Skip to main content
Some of the most important metrics aren’t a simple average over users. Revenue per session, average order value, items per order — each is a ratio where both the top and the bottom change from unit to unit. Ratio metrics measure these correctly.

​
Why a ratio is different

Take revenue per session. If you divided total revenue by total sessions and stopped there, you’d have a number — but no honest sense of how uncertain it is, because the denominator (sessions) is itself random and varies between users. A ratio metric accounts for that. Traffical estimates the ratio and its uncertainty using the delta method, which combines the variability of the numerator, the variability of the denominator, and how the two move together. The result is a lift and a confidence interval you can trust, just like any other metric.
Use a ratio metric whenever the quantity you care about is naturally “X per Y” and Y isn’t fixed — per session, per order, per visit. If the denominator is simply the number of exposed units, an average or conversion-rate metric is the simpler choice.

​
What you configure

A ratio metric is defined by two parts:
  • Numerator — what you’re summing (for example, revenue, or items).
  • Denominator — what you’re dividing by (for example, sessions, or orders).
Each is sourced from your events the same way other metrics are. The engine sums both per unit, then forms the ratio and its standard error across units.

​
Examples

Numerator: purchase revenue. Denominator: sessions. Answers “did the variant change how much each session is worth?” — independent of whether it also changed the number of sessions.

​
Works with the rest of the engine

Ratio metrics flow through the same analysis as everything else:
  • Sequential testing — peek at a ratio metric as often as you like, with the same anytime-valid guarantee.
  • Confidence and significance — the same two-sided confidence level and interval-based decision described in Significance.
Pair a ratio metric (revenue per session) with a count metric (sessions per user) when you want to see both whether each session got more valuable and whether users came back more often. They answer different questions.