Metrics glossary
Every metric page in this section follows the same template, so you always know where to look for an answer:
- What it is — a one-sentence definition.
- How it is computed — in plain language, no formulas, but precise.
- How to interpret it — what the number tells you and what it does NOT.
- Common misreadings — real questions we have been asked.
The most important reading rule: event vs state
Many numbers in the console measure events in a period (how many times something happened), while others measure a current state (how many customers are in some state right now). The two do not have to match — and should not.
A real-life example: Insights shows 72 won-back customers for the period, but the Audience filter "recently won back" shows 7.
- 72 = customers who entered the won-back state during the period (an event count from history).
- 7 = customers who are in that state right now (a snapshot).
The state is deliberately short-lived: when a customer buys, they move on to "hot"; when their rescue click ages out without a purchase, they fall back to "lapsed" or "dormant". The other 65 have simply moved on through the lifecycle — nothing is lost.
Chapters
| Topic | What it covers |
|---|---|
| Attribution | direct purchase, later exact purchase, indirect purchases, view-assisted purchase, influenced purchase, control purchase — and why "influenced" does NOT count toward engine revenue |
| Engine revenue vs total revenue | why the engine's number is deliberately conservative |
| Lift and incrementality | descriptive lift vs holdout-based incrementality; why the confidence interval starts wide |
| Engagement states | hot → warm → cooling → lapsed → dormant + win-back; the day thresholds |
| Email metrics | open rate (and Apple privacy-protection noise!), click rate, unsubscribes |
| Cold start | why a brand-new customer initially gets bestseller-based picks |
| Replenishment | buying-cadence prediction and the automation fields |
| Segments | the eight buying-behaviour patterns and how the engine uses them |
If a number looks "wrong", always start with the question: does it measure events in a period, or a current state? Most confusion resolves right there.
Last updated: 2026-07-06