It stopped being an experiment
Retail media in South Africa has crossed the threshold from test-and-learn line item to a genuine top-five channel by spend for several FMCG advertisers we work with. The growth has been fast enough that planning practice has not caught up, which means a lot of money is moving on the strength of a sales narrative rather than a measurement case.
The proposition is genuinely strong. These networks sit on first-party transaction data at the point of purchase decision, in categories where the purchase decision is made in-aisle or in-basket. That is a real advantage and it deserves real money.
What we tested
We ran comparable activity across six networks in four categories — household care, beverages, personal care and packaged food — and scored each network on four dimensions: data granularity, incrementality evidence, reporting independence and inventory transparency.
The spread was much wider than the market conversation suggests. Two networks provided log-level or near-log-level data that let us build our own incrementality read. Three provided aggregated reporting only, which means you are accepting the seller's own measurement of the seller's own performance. One could not tell us with confidence which placements a given spend had actually run against.
The uncomfortable finding
The networks with the best-performing dashboards were not the networks with the best-performing media. There is an inverse relationship in our sample between the polish of the reporting interface and the willingness to expose the underlying data. That should worry anyone planning this channel on reported ROAS.
Where we could measure incrementality independently, the honest returns were good but not extraordinary — clearly worth the investment, clearly not the multiples quoted in category sales material. Where we could not measure independently, reported returns were dramatically higher. Draw your own conclusion.
How to buy it properly
Make data access a condition of the buy, not a nice-to-have in the annual review. Insist on a holdout or a geo-split for anything above a threshold you set. Treat the network's reported ROAS as a directional signal and never as a planning input.
And plan it as a conversion channel that also does light brand work, not as a substitute for the reach layer. Retail media is closest to the transaction, which is exactly why it cannot build the demand it harvests.