Measurement that holds up when someone asks how the number was made.

When two dashboards disagree, the cause is usually a definition

I have spent a lot of my career on this particular argument. Much of it was in private-equity-backed rollups, where you inherit a dozen ad accounts built by a dozen different people. Every one of them has a conversion action called something reasonable like "Lead." In one account that means a submitted form. In another it includes a click on a phone number. In a third it fires on the thank-you page and also on the form plugin's own event, so it counts every lead twice. Then someone asks for one number describing the whole portfolio, by Thursday.

The instinct in that situation is to argue about which platform is right. That is the wrong argument. The gap between two systems is almost always the sum of specific, findable counting differences: different attribution windows, different definitions of a session, one system deduplicating and the other not, bot filtering in one place and not another. My approach is to enumerate the differences until the gap is explained, then pick one written source of truth per metric and get everyone to agree to it. A stable gap you can explain is fine. A gap that moves every month is a bug.

GA4 is a directional instrument, not a source of truth

This gets people into trouble because GA4 presents its numbers with a lot of confidence. Underneath, it is doing behavioral modeling for unconsented traffic, applying thresholds that suppress small numbers in some reports, sampling above certain volumes, and making its own decisions about session attribution. None of that is a defect. It is a system built to describe trends across a lot of traffic, and it does that job well.

What it is not is a ledger. If your revenue reporting, your commission calculations or your board deck are pulling from GA4 as though it were an accounting system, that is a problem waiting for a quarter-end. The rule I work to is straightforward: your CRM or order system is the ledger, and GA4 tells you which direction things are moving and where to look next. Both are useful. They answer different questions.

Attribution models are opinions with math attached

Every attribution model encodes a belief about who deserves credit. Last-click believes the final touch did the work. Data-driven models believe the pattern in your conversion paths is a reliable guide. Neither is objectively correct, because credit assignment is not a fact waiting to be discovered. It is a decision about how you want to run your business.

That means choosing a model is a business conversation and not a technical one, and I will treat it that way. What I care about is that you pick one deliberately, understand what it over-rewards, and stop switching models to justify decisions after the fact. The second most expensive thing I see, after last-click killing upper-funnel work, is a team that changes attribution model whenever the current one makes a favored channel look bad.

The work itself

• Conversion tracking audits: what fires, what double counts, what silently stopped working after a site release
• GA4 configuration, including events, key events, and the settings that quietly change what you see
• Google Tag Manager builds and cleanups, including containers nobody has documented
• Enhanced conversions and offline conversion imports, so closed revenue from your CRM makes it back into the bidding
• Consent Mode v2 for teams with European traffic, including the region defaults and the banner behavior
• Reconciliation work when the ad platforms, GA4 and the CRM all report different numbers

If you want to see the approach rather than read about it, this site runs the same setup I would build for you: consent defaults ahead of the tag manager, a single conversion event fired only after the server accepts the lead so bot submissions never count, and enhanced-conversion data assembled in the browser. The privacy page documents exactly what is collected, which is the part most implementations skip.

How engagements usually work

Most analytics work is project-shaped: an audit, a rebuild, a migration, an implementation. Some clients keep me on monthly to hold the measurement steady while their site and campaigns keep changing, which is usually the right call if you release often. Analytics work also pairs naturally with paid search, because bidding is only as good as the signal underneath it.