For AI agents: the complete documentation index is available at https://docs.clickmax.io/en/llms.txt, the full documentation bundle is available at https://docs.clickmax.io/en/llms-full.txt, and this page is available as Markdown at https://docs.clickmax.io/en/features/create/funnels/ab-test.md.
  • English

      • Create a funnel A/B test

      Turn a funnel page into an A/B Test: traffic is split between two or more variations and you compare the results. You do this in the funnel builder, on the card of the page you want to test.

      Role: anyone with access to the funnel in the workspace can edit it — no specific role is required.

      Steps

      1. Open the funnel in the builder.
      2. On the card of the page you want to test, open the menu and turn it into an A/B Test. The card becomes a test with Variation A and Variation B.
      3. Adjust the variations. Each one points at a funnel step — a page or a Quiz — and you add a traffic splitter to include more. A test takes a minimum of 2 and a maximum of 5 variations.
      4. Set each variation's traffic split. The percentages must add up to exactly 100% — the builder warns you when the total is off.
      5. Open Test Settings to review the Slug (URL) and, if you want, a custom domain with Use another domain (the default is the funnel's domain). The test starts out with a slug of its own, shaped like teste-ab-k3m9x2 — the trailing part is drawn at random, so two tests never compete for the same URL. Change it only when you want an address that is easier to share. As you type, the field checks availability and shows This slug is available or This slug is already used by another page or test. While the slug is taken — or empty — the Save Settings button stays disabled.
      6. Under Test objective, pick the conversion the test measures on each variation. In a new test, the default is Moved on to the next step; switch to Sign-up on the page, meetings, sales or another option if the funnel has a different goal. See test objective.
      7. In the same section, if you want the test to pick the winner on its own, turn on Pick the winner automatically. It starts off: without it, traffic stays split between the variations. When on, pick By {unit}, for example By sign-ups, and enter the target in Decide when it reaches, summed across variations (at least 10); or By date, and pick the day on the calendar and the Time. See pick the winner automatically.
      8. Click Save Settings.

      The test starts splitting traffic between the variations. Variations are ordinary funnel steps — edit pages in Pages and quizzes in Quiz.

      When a variation is a Quiz

      A variation can point at a Quiz card. Whoever lands on that variation goes to the quiz's public address, and the answers are still counted on the quiz's own card.

      The quiz needs a public address to enter the split. It starts out on the system's default address; if you set a custom domain under Change domain, the test sends visitors there instead. A variation whose quiz has no address yet stays out of the split, and its traffic goes to the other variations.

      On a quiz variation's line, visitors usually sits at 0: the quiz is served outside page tracking, so there are no page visitors to count there. The traffic the test sent to it shows in the arrow's Redirects number, and the answers live under Accesses and Completed on the quiz card. For the same reason, people sent to the quiz do not count toward the views limit of older tests: only visitors of page variations count.

      The test's numbers follow the panel's date range

      In analysis mode, everything the test shows counts only the selected range in the panel's date picker: the variation lines on the card and the Redirects number on each variation's arrow. Change the range and both recalculate.

      • visitors — distinct people who opened that variation's page in the range. Someone who reloads counts once.
      • conversion — the variation's rate in the range: contacts captured when the variations lead to ordinary pages, or purchases when they all lead to checkout, upsell or downsell. The divisor is the line's visitors.
      • Redirects, on the arrow — how many times the test sent traffic to that variation in the range.

      The arrow and the line count different things on purpose, so they need not match: the arrow counts openings, the line counts people. Someone who opens the variation twice adds two on the arrow and one to visitors — the same rule every funnel connector follows.

      A variation leading to a Quiz is where the gap shows most: the quiz is served outside page tracking, so its line usually sits at 0 visitors even while the arrow counts traffic. The quiz's own numbers live on the quiz card.

      A range with no traffic shows 0 on the line and — on the arrow: that is the filter, not a stalled test. Widen the range to see the running total.

      Once the test closes, the card freezes the total and the winning variation from that closing — those two stop changing with the range.

      Test objective

      In Test Settings, the Test objective is the conversion the test measures on each variation, like a campaign's objective. In a new test, it starts on Moved on to the next step. A test created before this option opens on the conversion it already measured, Sign-up on the page or Sale on the page depending on the page types, so saving another setting, such as the slug, does not change what it measures. The options are listed in funnel order:

      • Moved on to the next step — people who landed on the variation and went to the step connected to it in the funnel, whatever the page type. The variation needs a visitor step connected right after it; links only to an automation or a note do not count. If it is missing, the field warns No next step connected, naming the variations, because they never convert with this objective.
      • Sign-up on the page — contacts captured on the variation's page (the same number as always).
      • Booked a meeting — people who landed on the variation and, after the visit, had a meeting booked in the CRM, whether through the schedule's public link, by someone on the team, by the AI SDR or by an automation. Cancelled and no-show meetings still count.
      • Meeting held — the same, but only meetings marked as held.
      • Sale on the page — sales made on the variation's page, including external checkout sales (the same number as always).
      • Bought later in the funnel — people who landed on the variation and bought afterwards, at any checkout further on, for example capture → sales → checkout.

      With Pick the winner automatically on, Statistical settings, collapsed by default, sets the Confidence level (80% to 99%, default 95%) and, with a date limit, the Minimum conversions (total across variations, default 30, at least 10). With a conversions target the minimum is hidden: the target itself plays that role. At the end of the section, a summary says how the test will end with the current choices. When the test closes at its limit, the variation with the highest rate wins; it wins with confidence when the gap reaches the level and the total conversions reach the minimum. Otherwise it wins anyway and the card shows No statistical confidence: won on the highest rate. Either way, all traffic then goes to the winner.

      The variation lines on the card count the whole life of the test, the same way the closing does, and do not change with the panel's date range. Below them the card shows which metric decides the test, the confidence reached against the required level and what is left: the Test In Progress bar with the progress to the target, like 18 / 100 sales, or, without a target, how many conversions are still missing to reach the minimum.

      With Bought later in the funnel, if the checkout after the variations is external and no sale reaches Clickmax, the card shows External checkout without sales data.

      Meetings as the criterion

      With Booked a meeting or Meeting held, the count is per visitor: someone with three meetings counts as one conversion. The meeting has to belong to the contact the visitor became and be created after their first visit to the variation. If the same contact went through both variations, for example on their phone and on their computer, the meeting counts only for the variation they saw first.

      With Meeting held and a By date limit, when the date arrives the result does not come out right away: Clickmax waits up to 14 days for meetings already booked to take place, and the card shows Waiting for N booked meetings with the expected decision date. The round ends as soon as those meetings are marked as held, cancelled or no-show, or when the deadline passes. Only visits and meetings created up to the limit count; anything that arrives during the wait is left out. Traffic keeps being split between the variations in the meantime.

      A target of meetings held has no wait: it only counts meetings that already took place, so the test decides as soon as it gets there.

      Ending before the limit, described below, also applies to Meeting held, even with meetings still booked: it only happens when the winner is already clear from the meetings that already took place.

      Pick the winner automatically

      Pick the winner automatically sits in the Test objective section itself, below the field. Off, which is the default, the test only splits traffic between the variations: it never picks a winner on its own, not even through the early choice. On, the test picks the winner by the conversions target or by the date. Once the winner is selected, all traffic goes to that variation. When and under which conditions that happens is covered in the sections below. Turning the option off on a test that was waiting for meetings cancels that wait.

      Decide on reaching a target

      With a conversions target, the test decides when the total across variations reaches the number you chose, in the unit of the win: 100 sales, 30 meetings held, 200 sign-ups. That is the yardstick the statistics use; a number of views does not say how many people converted.

      Clickmax checks the target once a day, at 9am Brasília time; the queue can delay it a little. When the target is reached, the card shows Target reached with a Decide now button that requests the decision without waiting for the next check. The card stays on Decision requested… until the round closes. If the decision does not come through within a few minutes, for example because someone edited the target in the meantime, the card shows The decision has not been confirmed yet. Refresh the test or try again. Below the target the button does not show, and only the early ending described below can close the test before the target.

      In Round history, a round closed by the target shows Ended on reaching 100 sales, with the target in force when it closed.

      Tests with a views limit, created before this option, keep closing on views: the test closes when the different people who opened the variation pages add up to the limit, the winner is declared and all traffic goes to it. Someone who opens the same variation several times counts once. When you open the Test Settings of one of them, a notice says the option no longer exists: saving makes the test follow what is selected in the modal.

      End before the limit and get notified

      With Pick the winner automatically on, Clickmax checks the test once a day. It ends the round before the target or the date when three conditions hold together:

      • the round has been open for at least one day;
      • the total conversions reached the Minimum conversions;
      • the winner reached 99% confidence.

      The 99% bar only applies to ending early. Checking the result every day and stopping the first time it looks good makes the test declare lucky winners more often; the higher bar makes up for that. At the target or the date, the Confidence level you chose applies, as before.

      When the round ends, early or at the limit, the workspace owner and the administrators get the A/B test finished alert in the Clickmax bell and as a push on the Clickmax mobile app, with the winning variation, the funnel and the confidence reached. Clicking opens the funnel, where the test card shows the suggested challenger. In Round history, a round that ended early shows Ended before the limit: confidence was reached. To turn the alert on or off, see notifications.

      A test that still decides by the previous rule, with no confidence on the card, only ends at the limit and sends no alert. It starts measuring by the Test objective when you save its Test Settings.

      Start a new round from the winner

      When the test closes, the card shows Next round with the suggestion to turn off the losing version and create a challenger from the winner. Nothing changes until you act.

      1. On the test card, click Create challenger. AI analyzes the winning page and the round result, picks one test hypothesis and changes one text at the top of the page (headline, promise or subtitle) to test it, keeping the page's tone. One change per round is what tells you what caused the gain. This takes a few seconds: you can close the window and keep working on the funnel. Meanwhile the card shows Generating challenger...; when it is ready, the notice The challenger is ready to review appears and the card shows Review challenger.
      2. In Review the challenger, compare the Winner and Challenger pages side by side, as visitors will see them. The preview opens in Desktop; switch to Mobile to see the mobile version. The texts that changed are outlined, and both pages scroll together. Use the zoom buttons to enlarge the page and Fit to see the full width again. Above the list of changes, Hypothesis tested shows the chosen hypothesis, its category, the trigger, the metric it should move and why AI picked it. In What changed, the change shows why it was made.
      3. Choose one action:
        • Accept and create: goes on to create the challenger with the new text.
        • Try another: generates a new version. Optionally, first write what you want different in What should change next time?. Only one version is generated at a time: the button shows Generating... until it arrives.
        • Discard: closes the review and nothing changes in the test.
      4. Choose the slug for the copy of the winning page and confirm.
      5. Publish the funnel. The new round splits traffic between the winner and the challenger, half each.

      Hypotheses come from a catalog, in this order: benefit-led headline, pain-led, social proof with real buyers, headline that answers an objection, unique mechanism, measurable result, with a time frame, as a question, before and after, AI-led, authority of who is behind it, and research that backs the promise. AI starts with the first one not yet tested and only skips one when the page already does what it proposes. Measurable result and time frame only apply when the page already states that number.

      Each trigger is tested alone, in a round of its own:

      • Social proof: AI adds to the subtitle or the promise how many people already bought the product, using the real number of your approved sales, rounded down and always below the total ("more than 1,200 students"). It counts distinct buyers from any offer of the product linked to the page. It only applies when the page has an offer and the product has at least 100 buyers; below that, the hypothesis is left out.
      • Authority and research: AI only brings to the top who is behind the offer, or the study that backs the promise, when the page already states that fact. Without it, AI skips the hypothesis.

      Countdown timers, logo strips and customer photos are not used as challengers yet.

      Each round remembers the hypothesis it tested. The one that won becomes the baseline and is not tested again. One that lost only comes back after all the others were tested. With custom winning criteria, a round with no clear difference lets the hypothesis come back after two rounds.

      AI only changes a headline, promise and subtitle written as plain text. Price, offer, checkout, forms, buttons and links stay the same as the winner's, and the numbers in the text (deadlines, quantities) don't change and no new numbers appear. Once created, the challenger is a regular page: edit anything you want in the editor before publishing.

      The copy takes the losing version's place on the canvas and is already connected to the same next step as the winner. The losing page moves just below, loses all its connections and turns gray with the notice Left the A/B test in round 1 (hover over the notice to see why). Delete the page if you no longer need it, or connect it to a variation again to bring it back into the test.

      Once the test ends, the card in edit mode marks the winner with the Winner badge and the losing version with Traffic ended. Click either one to open the result in Analytics mode.

      The test limit still applies; if it was a date that has already passed, it moves forward by the length of the previous round (or 7 days). Check it in Test Settings before publishing.

      In the new round, the test's numbers only count from the moment you created the challenger, so the winner does not start with the previous round's visitors. The views limit of older tests also starts from zero each round. The card shows Round 2, Round 3 and so on, and Test Settings lists the Round history with the winner, each variation's rate, the hypothesis tested and whether it won and, with custom winning criteria, each round's confidence.

      A test created by Max follows the same rule: it gets a URL of its own automatically. Ask for a specific address that is already taken and Max tells you, instead of creating the test on the wrong URL.

      If it didn't work

      • Saving is blocked with a percentage warning: the splits must add up to 100%, across 2 to 5 variations.
      • The slug shows up as taken: another page or A/B test in the same project already answers on that URL, even if it lives in a different funnel. Pick a different slug and Save Settings becomes available again.
      • The test card shows the duplicate URL warning: two funnel steps share the same slug. Open Test Settings and change one of them.
      • The test URL does not open: none of the variations has a valid destination — usually a quiz with no public address. Open each variation, confirm its destination, then publish the funnel again.
      • A quiz variation gets no traffic: its quiz has no public address yet. Open the quiz card, check the address under Change domain, then publish the funnel again.
      • The card shows "No statistical confidence": with custom winning criteria on, the gap between variations did not reach the confidence level, or the total conversions stayed below the minimum. For the next test, raise the target or push the date, or lower the Confidence level.
      • An older test closed on its own sooner than expected: with the views limit, the test closes when the people across the variations reach the saved number, including tests published long ago. To keep splitting traffic, start a new round from the card, or open Test Settings, pick the new rule and save before publishing.
      • A variation's arrow shows —: that variation got no traffic in the selected range. Widen the range in the analysis panel's date picker.
      • The review says every hypothesis has already won: each text hypothesis that fits the page was tested and won. Use Create an exact copy and edit the challenger by hand in the editor.
      • The review says it couldn't find a headline, promise or subtitle: the texts at the top of the page have formatting (bold, color on part of the text), a price, a percentage or a link, or sit inside buttons, and AI does not touch them. Use Create an exact copy and edit the challenger by hand in the editor.
      • The review doesn't show the page preview: the winner hasn't been published yet, so only the list of changes appears. If the preview says a change isn't shown, the text was edited after the last publish; publish the funnel and open the review again.
      • The review doesn't generate the challenger: AI didn't answer in time or is unavailable. Click Try again; if it keeps failing, create the challenger the same way later.
      • A notice says the new text couldn't be applied: the copy was created but stays on the canvas outside the test, identical to the winner. Edit the copy by hand and connect it to the variation, or delete it and start again.
      • The card suggests the next round but has no Create challenger button: the winner is not an editor page (for example a quiz or an external page), so it cannot be copied from here. Replace the losing version manually: connect a new page to the variation and publish the funnel.
      • You cannot edit the funnel: you lack access to the workspace — see no permission.