First, confirm which server events are missing
Open the dataset the ad account actually uses, not one that happens to be first in the list. In the overview, read each event by connection method: the Conversions API guide describes how Events Manager lists browser and server copies with the counts received by each. Note which events have no server count, which have a smaller one than expected, and when the server count last changed. A route that never sent anything is a setup problem; a route that sent normally and then stopped is a change problem, and the date it stopped is the best clue you will get. If browser events are missing as well, the diagnosis for pixel events not showing comes first, because the browser side usually breaks for the same release.
Several patterns look like missing server events and are not.
| What you see | Why it can be expected | What to check instead |
|---|---|---|
| Server copies only for visits where the pixel also fired | The Gateway mirrors browser events, so it cannot report what the browser never sent | Whether a partner or direct integration is needed for full coverage |
| Server events with no browser count beside them | CRM, offline and other server-only events have no browser twin | That the server count follows the system that records the outcome |
| Browser and server counts that do not add up to the total | DeduplicationRemoving double counts when one conversion is reported by more than one source. keeps one copy of each matched pair | The deduplication share in the event detail, as the duplicate events diagnosis explains |
| Server events below orders | Visitors who declined consent should produce no server event | Whether consenting visitors produce one |
Then compare counts outside Meta. Over a closed week, the number of server purchase events received should sit close to the shop's order count for the same days, before any attribution is applied; the guide to reconciling Meta results with orders sets out how to run that comparison so ordinary differences are not mistaken for a fault. Finally, send one test event through the route with the test event code and look at what arrives. An event that reaches the test view proves the connection works, which narrows the search to what differs between the test and live configurations.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Server events visible in Test Events, absent from the overview | The live payload or integration settings for a test event code | A test event code left in the production payload | Remove the code from live sends; keep it in a separate test configuration |
| The integration reports successful sends, the dataset shows none | The dataset identifier in the server configuration against the one the ad account uses | The server integration sending to a different dataset | Point the integration at the account's dataset |
| Server events stopped on a known date | Platform changes, disconnections and staff or agency changes on that date | A partner integration or CRM job that stopped sending | Reconnect or rebuild the route on the system that now holds the orders or leads |
| Browser tags fire, the Meta server tag reports nothing | Requests reaching the server container; which client claims them; tag status | A server-side tag manager route that receives or forwards nothing | Repair the hop where the request disappears |
| Some server events arrive, others vanish, diagnostics show warnings | The diagnostics tab and one test payload's fields | Payloads Meta cannot use | Send every required field, format values correctly and send events as they happen |
| Server events stop for everyone after a consent change | Server events from a session that accepted consent | A consent check that blocks every event | Pass each visitor's recorded choice to the server and honour it per event |
Fix, verify and prevent each cause
A test event code left in live payloads
The Conversions API guide describes the test events view: it accepts a test event code included in server payloads during development, and events carrying the code appear in that view without entering reporting. That is useful while building and costly afterwards, because a code that survives into production sends every real purchase or lead to the test view instead of the dataset's reporting. The duplicate events diagnosis lists the same leftover among its common mistakes.
Fix: remove the code from the live configuration, wherever the route keeps it: the integration's settings, an environment variable or the server tag.
Verify: the overview shows server events again within the following reporting days, and the test view stops filling with real orders.
Prevent: keep test and production configurations separate, so the code can only exist in the one that never sends live traffic, and make its absence a release check.
Events sent to another dataset
Every server send names a dataset, and a business with several portfolios, a test dataset or a dataset left from an old site has more than one to choose from. The pixel diagnosis treats a wrong identifier as a cause on the browser side; the server side fails the same way, and less visibly, because nobody can see a server request in a browser. The guide to moving a website names a second dataset created by a new platform as one of the things a move breaks.
Fix: set the server configuration to the dataset the pixel and the ad account use, and confirm the ad account can select its events.
Verify: the same dataset lists browser and server copies of the key events, and the deduplication share appears in the event detail.
Prevent: record the dataset identifier, its owner and every route that sends to it in the account's measurement plan, as the measurement plan framework recommends for every source and destination.
An integration that stopped sending
A partner integration is the platform's own code. The migration tutorial points out that it belongs to the platform that runs it, so when a shop moves to another platform, its server events end on the day the old shop closes. The same silence follows a disconnected integration, an expired connection or a CRM job nobody owns after a change of agency. For lead businesses the loss is easy to miss: the lead generation setup notes that conversion leads starved of stage updates behave like an ordinary lead goal, so a stage feed that stops undoes the main advantage without any error on screen.
Fix: reconnect the integration, or rebuild the route on the system that now holds the orders or the lead stages; the guide to connecting a CRM to ads and analytics covers the CRM side.
Verify: server counts return and track the shop's orders or the CRM's stage changes over a closed week.
Prevent: give each route an owner, and add a check for server events to every platform, agency or CRM change.
A server container that receives or forwards nothing
Where events reach Meta through Server-Side TaggingRunning tags on a server the business controls instead of in the visitor's browser., the Meta tag is the last hop of a relay: the browser sends a request to an address the business owns, a client in the server container claims it, and the tag forwards it. The diagnosis for a server container that receives no data follows a single event through each hop and names the usual failures, from an unpublished container to hosting that cannot keep up. The Conversions API guide adds that this route only works when the client sends the identifiers the Meta tag needs.
Fix: find the hop where the request disappears and repair that hop, not the whole setup.
Verify: the server preview shows the request claimed and the Meta tag firing, and the overview lists server events.
Prevent: run the server-side tagging go-live checklist for every new destination or container release, so the Meta tag is tested before traffic depends on it.
Payloads Meta cannot use
The Conversions API guide lists the fields Meta documents for every server event: event name, event time, action source and enough customer information to match. Personal fields are normalised and hashedConverting an identifier such as an email address into a fixed string with a one-way function, so platforms can match records without the raw value. before sending, while the IP address, user agent and the browser identifiers the pixel sets are sent as they are. The diagnostics tab flags what Meta detects: missing required parameters, badly formatted values, events received too long after they happened, and deduplication failures. A batch job that sends yesterday's orders in one upload, a value sent in the wrong format or an Event ParameterA named value sent with an analytics event that describes it, such as a page location, form name or item price. left empty can each leave events missing or unusable. The exact field rules and how late an event may arrive are set out in Meta's own documentation; this page does not repeat them.
Fix: send each required field on every event, apply Meta's documented formatting rules before hashing, and send events as they happen with their original event time.
Verify: the diagnostics warnings clear, and a test payload shows every field populated as expected.
Prevent: agree a written list of fields with the developers who own the integration, as the low event match quality diagnosis recommends, and test one payload after each release.
A consent check that blocks everyone
Moving events to the server moves the consent decision with them. The Conversions API guide explains that the choice recorded by the banner has to travel with the session into the order record and be respected by the server code, so a visitor who declined gets no server event, exactly as they get no pixel event. The opposite mistake is common after a banner change: a check that reads a missing or renamed value as a refusal and stops every server event. The guide to privacy, consent and measurement covers how the recorded choice should flow between systems.
Fix: pass each visitor's recorded choice to the server and apply it per event, so consenting visitors produce server events and declining ones do not.
Verify: test one session that accepts and one that declines; only the first produces a server event.
Prevent: add both branches to the release test whenever the banner, its categories or the checkout change.
Prevention: a server-event check after every release
Every cause above follows a change: a release, a platform move, a new container, a new banner, a new agency. None of them produces an error anyone in marketing will see, and the guide to troubleshooting performance drops starts with the same warning: check that the optimisation event still arrives from both browser and server before reading anything else. A short routine catches the break in days rather than months.
| When | Check | What it prevents |
|---|---|---|
| Every site, checkout or banner release | One test purchase or lead through each route in Test Events, then the live overview | Test codes, consent checks and broken payloads reaching production |
| Every platform, agency or CRM change | Which routes send to the dataset, who owns each, and whether each still sends | Integrations that end silently |
| Weekly | Server events received against orders or CRM stages for a closed week | Slow leaks and late batches |
| Monthly | The diagnostics tab and the deduplication share for the key events | Formatting faults and drift between browser and server |
| When a route is added | The dataset identifier and the fields agreed in the measurement plan | Events sent to the wrong place or without what Meta needs |
Write the routine into the tracking outage response SOP or its equivalent, so the person who notices a gap knows who owns each route and what to record. Keep a note of every date on which server events stopped or returned, because reported conversion rate rises when missing events come back, and a later analysis should not credit that rise to the ads; the conversion rate calculator helps compare the rate before and after against the shop's own records. The comparison of the Meta Pixel and the Conversions API explains why both routes are worth protecting: the pixelAn ad platform's browser code that reports page views and events on a site back to the platform. carries browser identifiers that match well, and the server reports what browsers drop. Once server events arrive again, read their match quality, and if they arrive twice, the Meta Ads troubleshooting collection routes to the right page. The pixel setup tutorial remains the reference for the browser half of the pair.
A tile shop finds its silent server route (illustrative)
Yentrish Tiles
Yentrish Tiles, an invented bathroom tile shop, runs the pixel and a direct server integration. Three weeks after a checkout update, its weekly comparison shows 412 orders against 297 browser purchase events and no server purchases at all; the week before the update, the server had sent 405. Test Events shows server purchases arriving with full customer information, which rules out the dataset and the payload. The developer finds the test event code added while testing the update still set in the production configuration, so every live server purchase since that day had gone to the test view. Removing it brings the server column back the next day. The shop adds two checks to its release list, no test code in production and one live purchase visible in the overview, plus a weekly count of server events against orders. In the following closed week, 398 server purchases are received against 409 orders, and the deduplication share returns to its level before the update.
Common mistakes
- Reading the wrong dataset. A business with several portfolios or a leftover test dataset can look at an empty one while events arrive elsewhere.
- Expecting the Gateway to fill every gap. It mirrors what the browser sent; visits the browser missed produce no server copy.
- Leaving the test event code in production. Every live event then lands in the test view.
- Treating a consent refusal as a fault. Declining visitors should produce no server event; check the accepting branch instead.
- Fixing the whole setup at once. Follow the event to the first hop where it disappears and repair that hop.
- Forgetting the dates. Reported results rise when missing events return, and without a note the rise is credited to the ads.
