A metric is one number per visitor, worked out from an event. It is what a test is judged on.
An event records that something happened. A metric counts those events into one number per visitor, so two versions of a page can be compared. "Order completed" is an event. "The share of visitors who completed an order" is a metric.
Build your metrics before you build a test. Step three of every test asks you to pick from the metrics you already have.
The three metric types
| Type | What it counts | Your results screen shows |
|---|---|---|
| Unique conversions | Whether a visitor did it, once | Conversion rate |
| Total conversions | How many times a visitor did it | Conversions per visitor |
| Value per visitor | A number the event carries, added up | Value per visitor |
An example. Ten visitors see your page. Four of them buy, placing seven orders worth 400 between them.
| Type | The sum | Shared between | The number |
|---|---|---|---|
| Unique conversions | 4 buyers | 10 visitors | 40% |
| Total conversions | 7 orders | 10 visitors | 0.7 |
| Value per visitor | 400 | 10 visitors | 40 |
Every type is shared between all ten visitors, including the six who did nothing.
Which to pick:
- Unique conversions for sign-ups, purchases and trials, where doing it once is the point. Right for most tests.
- Total conversions for searches, video plays or cart adds, where doing it more often is the win.
- Value per visitor for revenue, quantity or a score, where you care how much rather than whether.
How to set one up
- Open Metrics in the sidebar and click New metric.
- Pick the Event this metric counts.
- Pick the Metric type.
- For Value per visitor, pick the Event parameter holding the number.
- Set the Winning direction.
- Check the Shows in results as panel, then click Create metric.
The Shows in results as panel names what your results screen will call this metric and says what it shares the total between. Read it before you save.
The conversion window and the cap sit under Advanced settings. The row says what each is set to, so you only open it to change one.
For Value per visitor, the amount has to arrive from the start. We cannot work the amounts out later from events that never carried one. Send the number with the event before you start the test. See Events.
What each label means
| Label | What it means |
|---|---|
| Metric name | The name on every report. No two metrics in a project can share one, and capital letters do not separate them. |
| Description | Optional. One line on what the metric is for. |
| Event | The event this metric counts. One only, and it has to be active. |
| Metric type | Unique conversions, Total conversions or Value per visitor. |
| Event parameter | For Value per visitor only. The number to add up, such as amount. Only number parameters are listed. |
| Unit of measurement | Standard for a plain number. Currency counts in your project currency and formats the report as money. |
| Winning direction | Increase when more is better. Decrease for refunds, unsubscribes, cart removals and errors. |
| Filters | Narrows the metric to events carrying a certain value. Up to three. |
| Conversion window | How long after a visitor enters the test a conversion still counts. Default Test duration, or 24 hours, 3 days, 7 days, 14 days or 30 days. Ninety days is the longest allowed. |
| Cap extreme totals at | For Value per visitor only. A percentage from 90 to 99.99. One visitor's total counts no higher than that ceiling. |
| Maximum value | A fixed ceiling you type yourself. It wins over the percentage. |
Set Winning direction wrong and your report names your worst version the winner, because it reads the number the way you told it to.
Filter comparisons
| Comparison | What it matches |
|---|---|
| is | An exact match |
| is not | Everything except the value you type |
| is more than | A number greater than the one you type |
| is at least | That number or greater |
| is less than | A number smaller than the one you type |
| is at most | That number or smaller |
| contains | Part of a piece of text |
| was sent | Any event that carried the parameter |
| was not sent | Any event that did not carry it |
Only parameters the event already declares are listed, so add them to the event first.
Good to know
- You can edit a metric at any time, and it cannot break a running test. A test keeps the meaning the metric had on the day it started.
- A metric is never deleted. Archive takes it off the pickers, so no new test can be judged on it. Every test that already reads it keeps it, and their reports read the same as before. The Used by column counts those tests.
- Archiving also withdraws a guardrail. A test that starts afterwards is not handed it. A test already running keeps the one it was given on its first day.
- An archived metric is under Status on the metrics list, which opens on the active ones. Restore puts it back on offer.
- An archived metric keeps its name. Two metrics in one project cannot share a name, archived or not.
- A test reads one metric as its goal and up to five more alongside it. You pick those on step three of the test.
- Every metric reads what each visitor did in the 14 days before they entered a test, to sharpen the result. The method is called CUPED, and there is nothing to set. It takes out the part of the spread that a visitor's own past explains, so a result is clear sooner. It can change which variation is ahead. If one variation happened to get more visitors who had done this before, its number is corrected for that head start. The report says how much narrower that made the range, in a line under the goal's table. A test reads its numbers as they are until 1,000 of its visitors had some earlier activity. That night it is decided once whether it helps, and the answer holds for the rest of the test. It does nothing on a test where nearly everybody is new.
Related articles
- Events: set up the events your metrics count, and check they arrive.
- Your first A/B test: build the test your metric decides.