You Lost 9 Purchases and Gained 16 New Customers. Which Number Do You Trust?
The same account, the same week, two numbers moving in opposite directions. Both came from the same platform.
The scenario is more common than it sounds
A week-over-week reporting pull shows total purchases down. In the same account, in the same week, the platform's own new-customer count is up. Both numbers came from the same pixel, the same account, the same reporting window. If the business genuinely sold less, new customers shouldn't be climbing. If it genuinely gained customers, purchases shouldn't be falling. The platform reported both anyway.
What's actually going on underneath
Purchase counts and new-versus-returning classifications are usually built from different modeling layers, refreshed on different schedules, and reconciled against different identity graphs. A visitor who declined cookies on one visit and consented on the next can get counted as a new customer twice, or dropped from the purchase count entirely, depending on which model handled which event. Neither number is lying exactly. They're both estimates, built on different assumptions, and nothing forces them to agree with each other.
The fix is one reconciled record, not two competing ones
The way out isn't picking whichever number is more convenient that week. It's building one first-party record, tied to real order and customer IDs from your own systems, that both purchase volume and new-customer status get measured against. When there's only one source of truth, the two numbers stop contradicting each other, because they're both derived from the same verified ledger instead of two separate guesses.