What happened

On September 9, ClickZ[1] published a report from a discussion in London. A Wolt representative described a Unilever campaign that combined the brand's first-party data with platform audiences in a secure data-sharing environment. According to the representative, the new strategy increased advertising-attributed revenue by just over 30%; it is being extended to 15 additional markets. [1 · ClickZ]

What is known about effectiveness

The metric concerns attribution of sales to advertising, not a proven increase beyond purchases that would have happened without it. The article does not provide the campaign period, sample size, comparison rules or control group. [1 · ClickZ]

How to assess the result

Before comparing campaigns, a consistent sales attribution window should be set and double counting of the same order across channels excluded. The causal effect can be tested using a randomly selected group that does not receive the new advertising or comparable geographic areas. It is also useful to measure the share of genuinely new buyers and their repeat orders. This helps clarify whether combining data expands relationships with an audience or simply finds people already ready to buy more effectively. [1 · ClickZ]

Expert commentary

I assess the Wolt and Unilever case primarily as a change in how a brand and a commerce platform coordinate. The brand's first-party data can add information absent from the platform's order history and make audience selection more meaningful. However, an improvement in advertising-attributed revenue can reflect either new purchases or more accurate identification of existing demand. These are different sources of value, requiring different management decisions and advertising payment methods. [1 · ClickZ]

Access to compatible data and the ability to repeat measurement across markets matter for competition. If the incremental effect is confirmed, partnerships in which the brand understands audience needs and the platform sees purchases will be stronger. Large manufacturers may have more initial data and integration resources, however. Without accessible onboarding terms, the new model could widen the gap with small brands and strengthen the platform's bargaining position as an intermediary in customer relationships. [1 · ClickZ]

Scientifically, combining datasets does not remove the causality problem. Gordon, Moakler and Zettelmeyer studied hundreds of advertising experiments and showed that even rich data did not allow observational methods to reliably reproduce experimental effects. The implication for this case is specific: the new audience-selection method should be compared with the old one under identical order-attribution rules. Otherwise, an improved predictive model can easily be mistaken for proof that advertising influenced a person's decision. [2 · Gordon, Moakler, Zettelmeyer, 2022]

For customers and society, secure processing is useful only alongside understandable control over how information is used. Tucker's study links changes in responses to personalized advertising with stronger perceived privacy control; it studied a social network, not Wolt. My hypothesis is that a transparent explanation of the purpose of combining data will help preserve trust. The phrase 'secure environment' alone does not show how well customers understand what is happening or whether they can influence it. [3 · Tucker, 2014]

In relationship marketing, I would distinguish loyalty to Unilever from the habit of ordering through Wolt. A buyer may return for the platform's convenience and choose a different brand each time. Success for a joint campaign should therefore be tied to repeat selection of a particular brand, purchase satisfaction and the absence of intrusive contacts. If personalization merely captures a buyer who is already ready to purchase, it may improve the advertising channel's report without strengthening the brand's own relationship with the person. [1 · ClickZ]

I would treat expansion into new markets as a series of tests, since audience composition and data availability may differ. The necessary measures include additional purchases relative to a control group, margin after advertising, the share of customers new to the brand and their repeat orders. Sales across all available channels should be considered: growth within Wolt may accompany a shift of purchases from another store. Scaling is justified when total value increases, rather than simply changing where it is recorded. [2 · Gordon, Moakler, Zettelmeyer, 2022]

Sources

  1. ClickZ — Wolt and Unilever case study — September 9, 2026. Report by the moderator of the Retail Media Pioneers panel.
  2. Gordon, Moakler, Zettelmeyer, 2022 — limitations of observational advertising measurement — Primary research, author version on arXiv; it did not test Wolt.
  3. Tucker, 2014 — personalization and privacy control — Primary research on advertising response; its application to Wolt is a hypothesis.