Marketing mix modelling is back at the centre of the CFO conversation- twenty10Multi-touch attribution alone can't survive cookie deprecation: triangulation is the new default- twenty10Geo lift tests and incrementality studies are calibrating the next generation of MMM- twenty10Clean rooms + MMM: how retailers and CPGs are closing the measurement gap- twenty10The UK MMM landscape: who's serious, who's selling a dashboard- twenty10Comparing marketing mix modelling providers in the UK- twenty10Brand vs performance: measuring long-term equity effects in a P&L your CFO can read- twenty10Marketing mix modelling is back at the centre of the CFO conversation- twenty10Multi-touch attribution alone can't survive cookie deprecation: triangulation is the new default- twenty10Geo lift tests and incrementality studies are calibrating the next generation of MMM- twenty10Clean rooms + MMM: how retailers and CPGs are closing the measurement gap- twenty10The UK MMM landscape: who's serious, who's selling a dashboard- twenty10Comparing marketing mix modelling providers in the UK- twenty10Brand vs performance: measuring long-term equity effects in a P&L your CFO can read- twenty10
All insightsAI & Measurement

Measuring the Impact of AI on Marketing Performance

How to measure the real, incremental impact of AI in marketing: on creative, targeting and mix, without falling for platform-reported uplift or vendor case studies.

twenty10··6 min read

Every marketing team is being asked the same question this year: what is AI actually doing for us? The honest answer, for most, is: we do not know. The dashboards say something is working, the vendors say a lot is working, and the P&L says less than either.

That is not a failure of AI. It is a failure of measurement. The tools most marketers use to score performance were never built to isolate the effect of a single change to creative, bidding, or audience. Layering AI on top of them just gives the same tools a new number to over-attribute.

Why platform-reported AI uplift is misleading

Every major platform now ships an AI-driven product: Performance Max, Advantage+, Smart Bidding, Generative Creative. Each reports its own uplift, measured against a counterfactual the platform itself defines. None of them see each other. None of them model demand that would have converted anyway. And the "control" is often a lower-quality version of the same platform's own targeting, which flatters the AI variant by design.

The result is a pile of double-counted, self-scored wins that do not reconcile to revenue. When finance asks whether AI is paying back, no one can produce an answer that ties.

What actually works

Three tools, used together, produce a defensible read on AI's marketing impact:

  • Geo experiments. Split markets into treatment and control. Turn AI creative or AI bidding on in one, off in the other, and measure the delta on total revenue: not platform-reported conversions. Two to four weeks is usually enough for a paid social test, six to eight for TV or CTV.
  • Incrementality tests. Ghost bids, holdout audiences and PSA tests measure the counterfactual directly. They are the only way to know whether an AI-optimised campaign is winning incremental customers or just harvesting demand that would have converted anyway.
  • Marketing Mix Modelling. MMM will not tell you whether Advantage+ is better than manual: the granularity is wrong. But it will tell you whether the channel's overall ROI has moved since AI was rolled in, which is the number that matters to the CFO.

The right sequence

Run the experiment first. Get a clean, causal read on the AI variant. Feed the result into the MMM as a prior. Then let the MMM tell you what the new channel-level economics imply for the plan. That is the loop: experiment, model, allocate.

Doing it the other way round: trusting platform-reported uplift, then explaining afterwards why the P&L did not move: is how marketing loses the AI argument in the boardroom.

What to report

Finance does not need a slide on the AI. It needs three numbers:

  1. Incremental revenue attributable to the AI variant, with a confidence interval, from a real experiment.
  2. Change in channel ROI since the AI rollout, from MMM.
  3. Payback period on the licence cost and the incremental spend the AI directed.

If a vendor cannot produce those three, they are selling a dashboard, not measurement. And AI without measurement is just a faster way to spend money.

More in AI & Measurement