Auto-Optimize a Proof Match widget
Let a Proof Match widget test its rules against each other and show each visitor the proof that leads to more conversions.
Auto-Optimize finds out which of a Proof Match widget's rules converts best. It splits the visitors a rule matches at random between that rule, the other rules that match them and the widget's default selection, counts the conversions that follow, and sends more visitors to the option that does better. It's on Business. The workspace owner and the project's Admins switch it on and apply winners; anyone who can open the project can read the results.
Before you start, you need:
- a published widget in Proof Match mode with at least one rule. See Set up Proof Match rules;
- the widget on your site through its script embed, or through the iframe code with Automatic sizing and Proof Impact ticked under Get code. An iframe without that helper gets no test and applies no rule: every visitor sees the widget's base pool;
- Proof Impact tracking with a Primary outcome event that your site sends. See Set up Proof Impact tracking.
Switch it on
Open the widget in the builder and find the Auto-Optimize card in its Proof Match settings. Switch it on and publish the widget.
Under Apply winners, choose what happens when a test has a winner:
- Ask me first, the default: you're told, and nothing changes until you apply the winner.
- Automatically: the winner is shown to most visitors as soon as it's found.
You can also switch a widget on from Proof Impact under Analyze in your project's sidebar: click the Auto-Optimize tab and use the switch beside the widget.
Keep a rule out of the test
While a test runs, some visitors on a page with a rule see the default selection or another matching rule instead. If a page must always show one rule, open that rule in the builder and switch on Always show this rule. Where it's the first rule that matches, every visitor sees it, and no test runs for those visitors.
Follow the test
Click See results on the card, or open the Auto-Optimize tab of Proof Impact. Each widget lists its tests, one for each set of rules that can match the same visitor. A test's table shows, for each option:
- Share of new visitors: how the next visitors are split;
- Visitors, Conversions and Conversion rate;
- Chance it is best: how likely this option is the best one, given the numbers so far.
A new test shows no counted visitors at first. See Why visitors are not counted yet.
Above the table, the test says where it stands. See What each status means.
Apply or undo a winner
With Ask me first, a test that shows Winner found has a button that starts with Apply and names the winner. Click it. The winner is then shown to most visitors; 10% keep seeing the other options, so the test can notice if it stops being the best. If it does, the test starts again and you're told.
Click Undo on an applied winner to go back to testing. A winner you undid isn't applied again by itself.
Why visitors are not counted yet
A visitor is counted once their time to convert has passed. That time is the primary event's Attribution window, and 7 days at most. Until then the visitor appears under Visitors as not counted yet, and a line under the table says how many are waiting. So a new test shows no counted visitors for its first hours, or its first days with a longer window, although your visitors already see the widget.
What each status means
A test shows one of these above its table:
| Status | What it means |
|---|---|
| Still learning | The split is even. It stays even until every option has 100 visitors and the test has run 7 days. |
| Testing | Better options now get more of the new visitors. Every option keeps at least 10%. |
| Leading | One option has passed the winner test on some daily checks, but not yet on 7 in a row. |
| Winner found | One option won. With Ask me first, it waits for you. |
| Winner applied | The winner is shown to most visitors. |
| Waiting for conversions | No primary conversion has arrived in the last 14 days, so the test waits. Once your site reports conversions from its server, a test counts only those, and waits when none of those arrive. |
How a winner is decided
An option wins when all of this holds:
- its chance of being the best is 99% or more;
- it converts at least 10% better than the next best option;
- that has been true on 7 daily checks in a row;
- every option has at least 500 visitors, and the test has at least 100 conversions.
The bar is high on purpose. A test is checked every day, and a lower bar would name a winner between two options that are in fact the same.
A visitor is given one option at random and keeps it on later visits in the same browser. At most four options compete for a visitor: the default selection and three rules. A rule joins a test only where it would have enough testimonials to show. Visitors who haven't agreed to tracking still see the split, but aren't counted.
What starts a test again
- Editing a rule's conditions or what it shows, or the widget's default selection.
- Moving a rule above another rule of the same test, because that changes what the page shows without Auto-Optimize.
- Changing the project's primary conversion event.
Renaming a rule doesn't.
While an experiment is running on the widget, Auto-Optimize waits, and you can't apply or undo a winner until the experiment is paused or has ended.
What happens next
You get a notification when a test finds a winner, when a winner is applied, and when an applied winner stops being the best. The Auto-Optimize tab also lists each change with its date.
Was this page helpful?
Run a Proof Impact experiment with a no-widget group
Split your visitors between a published widget and no widget at all, then compare how often each group converts.
Proof Impact shows no data or unmatched conversions
Find out why Proof Impact records no widget views or conversions, or can't match them, using its setup checks and Diagnostics page.