Goal Count
Overview
A Goal Count metric answers the question: how many times did each user do this?
Every goal event counts. A user who buys three times adds 3, a user who buys once adds 1. This makes it a measure of volume, not of how many users converted.
How it's calculated
Say you track a purchase goal and create a Goal Count metric on it.
context.track("purchase", { category_id: "ZYZ123" });
Three users in the same variant are exposed to the experiment:
| User | purchase events after exposure | Value |
|---|---|---|
| Ana | 2 | 2 |
| Ben | 9 | 9 |
| Cara | none | 0 |
In the experiment results, this variant would show a Value of 11 (2 + 9 + 0 purchases in total) and a Mean of 11 / 3 = 3.6667 purchases per participant.
Compare that with a Goal Unique Count on the same goal, which only asks whether each user purchased: Ana 1, Ben 1 and Cara 0, so a Value of 2 and a Mean of 66.67%. The unique count says two out of three users bought something. The goal count says how much buying happened, and most of it came from Ben.
Creating the metric
When creating a metric, choose Goal Count as the Metric type, then:
- Select the Goal you want to count, for example
purchase. - Optionally set an Outliers limit method to cap heavy users (see below).
- Optionally add Metric property filters to count only matching events, for example purchases where
category_idisZYZ123. - Optionally use Add time filters to count only events within a time range relative to exposure.
- Optionally use Add cancellation under Goal relations so that cancelled events stop counting. The goal and the cancellation event need a shared key property, such as an order number. See Cancellation.
The goal is the only required field. Add refund and Add replacement are not available for this metric type.
Capping heavy users
A few very active users can dominate a count and make the results noisy. Outlier limits cap each user's total count, not individual events.
In the example above, with the Fixed method, a Lower limit value of 0 and an Upper limit value of 5:
| User | Purchases | Value after capping |
|---|---|---|
| Ana | 2 | 2 |
| Ben | 9 | 5 |
| Cara | none | 0 |
The results would show a Value of 7 and a Mean of 7 / 3 = 2.3333. Value and Mean show the capped numbers.
Users with no events are never raised to the lower limit. They stay at 0.
Reading the results
In the metric results table, Value is the total number of events across all participants in the variant, and Mean is that total divided by the number of participants. The impact is calculated on the Mean, as a continuous metric. If the metric uses CUPED, both Value and Mean are CUPED-adjusted. Value is still shown rounded to a whole number.
A change in the Mean can come from more users doing the action, or from the same users doing it more often. To tell those apart, look at a Goal Unique Count on the same goal next to it.
More examples
Total purchases: goalpurchase. How many orders users placed.Purchases in category ZYZ123: goalpurchasewith a property filter oncategory_id. How many orders in one category.Searches: goalsearch. How often users searched.Purchases per visit: a Ratio metric with aGoal Countofpurchaseas the numerator and aGoal Countofsession_startas the denominator.
Good to know
- Good for volume questions: did users buy, search or share more often?
- Sensitive to heavy users. Consider an outlier limit if a small group triggers the goal far more than everyone else.
- Each cancelled event takes 1 off the user's count. If Ana cancels one of her two purchases, her value drops to 1. A
Goal Unique Countwould still count her as 1 until both were cancelled.