When their tracking vendor gave notice, WoodWatch had seven working days to replace the infrastructure every conversion depended on. We rebuilt it on a server they own, fixed the defects the old one had been hiding, and cut over without losing an order.
WoodWatch · Wooden watches, direct to consumer · Europe and global · woodwatch.com
01 · The problem
WoodWatch sells wooden watches across two storefronts, one for Europe and one for the rest of the world. Their server-side tracking, the layer that carries every purchase to Google, Meta and Microsoft, was run by an outside vendor on the vendor's own infrastructure, for a fixed monthly fee. It worked, mostly. Nobody at WoodWatch could see inside it.
We had already audited the setup a few weeks earlier. That audit recommended a phased repair: fix the worst defects first, clean up second, migrate when convenient. The notice removed the option of doing it slowly, so the whole programme collapsed into a single build.
02 · What the audit found
Worth saying plainly, because what follows is critical of specific components: the measurement was fundamentally working. Orders were recorded, the ecommerce funnel fired end to end, and reported revenue matched the back office at topline. The problems were specific, and every one of them was distorting a decision.
Critical
Found. The browser pixel and the server were both reporting purchases with no reliable shared key, so Meta could not tell that the two were the same event. Some data layer events were also pushing two or three times per action, compounding it.
Why it mattered. Meta's bidding optimises against the conversions it is told about. Every ROAS figure, every cost per acquisition and every budget decision on the account was being made against a number inflated by roughly a third. This was the single most expensive defect in the setup.
Critical
Found. WoodWatch's own consent platform and their storefront's own settings were both configured correctly. But the vendor's script loaded ahead of both and set its own tag behaviour first, so what the site's own configuration said and what the tags actually did could differ.
Why it mattered. This is the ownership problem in miniature. The client had made the right decisions and could not make them stick, because the code that overrode them belonged to someone else. It also meant the recorded audience sizes and event counts were larger than the real thing. Moving this onto their own server made their configuration the one that actually applies.
High
Found. Events pushing multiple times per action. Currency written to several different keys, so some hits arrived with none. Checkout value summing unit prices while ignoring quantity, so any multi-item basket under-reported. And two different product identifier fields used interchangeably, which split item reporting and broke catalogue matching.
Why it mattered. The under-reported basket value is the one that costs money directly: it under-feeds value-based bidding on both Meta and Google, so the platforms were being taught that WoodWatch's customers are worth less than they are.
High
Found. Tagging conventions had never been standardised across platforms, so a share of paid social traffic was not being recognised as paid. Separately, the payment provider was appearing as a referral source, a common default that quietly takes credit for the orders that pass through it.
Why it mattered. Channel reporting is what budget conversations run on. If paid social is being credited to direct and the payment gateway is being credited for purchases, the channel mix on the page is not the channel mix in reality.
Medium
Found. More than ten analytics properties. A tag for a Google product retired in 2023. Duplicate conversion tags passing different order IDs and values. Enhanced conversions hardcoded to a single market's account. A third of the advertising audience library empty, and a majority of tracked events with no report, audience or import reading them.
Why it mattered. None of it was urgent on its own, and none of it is unusual: this is what every setup looks like after a few years of platform changes and people moving on. It makes the data harder to use than it needs to be, and it is exactly the kind of housekeeping that only ever happens when a migration forces the issue. The rebuild was the moment to do it.
What Meta was reporting, versus what happened
03 · The rebuild
The principle was simple. The platforms stop disagreeing when they stop being fed separately. Everything routes through a single server container that WoodWatch owns, on their own subdomain, in their own account, with version history and a rollback button.
04 · The cutover
The audit had assumed three to four weeks of running old and new side by side. We had days. The answer was not to skip the parallel run but to compress it, and to make every step before the deadline reversible.
Cutover sequence, built so every step before the deadline could be undone
The old and new paths ran together for the final three days, which temporarily double counts. We flagged those three days as unusable for reporting in advance. That is the price of a safe cutover, and it is worth paying.
05 · What they own now
Recurring cost of the server layer
| Before | After |
|---|---|
| Server run by a vendor, on vendor infrastructure | Server owned by WoodWatch, on their own subdomain and in their own account |
| No access, no version history, no export | Full version history, preview mode and one-click rollback |
| Tag behaviour set by a script nobody at the company could edit | The client's own configuration is the one that applies, enforced again on the server |
| Meta reporting roughly a third more purchases than happened | One deduplication key shared by browser and server, and event ID reportable so it stays fixed |
| Basket value ignoring quantity, currency written several ways | Value recalculated from price and quantity, one currency resolver, one product identifier |
| Enhanced conversions working in one market | Market-driven lookups, working in every market |
| A fixed monthly fee | Roughly 90% less, on metered infrastructure, with the container portable to any future partner |
How this was verified. The full page load sequence recorded live and read line by line, so the order in which each script ran could be established rather than assumed. Container previews on both storefronts through a complete funnel. Outbound API request and response bodies inspected for every platform, not just tag status. Meta test events confirmed as deduplicated in Events Manager before publish, with the test code removed afterwards. Daily reconciliation against the client's back office from three days before cutover to three weeks after.
If you are paying monthly for infrastructure you cannot see inside, a tracking review will tell you what is actually running, what it is costing you in distorted bidding, and what it would take to own it.
Book a tracking review