A purchase can be counted in GA4 and still be useless to the system buying the next click. The sale happened. Analytics recorded it. Google Ads did not connect it to paid media. Nothing in that sequence requires the campaign to have deteriorated.
The dangerous mistake is treating the dashboard mismatch as the diagnosis. It is only the symptom. The marketer's next decision should be smaller and more concrete: follow one affected purchase through the identity handoffs that attribution depends on.
Prove the event and the attribution separately
GA4 event collection and Google Ads attribution answer different questions. One proves that an event arrived. The other needs enough paid-media context to connect that event with an earlier ad interaction.
Depending on the implementation, GA4 may retain campaign context while the relevant click identifier or associated identity does not survive on the request used by Ads. That can leave the analytics total steady while imported conversions fall. It does not prove which handoff failed, and it does not make the visible GA4 count a substitute for the missing attribution.
Start with a purchase that should have appeared in both systems. Confirm which request carried the conversion, which campaign context remained attached, and where that context stopped being usable. A click identifier visible somewhere in a server payload is not the same as identity surviving on the conversion that matters.
Trace the failing path, not the total
Overall conversion counts can hide a break that affects only a browser, device class, entry path, or cookie condition. Segment the mismatch before rebuilding the entire setup.
The useful comparison is one working route against one failing route. Follow the paid-media context from landing through purchase. Check whether an identity or cookie change separated the conversion from its earlier session. Then allow the normal import window to pass before treating an apparent recovery as causal.
Changing how the GA4 client manages cookies may repair fragmented context, but an early improvement does not establish why the count moved. The change could coincide with normal reporting variation. That uncertainty is why the complete conversion window matters more than a promising preview.
A platform bug is the last label, not the first
A server-side Google Ads tag can appear to fire while sending no conversion request. An upstream customer-data flag and an empty augmentation step are one credible explanation for that abort, not a general diagnosis for server-side conversion loss.
Transformations, consent parameters, browser-side customer-data settings, and upstream fields still need to be ruled out before escalation. If those checks do not resolve the abort, escalation is justified—but another implementation may still fail for an ordinary local reason. Declaring a platform defect too early ends the investigation at the point where it should become more precise.
A workaround changes the failure profile
Moving a conversion back to the browser can protect continuity while the server path is investigated, but it is not lossless. The browser path may again face blockers and privacy controls that the server implementation was intended to reduce.
The choice is not server good, browser bad. It is whether the known failure in the current path is worse than the expected loss in the temporary alternative. Run the workaround as a bounded continuity measure, keep its conversion separate enough to inspect, and do not call the migration complete because a number returned.
When GA4 and Google Ads disagree, pick one affected purchase and trace every identity handoff before touching the setup again.