300° · Measurement
What did the advertising actually cause?
It is the only question that matters and the one most reports avoid. We treat measurement as its own practice, with the standing to tell the other five sectors they are wrong.
What we build
Three layers, each checking the one above.
- CollectionAnalytics configuration, tag management, server-side event delivery and consent handling. Events are named once, documented, and tested before launch.
- AttributionPlatform and analytics attribution used for what it is good at: fast, directional signals for day-to-day optimisation.
- IncrementalityHoldout groups, geographic experiments and platform lift studies that measure the difference advertising made against a group that did not see it.
- ModellingFor larger budgets, media mix modelling to estimate channel contribution over time, calibrated against experiment results.
- ReportingOne dashboard built from your order or CRM data outward, so the totals match what finance already believes.
- Data collaborationPrivacy-safe audience matching and measurement in clean-room environments, where a partner offers one.
Anatomy of a lift test
The closest marketing gets to a controlled trial.
The principle is older than the internet: change one thing for one group, leave another group alone, compare.
- QuestionOne decision the result will change. If nothing would change, do not run it.
- SplitRandomly assign people or regions to exposed and holdout groups.
- RunLong enough to cover the purchase cycle. No peeking and re-cutting.
- CompareDifference in outcome between groups, with the uncertainty stated.
- DecideScale, cut or re-test. The decision is written down beside the result.
Reading the instruments
Three tools, three kinds of answer.
None is "the truth". Used together they bracket it.
Attribution
Good for: comparing ads and audiences within a channel, day to day.
Blind to: anything it cannot track, and whether the sale would have happened anyway.
Experiments
Good for: a defensible answer about one channel or tactic at one point in time.
Blind to: everything you did not test. They also cost reach while they run.
Mix modelling
Good for: the whole budget, including offline and untrackable media, over long periods.
Blind to: fine detail. It needs history, and it is only as good as its assumptions.
Straight answers
Questions about measurement.
Will the numbers match our ad platforms?
No, and they should not. Platforms each claim the conversions they touched, so their totals overlap. Our reporting starts from your own records and allocates from there. Expect the platform figures to be higher.
Are we big enough to run incrementality tests?
Many simple tests need less volume than people assume, email holdouts especially. Geo experiments and modelling need more. We will tell you what your volume supports and what it does not.
How do you handle privacy and consent?
Collection respects the choices people make in your consent banner and the laws that apply where they live. We prefer aggregate and modelled methods to anything that depends on identifying individuals.
Can you measure work run by another agency?
Yes. Independent measurement is one of our audit engagements, and it is often more persuasive coming from someone who did not buy the media.
Bring us the number
you actually answer for.
Tell us what you sell, who buys it and what the business needs from media this year. We reply with a point of view, not a deck.
Start a conversation