Field notes on tracking.

Cost sits in the ad platforms, revenue sits in the CRM, and campaign name is not a join key. What it takes to put them in one row.

MMM is pitched as the answer to signal loss. What it actually fixes, what it quietly inherits from your data, and where it leaves a gap.

Vague tracking requests come back as another sprint. What a data layer brief has to pin down before a developer touches it.

More conversions than the client has orders. Here are the four things that make Google Ads count one event twice.

The checkout domain shows up as a traffic source, sessions double, and Google Ads disagrees with the order table. Here is why.

One conversion traced from the browser to the ad platform, and the three hops that decide whether a migration pays off.

Same property, same week, three answers. What model disagreement really tells you, and which part of it is missing data.

The Core Web Vitals objection, answered honestly: what really leaves the browser, what stays, and when a container costs you speed.

A dip can be August, a broken tag, or a report that is not finished yet. The order you check in decides how fast you know.

An email address in a URL, a phone number in a site search term. How personal data reaches analytics, and what it costs you.

The honest answer is not a yes or a no. It is a description of what your setup does, and most agencies cannot produce one.

Cleaning the export is the last step, not the first. The five things that change in a client dataset, in the order that works.

A tracking break makes no error, only a smaller number. Why silent data loss reaches the client report before it reaches yours.

The drop has a date on it and the banner is the obvious suspect. Reading the shape of the drop tells you what actually broke.

Four different empty fields hide behind one label in GA4, and three lookalike values get blamed for it. Telling them apart is the fix.

Shopify is not right and GA4 is not broken. They are two ledgers, and only one of them decides how the client's budget gets bid.

Spend held. The client's own order count held. Meta fell anyway. That is a measurement break wearing a performance break's clothes.

Nobody switched these features off. They were never switched on, and the data to feed them has been sitting there for months.

The slow part of monthly reporting is not the copying. It is deciding which of your three conversion numbers the client sees.

Quota errors and numbers that disagree are not chart problems. They are data source problems, and they share one fix.

One banner, five markets, five different numbers. The rule your setup has to satisfy changes at the border, and so does the data.

Thirty accounts, thirty dialects. Keeping tracking consistent is a specification problem, not a dashboard problem.

Build versus buy is the wrong question. Sort your measurement stack into layers, then decide which ones the client has to keep.

You went first-party, the container sets the cookie server-side, and Safari still caps it at seven days. The reason is DNS.

Invalid click filtering does not cover the conversion that click triggered. What that does to your bidding, and where to filter instead.

The two counts measure different objects, after different filters, in different time zones. Which one to trust, and how to shrink the rest.

More leads at a lower cost per lead, and the same revenue. The optimization target is usually the reason, not the campaign.

Lead volume up, revenue flat. What has to be true before closed deal value can reach the ad platform, and when it cannot.

Every server-side post is written for a webshop. What actually changes when the conversion is a form fill, and where the case falls apart.

ROAS and POAS divide by the same number, and it is the wrong one. How deep a profit figure has to go before it changes a decision.

The logins are the easy part. The question that decides whether a client's measurement survives the handover is who owns it.

Taking over an account nobody documented, without pretending you can repair the numbers the client has already seen.

Your redirect map protects rankings. It does nothing for visitor identity, ad click parameters or the before-and-after comparison.

Every release quietly moves something your tracking depends on. Why nothing errors, and what actually catches it in time.

It runs real code on your file, so the arithmetic is right. What a GA4 export left out before the upload is the actual limit.

Two of the three options share one owner, and none of them decides your consent rate. Here is what the vendor choice really settles.

The conversions in your client report are part counted and part estimated, and no column tells you where the line sits.

The accept rate on your client's cookie banner scales every number you report, and it is the one metric nobody at the agency owns.

Every tracking vendor says server-side, adblocker-proof and compliant. Five questions that tell you which one means it.

Platform ROAS can climb while client revenue stands still. Four documented reasons why, and the number that would have said so sooner.

You followed both setup guides and the numbers barely moved. The layer underneath decides what either feature can give back.

Sampling, (other) rows, thresholding and retention each bend a client report differently. How to tell which one you are looking at.

Compliant and measurable is not a trade-off. The four decisions inside Consent Mode v2 that decide which one you get.

An unowned tracking setup gets worse without anyone touching it. What owning the measurement layer actually involves.

Why tracking hours end up unbilled, and the three invoice lines that turn the work into something a client accepts.

The conversation agencies dread, in order: what to say first, what to put on screen, and why the correction usually runs in your favour.

Ads, Meta and GA4 will never agree, and that is fine. How to tell a counting difference from a real hole in your data.

Your totals look survivable because the loss is not spread evenly. It sits on Safari and iOS, where the browser deletes the evidence.

Cost is billed and complete. Conversions are collected and incomplete. That asymmetry raises a CPA no campaign change explains.

Four documented counting rules explain most of the gap. What is left after that is not a reporting quirk, it is missing data.

Sampling, 14-month retention and (other) rows. Three signs an account has outgrown GA4, and what to move to instead.

Both host the same Google server container. Here is the decision that actually changes how many conversions you recover.

Hosting is the cheap part. Here is the honest breakdown of what a server-side setup costs to own, and where the money really goes.

Threshold alerts fire on a quiet Sunday and stay silent when one conversion action dies. The fix is not a tighter number.

The model is rarely the problem. Three things in your data decide whether an AI answer holds up on a client call.

Some of the loss you agreed to, the rest your banner configuration chose for you. How to tell them apart, and recover the second half.

Server-side sounds expensive and the payback feels vague. Here is the honest business case, and when it is actually worth doing.

You cannot justify a full-time data hire, but the missing conversions keep costing you. Here is the real maths on each option.

Clients want server-side, consent-safe measurement. Here is how to offer it under your own brand without hiring a specialist.

A quick tag test and a real audit both get called an audit, but they find very different things. Here is what a proper one covers.

Two identical first-party cookies, one lasts a week, one lasts a year. The difference quietly shortens every client's attribution window.

GA4 and Google Ads never quite agree for a reason. Here is where the missing conversions actually go, and how to get them back.

Everyone says go first-party, few explain what it changes. Here is what a tag gateway actually does, in plain terms.