Conversions API (CAPI)

The Conversions API sends conversion events from your server to Meta directly, rather than relying on the browser to report them. It exists because browser-only tracking now loses a meaningful share of conversions to tracking prevention, ad blockers, and consent choices, and those losses are not random — they concentrate in particular browsers and particular audiences, which biases the signal as well as shrinking it.

The reason this matters more than the reporting improvement is that the same events train delivery. Meta optimizes toward the conversions it is told about, so a pixel seeing 70% of purchases is not simply under-reporting by 30%; it is teaching the algorithm to find people who resemble the 70% it can observe. Restoring the missing events changes who gets targeted, not just what the dashboard says.

Deduplication is the whole implementation, and it is where most setups fail. Server and browser both report the same purchase, and Meta collapses the pair using a shared event ID plus matching event name and timestamp. Get that right and one purchase counts once. Get it wrong and it counts twice, which inflates reported conversions, halves reported CPA, and quietly corrupts optimization by telling the algorithm that a single buyer was two. The failure looks like success: performance appears to improve immediately after launch, which is exactly why it survives.

The audit is simple and worth doing before trusting any lift. Compare Meta’s reported purchases against orders in your commerce platform for the same window. Roughly equal is healthy. Meta materially above your store is duplication, not incrementality. Then check the event match quality score, which reflects how many hashed customer parameters each server event carries — email, phone, name, location, external ID. More parameters means more events matched to real people, and match quality is the difference between events that arrive and events that are useful.

A worked illustration of why coverage compounds: at 70% event coverage, a campaign that genuinely produced 300 purchases reports 210, so its true CPA looks 43% worse than reality and the campaign loses budget arguments to a rival with better tracking rather than better performance. Coverage gaps do not merely understate results; they misallocate budget between campaigns unevenly.

Two closing cautions. CAPI does not exempt you from consent — events for users who declined tracking should not be sent, and a server-side channel makes it easier to send them by accident. And hash customer parameters before transmission, following Meta’s normalization rules, since incorrectly normalized hashes match nothing and produce the appearance of a working integration with none of the benefit.