Incrementality is a measurement model problem

Marketing incrementality has been the holy grail for as long as marketing has existed. The toolkit is better than ever: data-driven attribution, MMM, newer agile and open-source MMM approaches like Meridian and Robyn, platform lift studies, geo-based experiments. Big Tech had already been stitching together cookies, session IDs, user IDs, CRM records and transaction histories for years in an attempt to approximate the truth.

This becomes materially easier when you own most of the commercial journey yourself. If the bulk of demand happens in your own eCommerce and retail, you can arguably get a true picture of incrementality. However, many brands operate across eCommerce and retail, marketplaces, online and offline wholesale, social commerce and, increasingly, agentic commerce.

The good old question keeps popping up: what is driving demand, and what is merely capturing it? The persistent answer has been: test it. But how exactly do you test well when you do not have a holistic view of commercial outcomes? Marketplace visibility is partial. Wholesale lags and often arrives late and aggregated. Social commerce can collapse media and transaction into one platform, and agentic commerce may compress discovery, comparison and purchase.

In that world, someone can and will challenge the result of your test, however statistically significant, because the commercial system underneath is still only partially visible.

And then there is the organisational tension underneath all of this. On one side sits the performance marketer, still judged on ROAS. On the other sits the CMO, often rightly convinced that some of today's budget should be redirected towards demand creation. Both are reacting rationally to the system they are measured in. The problem front and centre is the measurement model, and you have to change it.

Start by defining the true north star KPI. Not the steering KPI, not the in-platform efficiency metric. Usually that is some version of profitable revenue over time, contribution margin, CLTV to CAC, repeat behaviour and the quality of customer acquisition. Then separate those from the steering KPIs that help teams run channels day to day. ROAS has its place, but it is not the truth. Useful, often necessary, but dangerous when mistaken for the objective itself.

Once that is clear, work backwards. Align leadership on the outputs that actually matter. Get them signed off high enough in the organisation that they become the operating truth rather than an analytical side note. Then build the best proxies possible for the channels you cannot fully see. Use partner sell-out where available. Use finance data where that is all you have. Reverse engineer where you must and triangulate where you can.

Until that prework is done, I would be cautious about anyone claiming they have solved incrementality. Unless you have taken a true bird's-eye view across commercial channels and outputs, somebody somewhere will always be able to challenge the conclusion.