Analytics & Tracking

Attribution tells you where to look, not what worked

Every platform reports the conversions it feels entitled to. Google Ads counts a view-through on YouTube, Meta counts a click from seven days earlier, and Google Analytics 4 splits the same sale across three channels using data-driven attribution. Add the platforms together and you will routinely find you sold 140% of what you actually sold.

None of them are lying. They are answering different questions, over different windows, with different rules about who gets the credit — and conversion lag means this week's spend often shows up in next week's revenue regardless. The job is not to pick whichever platform gives the number you like. It is to know which question each one answers, and to hold one figure above them that cannot be double-counted.

That figure is usually blended: total revenue over total spend, sometimes called MER or blended ROAS. It is unglamorous and it is the only number your bank account recognises. Platform ROAS then becomes what it should always have been — a steering signal inside a channel, not a measure of the business.

How do I know if my ads are actually causing sales?

You test it. Attribution models divide credit for conversions that already happened; they cannot tell you which of those would have happened anyway. That question — incrementality — needs an experiment, not a report.

The practical versions are cheaper than they sound:

• Geo tests. Hold a comparable set of regions out of a campaign, run for a few weeks, compare against matched control regions. The cleanest read available on most budgets.
• Holdout audiences. Exclude a random share of a retargeting or Customer Match audience and compare conversion rates. Usually the fastest way to discover that retargeting was claiming credit for people who were already buying.
• Platform lift tests. Conversion lift on Meta and Google where spend qualifies. Worth running, but read the result as a vendor's estimate of its own value.

Brand search is the one I would test first in almost any account. It reports beautifully, and a meaningful share of it is buying clicks from people who typed your name and were going to arrive anyway.

First-party data is the part that still works

Browser restrictions took most of the signal out of the open web. What survives is data you own and send back yourself, hashed, from your own systems.

• Enhanced conversions for Google and the Conversions API for Meta — hashed email and phone sent alongside the browser event, with matching event IDs so a conversion counted server-side is not counted a second time client-side
• Customer Match segments built from real purchase history: high-value buyers, lapsed customers, people who bought once and never came back — used for exclusion as often as for targeting
• Offline conversion import from your CRM, so an algorithm optimising for “leads” learns which leads actually became revenue

That last one changes accounts more than anything else on this page. A lead-gen account optimising for form fills will confidently find you the cheapest form fills in your market, and a good share of them will be worthless. Feeding closed-won values back is what turns a volume machine into a revenue one.

The audit

Most engagements start here. It is a fixed-scope piece of work and it produces a document, not a conversation:

1. Every tag, trigger and variable inventoried, with what it does and whether it still should
2. Each conversion action traced from user action to platform report, testing for missed and duplicated events
3. Reported conversions reconciled against your actual orders or CRM records
4. Consent flow tested in both states — granted and denied
5. A prioritised fix list, severity-ranked, with the revenue impact of each where it can be estimated

Tools

Google Analytics 4, Google Tag Manager (web and server), Google Ads conversion tracking and enhanced conversions, Meta Pixel and Conversions API, Microsoft Clarity, HubSpot, Looker Studio — and ConvoPixel, the free WordPress plugin I built for ChatGPT Ads conversion tracking.

How do I know my tracking is wrong?

The usual signs: platform-reported conversions that do not match orders in your back office, a conversion count that jumped or fell on a date nobody can explain, thank-you page refreshes counted as extra sales, or GA4 and Google Ads disagreeing by more than about 10%. If any of those sound familiar, the tracking is worth a look before anything else.

What is server-side tracking and do I need it?

Instead of the browser reporting a conversion directly to the ad platform, your own server does. It survives ad blockers, ITP cookie limits and closed tabs. If a meaningful share of your audience uses Safari or blockers — which is most audiences now — it typically recovers conversions the browser was losing.