One audit found twenty campaigns each claiming credit for a single order. The cause was not fraud. It was a tracking pixel guessing across a domain boundary it could not see past.
Plenty of stores run checkout on a dedicated third-party domain rather than their main site. A pixel-based attribution tool with no tag installed on that domain never directly observes the purchase event. Instead it reconstructs a conversion from whatever signal survives the handoff, and the session and UTM parameters that would normally identify the paid click frequently do not survive that domain boundary. A campaign that actually earned the sale can end up relabeled direct or unknown by the time the tool records it.
A CustomerLabs audit of one brand's account found inconsistent campaign ID and name formatting on the checkout side, loose enough that twenty separate campaigns each matched and claimed the identical purchase. The tool's total conversion count for that order did not reconcile to the one real sale sitting in Shopify. Nothing was broken on purpose. The matching logic was just permissive enough, and the domain boundary opaque enough, that one sale looked like twenty.
The mechanism generalizes to any pixel-based, rented dashboard reading click-shaped evidence across a boundary it cannot fully see, not to one vendor's tracking. Better signal engineering, server-side capture, a unified identity layer, narrows the gap but is still someone else's platform reconstructing an estimate. What actually closes it is an owned, reconciled record checked line by line against the real Shopify order, where twenty competing claims collapse to the one verified sale that actually happened.