Fair credit: which step actually earned the sale

One sale, three steps

Say your funnel tests three things: a headline on the landing page, an offer on the product page, and the label on the checkout button. A visitor sees headline A, offer 2, and the "Buy now" button, then spends $80.

Which of those three earned the $80?

By default Bandit gives all three the full $80. We call that per-step credit. It is simple, it starts learning on the first visitor, and for most funnels it points the right way.

The problem with giving everyone the full order

Not every step matters equally. Swapping your headline might change what visitors spend by $15. Swapping the button label might change it by 50 cents. Per-step credit can't tell those apart, because both get the same $80 on every order. A version can look good just because buyers happened to see it, not because it did anything.

What fair credit does

Fair credit gives each step credit in proportion to how much it actually moves the payout. A step that changes what people spend gets a lot of credit. A step that barely matters gets very little.

To work that out, it compares visitors who saw different versions of one step while the other steps were the same. If visitors who saw headline A spend $15 more than visitors who saw headline B, with everything else matched, headline A is credited with that $15. If the two button labels come out within a few cents of each other, neither label gets much credit, whichever one the buyer saw.

Here is how the same $80 order might look under each model:

Under per-step credit, all three look equally responsible. Under fair credit, the headline clearly earned the most, the offer helped some, and the button hardly mattered. Bandit then sends traffic based on those differences, so the steps that move revenue get optimized hardest, and a version with no real effect stops collecting traffic it didn't earn.

Why it matters more with Bandit

Bandit moves traffic toward what is winning, so the steps don't stay independent. If headline A is the strong one, most visitors see it. If one button was shown more often with headline A, that button collects credit for orders the headline won, and the extra traffic makes it look better still. Fair credit separates the two, so the button's credit falls back to the little it actually adds.

A resolved visitor is one with a final payout: a real order, or a $0 once they have been idle long enough that Bandit counts them as not buying.

When it unlocks

Separating the steps takes more data than crediting everyone. Fair credit stays locked until both of these are true:

Once both pass, the panel says Fair credit (ready) and an admin can turn it on. Bandit trains the fair credit model in the background from the first visitor, so turning it on uses all the history you already have.

Steps that work together

Sometimes two steps really do depend on each other. A playful headline might do best with a playful button, and a serious one with a serious button. After fair credit is on, Bandit watches for pairs of steps like that. When a pair has enough visitors in every combination, at least 30 each, and the effect is clearly real, an admin can turn on that pair, and Bandit will learn which combinations work best together.

Turning it off

You can switch back to per-step credit at any time with Revert to per-step credit. Nothing is lost. The per-step model kept learning the whole time fair credit was on.

Should you turn it on?

If your funnel tests more than one step and the Results panel says ready, yes. You get a truer read of which versions earn money, and traffic stops flowing to versions that only look good because of what they were shown with.

If you test one step, or the funnel is still short of 500 resolved visitors, per-step credit is doing the job. Keep collecting. For more on how funnel tests fit together, see funnel experiments.