Every dashboard in ecommerce has a number that says how much revenue a feature drove. Most of those numbers are generous. Ours are not, on purpose, and it's worth explaining exactly what that means before you compare it to a number from another tool.
The rule
A recommendation gets credit when a shopper opened that exact product from that exact block, and bought that exact product in the same visit. Search revenue works the same way, per search term. Recovery and win-back emails get credit only when the shopper came back through the link in that email and completed checkout.
Read that again slowly, because each clause is doing work. Not "a recommendation was on the page somewhere." Not "the shopper visited again within a week." The click has to lead to that product, and the purchase has to happen in the same visit.
What this rules out
It rules out crediting a sale because a recommendation block happened to be rendered on a page the shopper passed through. It rules out crediting a purchase that happens three days later, even if the shopper clicked a recommendation on their first visit — because there's no honest way to prove the block caused that later purchase rather than the shopper simply deciding to buy anyway. It rules out double-counting a sale to every feature that was technically present during the session.
Loose attribution flatters every feature on the page at once. If five different tools each claim credit for the same order, at least four of those claims are wrong, and there's no way to tell which one from the dashboard.
What it doesn't rule out
A product that a shopper found through search and also saw in a recommendation block can legitimately show up in both reports. That's not double-counting — both did real work in that visit, and it's noted on both pages rather than hidden. The goal isn't to make the numbers small for their own sake; it's to make each number defensible on its own.
Why this makes the numbers useful instead of just impressive
A stricter number is a number you can act on. If a block's attributed revenue goes up after you change its ranking logic, you can trust that the change caused it, because the measurement wasn't already inflated with credit it didn't earn. If you're deciding whether to keep paying for a feature, a number built on same-visit, same-product attribution is the one that survives scrutiny from a finance team asking hard questions.
The honest framing is that these numbers are a floor, not a ceiling. The true influence of a well-placed recommendation block is probably somewhat higher than what's attributed — shoppers do sometimes come back later and buy something they first noticed in a block. We just don't claim the part we can't show.