Why this matters
At the advanced performance marketing level you specified enhanced conversions and server events for two platforms. At expert level the proposal on the table is infrastructure, and here it arrives as an answer to every measurement complaint at once. The comparison of the two fixes says they close different gaps: enhanced conversions change what one platform's tag sends, while a server container sits beneath every vendor and someone must host and run it.
The saddlery and its two gaps
Vreiroth Saddlery is an invented online retailer of saddles, riding boots and stable supplies. A paid media manager runs Google Ads and Meta in-house, a freelance developer looks after the shop, and the owner, who also handles data protection, wants one place to decide what each vendor receives.
Vreiroth Saddlery: what the team found
| Evidence | What the team found |
|---|---|
| Orders in the shop system | About 1,150 a month |
| Google Ads purchases | About 760 a month, short since the account began |
| How riders buy | Saddle buyers return over weeks; many click an ad on a phone and order on a computer |
| Base tag | Fires once on each test order, from a phone and from a computer |
| Meta | Browser pixel only; the paid media manager wants the Conversions API |
| Web container | 13 vendor tags, 3 of which nobody can name; the reviews and chat widgets render on the page |
| Product pages | Slow on mobile, much of the weight from vendor scripts |
| Outgoing requests | The confirmation page sends the buyer's email to the affiliate network in a URL |
| Consent | The banner works, but the heatmap tag fires before the visitor chooses |
| Hosting and on-call | Nobody named; the developer works 2 days a week |
| Agency plan | All 13 tags moved in one release, the week before the spring sale; the busiest weeks run from March to May |
Match the fix to the gap
Google Ads counts fewer purchases than the shop although its base tag fires. The comparison lists long gaps before purchase and cross-device journeys, both how these riders buy, among the cases where the enhanced conversions guide says the gain tends to be large, and gives that gap to enhanced conversionsA Google Ads feature that sends hashed first-party customer data to improve conversion matching., which need no server container. The second gap is control (page weight, what each vendor receives, Meta's server integration), which the comparison and the server-side tagging basics guide give to server-side taggingRunning tags on a server the business controls instead of in the visitor's browser.. Both wait while consent wiring is broken, and with no owner for hosting the comparison allows enhanced conversions only.
Task
- Name each gap and the fix the comparison points to.
- Run the case test from the server-side guide: which requests leave the page, which vendors the server calls and who answers when the endpoint fails.
- List what must be true before any traffic moves.
- Plan the cutover one destination at a time, gated by the server-side tagging go-live checklist, and place it against the calendar.
- Say what the server container will not fix.
Worked answer
The Google gap. Switch on enhanced conversions after the heatmap fix, in the web container Vreiroth already runs: the developer passes the checkout's email address to the Google Ads conversion tag through the data layer, the tag hashes it, and the consent banner gates it. Diagnostics, one tagged test order and purchases against shop orders before and after show whether it works.
The case test. Requests removed: the unnamed tags go first, and the page-weight case rests on the scripts that can move; the reviews and chat widgets draw what visitors see, so they stay. Vendors called from the server: analytics, Google Ads conversions and Meta's Conversions API. The person on call is blank, so the case is not ready until the developer owns hosting, uptime and the domain, the paid media manager owns the container's contents, and a short runbook says who acts when the endpoint stops answering.
Before any traffic moves. Fix the heatmap tag's consent behaviour, because a server container inherits whatever consent state the browser sets, and the consent mode guide is where that state is wired. Stop the email reaching the affiliate network now, in the browser.
Vreiroth Saddlery: the cutover plan
| When | Step | Passes when |
|---|---|---|
| November | Enhanced conversions on in the web container; heatmap consent fixed; 3 unnamed tags removed; affiliate email stripped | Diagnostics read recording; tracking verification passes |
| December | Server on a first-party subdomain; developer named owner; publishing restricted to 2 people | Requests reach the endpoint and the right clients claim them |
| January | Analytics stream sent through the server beside the browser tags | Server and browser events agree over a 2-week parallel run agreed with report readers |
| January | Meta Conversions API tag, order number as the shared event ID | Server purchases deduplicate against the pixel; until they do, the pixel carries conversions |
| February | Google Ads conversion tag moved to the server | Hashed data still arrives; until it does, the web tag keeps running |
| February | Browser copies removed one vendor at a time; previous container version recorded | No vendor counted twice; cutover dates annotated |
| March to May | Nothing moves | Any destination that failed stays in the browser until June |
One destination at a time, with a way back. The go-live checklist treats each destination as a small cutover, so a failure usually holds that destination, not the project: a purchase counted twice through failed DeduplicationRemoving double counts when one conversion is reported by more than one source. would teach bidding to chase sales that never happened. The parallel runRunning the old and new versions of a process, platform or tag set side by side for an agreed period, compared on one measure, before the old one is retired.'s end is agreed before it starts, the previous web container version can be published again, and each cutover date is annotated in analytics, Google Ads and Meta, so a change in reported conversions is not blamed on the campaigns. The agency's single release before the sale leaves room for none of this.
What it will not fix. A server container passes on only what reaches it: events wrong at the source stay wrong, and visitors who declined consent stay unmeasured.
Common mistakes
- A server container as the route to enhanced conversions. The feature already works in the web container.
- Moving tags before consent is fixed. The server inherits the browser's consent state, faults included.
- Removing the browser tags on cutover day. Without a parallel run there is no baseline to show what was lost.
- A silent cutover. An unannotated date turns a measurement change into a story about performance.
