Loading page
About this tool
Define a metric by numerator and denominator, check its 95% margin of error and sample size, then build the AI prompt that documents it.
The KPI Definition Prompt Builder takes a metric's numerator, denominator, type and reporting grain, evaluates a worked example, and — for percentage metrics — computes the 95% margin of error using the normal approximation z × √(p(1−p)/n) along with the denominator you would need to hit a target margin. It then writes an AI prompt that turns all of it into a definition with the edge cases decided rather than caveated. For analysts and PMs tired of two dashboards disagreeing about the same number.
Open KPI Definition Prompt Builder on AltFTool — it loads instantly in your browser.
Fill in the Metric name, Type, Numerator and Denominator, a Sample numerator and Sample denominator (the default worked example is 820 of 1000), plus Grain and Direction.
Review the three result tiles — Value, MOE in ±percentage points and the prompt's token estimate — which recalculate as you edit.
Read the generated analytics-engineer prompt in the preview pane and press Copy prompt; Reset restores the worked example after a confirmation.
The definition starts from what is counted and what it is divided by, not from a dashboard name.
The 95% interval and required sample size show whether a movement is real before anyone reacts to it.
Ten common edge cases can be attached, and the prompt demands a rule for each, not a footnote.
State the numerator and denominator as specific events with their filters and time window, name the grain and direction, then decide every edge case as a rule. A definition is finished when two engineers reading it separately would write the same query.
For a proportion, the 95% margin of error is 1.96 × √(p(1−p)/n), expressed in percentage points. At p = 0.258 and n = 4,800 that is about ±1.24 points, so anything smaller than a 2.5-point swing is inside the noise.
Rearranging the formula, n = z²p(1−p)/e². To measure a 25.8% rate to within ±2 percentage points at 95% confidence you need roughly 1,839 in the denominator. The normal approximation also needs at least five in each outcome group before it is trustworthy at all.
Any metric with a denominator someone can shrink, or a numerator someone can trigger without creating value. The prompt asks explicitly for three gaming routes and the guardrail metric that would catch each, because pairing a goal metric with a guardrail is what stops the optimisation going sideways.