Display has a credibility problem: its reports can say almost anything depending on which column you look at. The answer is not to distrust everything, it is to ask the right question —what would have happened without this campaign?— and to have a method for answering it. This lesson closes the module with that method.
Why Display's figures mislead
- View-through conversions: somebody saw (did not click) a banner and converted later through some other channel. With remarketing, almost every buyer has "seen" one of your banners: the campaign claims credit for sales that were already yours.
- Generous windows: 30 days of view-through by default is a lot; one day is more honest.
- "All conversions" mixes primary, secondary and view-through.
- Accidental clicks in apps and small slots: high CTR, no conversions, but "clicks" that look like traffic.
The honest columns and windows
| To judge… | Column | Window |
|---|---|---|
| Direct performance | Conversions (per click, primary) | The business's own (30 days) |
| View contribution | View-through conv. separately, never added in | 1 day (or less) |
| Awareness | Reach, frequency, brand searches afterwards | — |
And in reports for management: always with a definition of what is being counted.
The right question: incrementality
Display's metric is not "how many conversions it claimed" but "how many conversions would not have happened without it". Three methods, from most to least rigorous:
1. Holdout experiment
Split the audience (a remarketing list, say) into two random groups: one sees ads, the other does not (it is excluded). Compare the conversion rate of both groups over the period. The difference is the real incrementality. You build it with lists and exclusions (or with Google's experiment tools when they are available for your campaign type). It needs volume: hundreds of conversions in the period.
2. Geo experiment
Run the campaign in some areas (regions, cities) and not in other comparable ones, for 4-6 weeks, and compare how total sales (not just attributed ones) evolve by area. It works for awareness, Demand Gen and video. It needs comparable areas and sales you can measure by area.
3. Reading the account total (before/after)
Turn the campaign on and see whether total conversions (all campaigns
- organic if you measure it) rise beyond the trend. It is the simplest method and the least clean (seasonality, other changes), but it catches the most common case: the remarketing campaign that "converts 50 a month" while the total does not move.
How to present Display to whoever pays
- Conversions per click with their CPA, like any other campaign.
- View-through separately, with a 1-day window, labelled as "estimated contribution".
- Whatever incrementality evidence you have (holdout, geo or total), with its result as a percentage: "remarketing contributes 12% of conversions that would not have happened".
- The decision: keep, scale or cut based on the cost per incremental conversion.
Honest remarketing usually comes out like this: excellent CPA per click, reasonable incremental CPA, view-through irrelevant to the decision. And that is enough to defend it — better than an inflated column nobody believes.
What isn't measured but exists
Awareness, brand searches rising after a video campaign, Search CTR improving because people have already seen you: real effects, hard to attribute. You look at them with time series (brand searches, direct traffic) and geo experiments, not with conversion columns. Acknowledging them without inventing numbers is part of being honest.
💡 Ninja trick: half of the Display "performance" that evaporates once you measure properly is not down to attribution: it is traffic that was neither human nor interested (farms, apps, accidental clicks). Cleaning up placements with Ninja Shield usually improves Display's CPA per click more than any bid adjustment — because you stop paying for what was never going to convert.
What you should remember
- View-through separately and at 1 day; never use "all conversions" to decide.
- The question is incrementality: holdout > geo experiment > account total.
- Present Display with CPA per click + incremental CPA, and view-through as an estimated contribution.
- Awareness effects are real; you read them in time series, not in columns.
End of Module 7. Module 8, Measurement II, closes the biggest gap in measurement: getting what happens after the form back to Google.