How the Conversions API works
A browser tag reports a conversion from the visitor’s device. The Conversions API reports the same conversion from your own systems — your website’s server, your ecommerce platform, your CRM, or a server-side tag container sitting between them. The event travels over a direct, authenticated connection to the advertising platform, so nothing about it depends on the visitor’s browser, connection or extensions.
Because the send comes from your side, you can attach information the browser never had. Meta’s version, usually written CAPI, accepts customer details such as email and phone in hashed form so the platform can match the event to an account. You can also send events that never touch a web page at all: an order confirmed by phone, a deal marked won in the CRM, a refund.
The important design point is that both sends are meant to coexist. The browser and the server report the same action, each carries the same event ID, and the platform uses that shared reference to count one conversion rather than two.
Why the Conversions API matters
Browser-reported events go missing for reasons you cannot control — blockers, closed tabs, cookie limits, a mobile connection that drops before the request completes. Every missing event is both an under-reported result and a signal the campaign never learned from. Automated bidding is trained on the conversions it is told about, so patchy reporting produces patchy optimisation, and the account quietly bids as though your best customers do not exist.
It also lets the platform learn from what actually happened rather than what happened on the website. If half your enquiries turn out to be unusable, sending only the qualified ones back through the API teaches the campaign to chase the good ones.
Common mistakes with the Conversions API
Turning off the browser pixel after connecting the API is the classic error. The two sources cover different gaps and the platform expects both. Sending events without a shared identifier is the next one, because the result is silent double counting that makes cost per acquisition look far better than it is — deduplication only works when both copies agree on the event name and the identifier.
The rest are process failures. Customer fields sent without hashing or without normalising them first. Consent ignored on the assumption that a server send is outside privacy rules, which it is not. And nobody watching the connection afterwards, so an integration that breaks during a site update goes unnoticed until the monthly report looks wrong.
How to act on it
Decide first which events genuinely need it: purchases and qualified leads, not every button click. Agree where each will be sent from, generate one identifier per action at the moment it happens, and reuse that same identifier on both the browser and server copies. Then verify in the platform’s event tools that the events arrive and are reported as deduplicated before you trust any number built on them.
Most businesses reach this point through a pixel and Conversions API implementation rather than building from scratch, because the work is less about code than about agreeing what counts as a conversion and keeping it working afterwards.
