Attribution PlaybookMeasurement that survives the browser

Double Counted Purchases: The event_id Mismatch Behind Inflated Meta Numbers

Ads Manager says 412 purchases. The store took 296 orders. The gap is not fraud and it is not optimism; it is arithmetic. Every purchase fired twice, once from the pixel and once from the server, and Meta never recognised the pairs as the same event.

The pairing rule, stated once and precisely

Meta's deduplication works on two keys. The browser event's name must match the server event's event_name, and the pixel's eventID must match the Conversions API event_id, for the same pixel, with both copies received within 48 hours of each other. When the copies do not meaningfully differ, Meta generally keeps the first one received. Miss any element, a name spelled Purchase in one place and purchase_complete in the other, an ID present on one side only, and the events are strangers. Both count. The dashboard is not lying about what it received; it received two events.

The failure that produces most double counts is the independently generated ID. The browser generates a random event ID, the server generates a different random ID for the same order, and the pair can never match, because matching is the entire mechanism. The ID has to be a value both sides already know. For a purchase that value is the order number, which is why Meta's own server event reference suggests an order or transaction ID. For events with no order behind them, the browser must pass its generated ID to the server with the request, deliberately, in the payload.

Diagnosing it in Events Manager before you touch code

Events Manager shows, per event, whether it arrived from browser, server, or both, and whether deduplication happened. The healthy signature for a redundant setup is a high share of events marked as received from both sources and deduplicated. The sick signature is server and browser volumes each roughly equal to your order count, added together in the total. Check the overlap before checking anything else, because a store can have perfect event coverage and terrible attribution purely from this one defect, the server side cousin of the offline purchase gap.

There is a fallback matching method based on fbp or external_id plus event name, and it is weaker in a specific way: it generally only works when the browser event arrives first, and it does not merge two events from the same source. Treat it as a safety net that catches some pairs, never as the design. If your deduplication rate depends on the fallback, your primary keys are broken and the fix is upstream, in how the ID is created and shared.

The repair, and the test that proves it

Pick the ID source per event and write it down: order ID for Purchase, a browser generated ID passed through the data layer for AddToCart and Lead. Send it in the pixel's eventID parameter, in the fourth argument position, not buried inside the event data where it is treated as a custom property and ignored for deduplication; that placement error alone accounts for a steady share of broken setups. Send the identical string as event_id in the server payload, with the identical event name, to the same pixel.

Then prove it with one test order. Place it, find the event in Events Manager, and confirm it shows both sources and a deduplicated status. One order, checked by eye, is worth more than a week of dashboard watching, because the dashboard total will drift down gradually as the fix takes effect and you will never be sure which day it started working. Date the test, keep the screenshot, and expect reported purchases to fall toward your order count. That fall is the measurement getting honest, and it will need explaining to whoever celebrated the old number.

Frequently asked questions

Why does Meta report more purchases than my store took orders?

Usually because the pixel and the Conversions API each reported the same purchases and deduplication failed, so both copies counted. Check in Events Manager whether events show as received from both sources and deduplicated, or as separate browser and server events.

What exactly has to match for Meta to deduplicate an event?

The event name and the event ID must match across the two copies: the pixel eventID equals the server event_id, the names are identical, both go to the same pixel, and both are received within 48 hours of each other.

What should I use as the event_id for a purchase?

The order number. Both the browser and the server already know it, which is the requirement. Meta's server event reference suggests an order or transaction ID for exactly this reason. Random IDs generated independently on each side can never match.

Does the fbp fallback make event_id optional?

No. Matching on fbp or external_id is a fallback that generally only works when the browser event arrives first, and it does not merge events from the same source. Use a shared event_id as the design and let the fallback catch stragglers.