3 min read
In brief
- Purchase is the event that most benefits from a server copy, because it is the moment you hold the most identifiers and the highest value at stake.
- Upper-funnel events gain little from server-side delivery and lose the browser signals that make them attributable.
- The wrong configuration is not 'no server events' — it is duplicate events with mismatched ids, which inflate counts and teach the algorithm the wrong thing.
The principle: match the event to the data you hold
A server-side event is worth sending when the server knows something the browser cannot report reliably. At checkout, the server holds the order value, the currency, a verified email, sometimes a phone number, and a durable order id. That is a far richer event than a browser firing into an ad blocker.
At the top of the funnel, the opposite is true. A PageView from an anonymous visitor has no email and no order. The browser at least carries the cookie and the device signals. Moving that event server-side strips the only identifiers it had.
| Event | Where | Why |
|---|---|---|
| Purchase | Both, same event id | Highest value, richest identifiers, and the one a lost browser hit genuinely hides |
| InitiateCheckout | Both, same event id | Reliable on the server, useful in the browser for optimisation volume |
| AddToCart | Browser primary | Server copy adds little; the browser is the only place the intent happened |
| ViewContent | Browser | Anonymous by nature; a server copy cannot match better than the cookie |
| Lead / Contact | Server copy worth adding | You hold the email; the browser often loses the submission |
The part that decides whether this helps or hurts
Two copies of a purchase are only correct if the platform can tell they are the same purchase. That is what event_id is for: the browser event and the server event carry the same id, and the platform merges them into one with both sets of signals.
If the ids differ — or if one side omits the id entirely — the platform counts two purchases. Spend looks efficient, revenue looks higher than the store, and bidding learns from a phantom.
- Generate the event id once, at the order, and pass the same value to both sides.
- Never regenerate the id on a retry; a retry is the same event.
- Log both copies with their ids so deduplication is provable rather than assumed.
- Check the platform's own event view for a duplicate flag before scaling spend.
What to actually send
- Hashed email and phone from the order — normalise before hashing (lowercase, trimmed) or the hash will not match the platform's.
- External id: a stable customer identifier, which improves matching across sessions without sending raw personal data.
- fbp and fbc from the browser cookie where you have them, passed through to the server event.
- Client IP and user agent captured at the request, not inferred later.
- Value and currency exactly as charged — a zero-value purchase quietly breaks optimisation and is one of the most common server-side mistakes.
On retries
Retry on 429 and 5xx. Do not retry a 4xx — a validation or policy rejection is not transient, and repeated attempts look like abuse rather than resilience.
Send purchase and checkout events from both sides with one shared id, keep the top of the funnel in the browser, and log the ids so you can prove the merge. That is the whole configuration — everything else is volume for its own sake.
