Reading paid media results without fooling yourself

An advertising platform reports on its own performance using its own definitions, within its own view of the customer journey. That is not a conspiracy; it is a structural limitation. Anyone making budget decisions from those reports needs to know where the numbers bend.

Attribution inflation

If you sum the conversions claimed by every channel, the total will usually exceed the conversions your finance system recorded. Each platform counts a conversion it touched, and several platforms touched the same one. Retargeting and brand search are the usual beneficiaries: they appear late in journeys that were already going to complete.

The practical response is not to abandon platform data but to hold a second reference. Compare the platform's claim to your own back-office numbers over the same period, and watch the relationship between total spend and total revenue rather than treating each channel's self-report as independent truth.

Survivorship in the creative report

Automated delivery concentrates impressions on the assets that perform early. By the end of a flight, one asset may hold most of the spend, and it will look like the best creative. Often it simply received the most opportunity. Testing creative properly means controlling delivery, or at least being honest that a comparison of unequal impressions is not a comparison.

If you sum the conversions every channel claims, the total will exceed what finance recorded.

Sample sizes that cannot support the conclusion

A campaign with forty conversions cannot reliably tell you that one audience outperforms another by fifteen per cent. The difference between two small numbers is mostly noise, and acting on it produces changes that feel decisive and are effectively random.

• Decide before launch how many conversions a comparison needs before it is worth reading.
• Resist restructuring accounts on a week of data; weekly variance in small accounts is large.
• When volume is genuinely low, prefer bigger, structural changes over incremental tests.

Seasonality mistaken for skill

Performance that improves in a strong season and declines in a weak one gets attributed to whatever was changed most recently. Year-on-year comparison, or comparison against an untouched control campaign, separates the calendar from the work.

What honest reporting looks like

Good paid media reporting states the objective agreed in advance, the result against it, the confidence in that result, what was changed during the period and what will be changed next. It shows the disagreement between platform and finance numbers rather than quietly picking the more flattering one.

Where volume allows, geo holdouts or scheduled pauses give a genuine read on incrementality: what happens to total demand when the spend stops. It is the least comfortable measurement available and usually the most informative.

None of this makes paid media less useful. It makes the decisions built on it more durable, which matters most when you are deciding whether to scale.