The engine, explained
How A/Bee works
A/Bee writes hypotheses about why your visitors might convert, tests them as copy on real traffic, and files what each one proved. That runs as a single cycle of six stages, for as long as you leave it on: you set the goal and decide whether to approve, and the engine does the other four.
It changes copy. Headlines, subheads, button labels, body copy. It does not change your page. Nothing moves, nothing is reordered, and no CSS is ever written.
The optimization loop
It closes. That is the point: the result of one test is what picks the next. Each pass narrows what is worth trying, so the first experiment is a guess informed by your site and the tenth is chosen from everything the nine before it proved or ruled out.
- 01Goalname the conversion
- 02Hypothesizeangles aimed at your audience
- 03Approveoptional, can be switched off
- 04Testreal visitors decide
- 05Promotethe winner becomes control
- 06Compoundthe belief is filed either way
and round to 01 again
- The planner
- Picks the next test, and the surface that can carry it: the whole funnel, one page, or a single element. Proven beliefs carry the most weight, and a share of every round is held back for an idea that is not a refinement of the current winner, because a loop that only ever polishes what already works stops finding anything.
- The registry
- Holds every belief and its status. Untested, promising, proven, disproven, exhausted.
What it measures
A goal is the action you want more of. The scan reads your site, finds the goals already on it, and suggests them. You choose which one an experiment optimizes, and which one must not suffer to get it.
The last row is the one thing a scan can never see: a thank-you page only exists after someone has already bought. A/Bee says so rather than guessing, and hands you the line to drop in. window.abee.track('purchase')
You can define your own goals the same way. A page visited, a form submitted, a click, or an event you fire from your own server when the sale is confirmed.
Goals found on a store, and how each one fires
- Viewed a producturl
- Visited pricingurl
- Newsletter signupform submit
- Added to cartclick
- Reached checkouturl
- Completed purchaseneeds one line
How a winner gets called
All three have to be true at once. Miss any one and the test keeps running.
- 01 A Bayesian posterior past 0.975
- 02 The same verdict on two consecutive reads
- 03 Enough data in each arm that a lucky handful cannot decide it
A result that shows up once and vanishes was noise, and that is how it gets treated.
- The guardrail
- You name a goal to optimize and a goal to protect. A change that lifts the first but costs the second is blocked, not shipped. That is what makes it safe to aim at a nearer step for faster signal, say reaching checkout, without quietly trading away the purchases at the end of it.
- After promotion
- A slice of traffic keeps seeing the old copy, so a win that quietly reverses gets caught instead of standing forever.
- When nothing separates
- A test that cannot tell the variants apart is retired with a reason rather than left running, and the reason feeds the next round.
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