the number is inconvenient.

• Enter a metric name and the tool recognises fourteen that cover most recurring business reporting: Monthly and Daily Active Users, conversion rate, retention, net revenue retention, gross margin, churn and gross churn, ARPU, average order value, CSAT and NPS. Common spellings and abbreviations are mapped to the same entry, so MAU, monthly active users and the spelled-out form all resolve to one definition.
• Each recognised metric arrives with a real formula rather than a placeholder, expressed as a query-shaped string such as a distinct count over a date window, plus a stated entity type so the arithmetic says users or customers rather than the vaguer word records.
• Every definition carries a watch-out field, and this is the part worth reading. The MAU entry notes that the active action has to be defined and that counting raw logins inflates the figure; the DAU entry suggests pairing it with MAU as a stickiness ratio and warns that the per-user definition must match across dashboards.
• Three versions are produced side by side. Strict is the narrow, defensible reading; Standard is presented as the industry default and annotates the formula with what is being counted per record; Broad is the inclusive version and widens the data source to include auxiliary joins.
• A metric the table does not recognise still gets three versions, built from a skeleton with the placeholders left visible — a deliberate choice, since a definition that shows you where to write your own column names is more useful than one that invents them.
• Domain and Goal emphasis selectors adjust the framing: choosing retention adds a note about keeping cohorts to thirty days or less and excluding reactivation noise, revenue adds a note about currency and excluding refunds and taxes, and quality adds a note about filtering to verified records.

From a Metric Name to Something a Data Team Accepts

1. Type the metric name. If it is one of the recognised fourteen, an abbreviation is enough — the alias table handles case, spacing and common variants.
2. Choose the Product or domain. Its effect is narrow and worth knowing: it adds a convention note only where there genuinely is one, such as flagging a SaaS default or the e-commerce convention for average order value and churn, rather than decorating every definition uniformly.
3. Choose a Goal emphasis. This is what most changes the watch-out text, because the failure modes differ — retention definitions break on cohort windows, revenue definitions break on refunds and tax treatment.
4. Press Generate Definitions and read Strict first. It is the version that is easiest to defend in a disagreement, and it is the one to put in a written document.
5. Use Standard as the working definition for dashboards and Broad only when you deliberately want the inclusive reading, such as a top-of-funnel health check where missing a segment is worse than double-counting it.
6. If you are defining something the table does not know, take the skeleton and fill in the bracketed placeholders. Do not delete the watch-out line it gives you — the instruction to replace placeholders with the exact columns and window your team agrees on is the actual work.

Strict, Standard and Broad Are Not Quality Grades

Alternative	Notes
Strict — narrow, defensible	The formula is given unmodified, with no qualifying annotation. Use it when the number may be challenged: an investor metric, a contractual service level, anything that appears in a document someone else signs. It under-counts by design, and under-counting is the safe direction when the number is a commitment.
Standard — industry default	Same core formula with an explicit annotation of what is being counted per record, which turns an implicit assumption into visible text. The right default for an internal dashboard, because the annotation is what prevents a reader from assuming a different denominator.
Broad — inclusive	Widens the data source to include auxiliary joins, so records that would be dropped by a strict join still count. Useful for engagement and reach questions where a missed segment distorts the answer more than a double count. Poorly suited to anything with a target attached, since an inclusive definition makes a target easier to hit without anyone deciding to make it easier.
Deciding between them	Whichever one you pick, write down that you picked it and why. A definition that changes version between quarters without a note is worse than any of the three, because it makes the trend line unreadable and the change invisible.

The Same Metric Written Three Ways

Take Monthly Active Users on a subscription product and watch the three versions separate.

• Strict gives the distinct count of users with a qualifying action inside the month window, with nothing added. If the number is challenged, this is the one with a defensible boundary: nothing outside the window counts and nothing is inferred.
• Standard repeats the formula with a trailing annotation naming the per-record unit, so the definition itself states that the count is per active user rather than per session or per event — the assumption most often left implicit in a dashboard caption.
• Broad keeps the same window but extends the data source with auxiliary joins, which is where a second system's activity can enter the count. That is the version that will disagree with a finance report, and the disagreement will be legitimate rather than an error.
• The watch-out line is identical across all three and stays useful in all three: the qualifying action has to be named, and counting logins inflates the figure. That sentence is the one to copy into the document, because it is the part that survives a change of tooling.
• Ask for churn instead and the tool recognises it, but note that churn and gross churn are separate entries with separate formulas. Reaching for the wrong one is a common and quiet error, since both produce a plausible percentage.

How the Definitions Are Produced

The named definitions are stored in the page as a compiled table, and the three versions are derived from each entry by adding or removing qualifiers. There is no model call in the default path, which is why the output is instant and identical every time you ask — two people defining the same metric on different days get the same text, and that predictability is the point. The trade-off is coverage: fourteen metrics are handled with real formulas and real watch-outs, and anything outside that set falls back to a skeleton that is structurally correct but has no domain knowledge in it.

• The goal emphasis notes are fixed strings, not generated prose, so the advice about cohort windows and refund treatment is the same advice every time rather than a fresh guess.
• Known metrics get a mapped formal name and abbreviation, so the definition is titled consistently even if you typed a lowercase shorthand.
• Unrecognised metrics keep their uppercase four-character abbreviation as a placeholder, which is a prompt to rename rather than a finished label.
• No key, no account and no connectivity are required. If you do connect a provider, the metric name, domain and goal are what get passed along, so keep the input to the metric itself rather than pasting internal figures into it.

Questions About Locking a Definition Down

Which version should go into a contract or a funding document?

Strict, without the annotation and without the auxiliary joins, because a commitment should be defined at its narrowest defensible reading. The failure mode with the other two is not dishonesty, it is ambiguity at the moment of measurement: a broader definition gives whoever reports the number room to choose, and that room is exactly what a written definition is meant to remove.

Why does an unknown metric produce placeholders instead of a guess?

Because the columns a metric counts are specific to your warehouse and cannot be inferred from its name. The skeleton gives you the structure — a count over a stated time window from a named table, with a data source and a watch-out field — and leaves the parts only you can fill. A generator that invented a column name would produce something that looks finished and is wrong, which is harder to catch than a visible placeholder.

Do the domain and goal selectors really change the output?

They change the notes rather than the formula. The goal emphasis is the more useful of the two, because it attaches the failure mode that matches your intent: cohort windows for retention, refund and tax treatment for revenue, verified records for quality. The domain selector is deliberately restrained and only adds a convention note where one genuinely exists, which is why switching it often appears to do nothing.

Can I define a metric that combines two of the recognised ones?

Not as a single entry. The table resolves a name to one definition, so a ratio of two metrics needs to be built from both of them — generate each side, keep the quarter-to-quarter note about consistent per-user definitions, and combine them yourself. Asking for the unrecognised name directly gives you the skeleton, which is the honest starting point for a derived metric.

Last verified: October 3, 2026

Launch KPI Definition Generator

All Tools · AI Tools · About · How We Test · Contact · Privacy

Maintained by Huang Bin · Last updated October 3, 2026