Skip to main content

Goal Unique Count

Overview​

A Goal Unique Count metric answers the question: what share of users did this at least once?

Each user either converted or didn't. A user who triggers the goal once and a user who triggers it twenty times both count as one conversion. This makes it a natural fit for conversion rates, such as "did this change get more people to buy?"

How it's calculated​

Say you track a purchase goal and create a Goal Unique Count metric on it.

context.track("purchase", { category_id: "ZYZ123" });

Four users are exposed to the experiment:

Userpurchase events after exposureValue
Ana31
Ben11
Cara00
Dan00

In the experiment results, this group would show a Value of 2 (two users purchased) and a Mean of 2 / 4 = 50.00%, the conversion rate.

Compare that with a Goal Count on the same goal, which counts every event: Ana 3, Ben 1, Cara 0 and Dan 0, so a Value of 4 and a Mean of 1.0000 purchases per participant. The goal count tells you how much people bought. The unique count tells you how many people bought at all.

Creating the metric​

When creating a metric, choose Goal Unique Count as the Metric type, then:

  1. Select the Goal you want to measure, for example purchase.
  2. Optionally add Metric property filters, for example on category_id. A user then converts only if at least one of their events matches.

The Outliers limit method can't be changed for this metric type, since each user's value is already either 0 or 1.

Reading the results​

In the metric results table, Value is the number of participants in the variant who triggered the goal at least once, and Mean is that number divided by the number of participants: the conversion rate. When CUPED is applied in an analysis, both columns show CUPED-adjusted values, so Value is no longer an exact count of converters.

By default the Mean is shown as a percentage with two decimals, for example 50.00%. You can change this with the Format, Scale and Precision fields under Mean.

More examples​

  • Purchasers: goal purchase. The share of users who bought something.
  • Sign-up conversion: goal sign_up. The share of users who created an account.
  • Purchasers in category ZYZ123: goal purchase, with a property filter on category_id. The share of users who bought from that category.
  • Purchases per purchaser: a Ratio metric with Goal Count of purchase as the Numerator type and Goal Unique Count of purchase as the Denominator type.

Good to know​

  • A few very active users can't skew the result, because each user counts at most once. That's also why outlier limits don't apply.
  • It doesn't show how often users do something. Use a Goal Count for volume, or a Goal Activity Period Count for how regularly they come back.
  • If you add a cancellation under Goal relations, a user stops counting as converted only when every event that counted for them has been cancelled. Cancelling one of Ana's three purchases leaves her at 1.
  • If you use CUPED, the pre-exposure covariate is whether the user triggered the same goal before they were exposed. When CUPED is applied in an analysis, the adjusted values are no longer just 0 or 1, so the metric is analysed as a continuous metric instead of a binomial one.