What the purchases metric counts
A purchase is a standard event your website sends to Meta when an order completes, usually from the order confirmation page, from the Conversions API on your server, or from both. Meta then credits it to an ad if the buyer clicked or saw that ad inside the attribution window in force.
Two things follow. The count is of events your site chose to send, so it is only as accurate as the tracking behind it. And it is attributed, so it answers the question “which ad did this buyer meet” rather than “which ad caused this sale”. Those are different questions, and only the first is answered by the column.
Why purchases matter
It is the event closest to money, which makes it the best thing to optimise for when there is enough of it. Delivery learns from whatever it is given; given real purchases, it looks for people who buy rather than people who browse or tap.
It is also the anchor for value reporting. Without a reliable purchase event carrying a value and a currency, return on ad spend cannot be calculated at all, and the account has to be run on proxies that are easier to game and harder to defend.
Common mistakes with purchases
Duplicates come first. When the pixel and the Conversions API both report the same order without a shared identifier, the sale is counted twice, cost per purchase looks better than it is, and delivery learns from a phantom. A matching event ID lets Meta remove the copy, which is what event deduplication is for.
Cash on delivery comes second, and it matters across Nepal and much of South Asia. If the purchase event fires when an order is placed rather than when it is paid for, refused deliveries and abandoned orders are counted as sales. The advertising then learns to find customers who order easily and pay rarely, which is a slow and expensive mistake. Where cash on delivery is normal, this deserves a deliberate decision rather than a default.
Third is treating attributed purchases as caused purchases. Some buyers were coming anyway: returning customers, people who searched for the brand by name. The column will claim them if they passed an ad. Only incremental testing tells you what the advertising added.
How to act on it
Get the event right before reading the number. Send it once, with a value and a currency, from a point in the process where the money is real, and run the Conversions API alongside the browser pixel with a shared event ID so the two are reconciled rather than added. Reconcile the total against your shop’s own order report every week.
Where cash on delivery is common, decide which moment counts as a purchase, and if the delivered share is poor, consider firing the event only on confirmed payment even though the reported volume falls. Fewer, truer events teach delivery better than a large count full of orders nobody ever paid for.
