Two Dashboards, One Campaign: Reading Platform Reporting Sceptically

Open Meta Ads Manager and Google Ads on the same week and they will not agree. Neither is lying. They are answering different questions, under different rules, about the same people — and if you add their answers together you get a version of the week that did not happen.

Every platform is its own witness

A platform reports the actions it believes it caused, inside a window it chose, using its own definition of a view. That is a reasonable thing for a platform to do and a dangerous thing to read as a total. Someone sees a film on TikTok, searches the brand name two days later, clicks a Google Search ad and buys. Both accounts will claim that purchase, each correctly by its own rules. The overlap is not extra work done; it is one event counted twice.

So the first discipline of paid media reporting is refusing to sum. Channel numbers describe channels. Only a source outside the channels — your own analytics, your own order records — describes the business.

Last-click is a convention, not a finding

An attribution model is a rule about where credit lands, chosen in advance. Change the rule and the ranking of your channels changes while the week itself stays exactly as it was. That is worth sitting with: if switching models changes which channel looks best, the model is doing the arguing, not the channel.

Last-click in particular rewards whatever stood nearest the door. Branded search is the clearest case — it converts beautifully because the person had already decided, somewhere further back, to type the name. Read it as a harvesting surface. Read it as proof of demand creation and you will lose a year congratulating the final step of a journey that was built earlier.

Read these before the headline

• The attribution window, and whether anyone changed it mid-flight. A metric that improves on the day the window widens has not improved.
• What counts as a view on that surface. A moment of muted autoplay is not a viewing, and the two often arrive under the same word.
• Frequency alongside reach. Reach alone cannot tell you whether you spoke to a great many people once or to a small group relentlessly.
• Whether the audience definition changed while the numbers moved.
• Whether creative was replaced in the same week the metric moved. If it was, you now have two explanations and no way to separate them.

Incrementality is the argument worth having

Last-click asks who was standing nearest the door. Incrementality asks whether the door would have opened anyway. Only the second question describes what the campaign added, and it cannot be answered from inside a dashboard, because a dashboard holds no record of the people who were shown nothing.

It can be answered structurally. Hold a comparable market out of the flight and watch what it does without you. Stagger a launch geographically instead of switching everything on at once. Turn one channel off for a defined period, long enough to clear its own attribution window, and read the unexposed baseline. A holdout is uncomfortable — it deliberately declines to reach people who could have been reached — and it is still the only design that answers the question instead of restating it.

What good reporting sounds like

It names the source of every figure and never blends two sources in one row. It reports reach and frequency together, and creative separately from audience, so a movement can be traced to a cause. It states plainly what was changed since the last report. And it distinguishes between what the platform claims and what the business can see — because those two drifting apart is not a reporting error. It is the thing you most need to know.