Autopilot
Automated split testing that picks its own tests
Most tools automate the plumbing: they split traffic, count conversions and leave the thinking to you. Deciding what to test next is the actual work. On autopilot, A/Bee names the next bet, writes it, runs it and files what it proved, then does it again.
What it does with nobody watching
Every one of these is a decision you would otherwise be making yourself, on a day you probably meant to spend on something else.
- Chooses the next question. Not the next variation of the same question. It reads what it already believes about your audience and picks the belief worth testing now.
- Chooses where to ask it. A headline, a page, or every page on the path to your goal, depending on what the idea needs to be expressed properly.
- Writes the variations. Several per idea, worded differently, so a win is attributable to the argument rather than to one lucky sentence.
- Calls it, or refuses to. Promotes a winner, retires a loser with a reason, and leaves a test running when the evidence is not there yet.
- Files the result and starts over. Win or lose, the outcome attaches to the idea it tested, and the next round is chosen from a shorter list of open questions.
- Keeps looking while it is winning. Every winner becomes the control, and the safest way to beat a control is a small change to something already working. Do only that and the page becomes the best version of itself without ever becoming a better one. So part of every round is held for an idea that is not a refinement of the current winner, including one written with no argument to serve at all. Most of those lose. The ones that do not are the tests that move a page somewhere it was not going.
Your standing input: the goal, and the one thing that must not suffer to reach it.
Autopilot is a setting, not a one-way door
Approval mode holds every variation until you have read it. Your brief sets what the engine may never say. A rejection with a reason changes what gets written next. And you can drop to picking tests yourself, or writing them yourself, whenever you want.
Common questions
- Do I lose control over what gets tested?
- No. Approval mode holds every variation until you have read it and said yes, and it can be switched on per experiment. Your rejections and your brief steer what gets written next.
- Does automated testing work on a small site?
- Yes. The engine waits for real evidence instead of calling a winner early, and what it is testing is a belief about your audience rather than a button position, so what it learns does not expire while it waits. Testing on low traffic →
- Can I still run tests myself?
- Yes. Fully manual mode is on every plan, including free: you write the variations, and the engine still splits the traffic, computes the statistics and calls the winner. The AI-written and autopilot modes are on paid plans. The statistics are the same in all three.
- How is this different from other A/B testing tools?
- Most tools hand you infrastructure and leave the thinking to you. Deciding what to test next is the actual work, and that is the part this automates. See the loop →
Try it on your own site
Paste a URL and A/Bee hands back its first hypotheses for that page. Free, and no signup.
See what it would test