On 14 November 2025, Kiri Masters published a column in The Drum naming the asset retail media networks will defend hardest through the AI-shopping reset. Not audiences. Not context. Loyalty data.
She is right, and the implication is sharper than most RMN decks admit.
The argument runs like this. Agentic shoppers, generative search, and AI-mediated discovery are rewriting the journey from query to basket. The signals that media buyers have leaned on for a decade — third-party audiences, contextual proxies, even most modelled segments — degrade fastest in that environment. What survives is the deterministic record of who actually bought what, when, and how often. That record sits inside loyalty programmes.
The Wise Marketer's state-of-RMNs review, published in the same window, reaches the same conclusion from the operator side. The networks pulling ahead are the ones with dense, consented, basket-level data tied to a logged-in identity. Tesco's Clubcard-powered Dunnhumby business, Kroger's 84.51°, Albertsons Media Collective, Sainsbury's Nectar360. The pattern is consistent. Loyalty is the moat.
For buyers, this is a repricing event. If loyalty data is the signal that holds up across an AI-shopping cycle, then media bought against it should command a premium and media bought against decaying proxies should not. Most plans still treat the two as interchangeable line items. They are not.
The second-order effect is more interesting. Loyalty data only earns its premium when it is activated against incrementality, not ROAS. IAB Europe's commerce media committee made that explicit in its November Q&A and the V2.1 standards that followed. A loyalty-matched audience served a discount on a product the shopper was already buying is not performance. It is subsidy. The combination that matters is deterministic identity plus an incremental read. One without the other is theatre.
This is where the practitioner work sits. Three pieces have to lock together.
First, an inventory view of which RMNs actually expose loyalty signal at the bid layer versus which ones bolt it on post-campaign as a measurement story. The gap is wide and the rate cards do not reflect it. Tesco Media and Sainsbury's Nectar360 sell against logged-in basket data at activation. Several others sell modelled lookalikes and call it the same thing. It is not.
Second, a creative operation that can metabolise the signal. Loyalty data is only as useful as the number of meaningful variants a brand can ship against it. Frequent-buyer, lapsed, category-adjacent, basket-complement. Each is a distinct creative brief. Most brands ship two. Persaic exists because that arithmetic is the bottleneck, not the targeting.
Third, a measurement contract written in incrementality. Geo-holdouts, matched-market tests, and clean-room lift studies are the table stakes now. If an RMN cannot return an incremental read against a loyalty-matched cohort, the loyalty premium evaporates. Buyers should walk.
The reason this matters beyond grocery is that the loyalty playbook is migrating. Boots, Uniqlo, IKEA, Decathlon, the airlines, the QSR apps. Anyone with a logged-in repeat customer base is now in scope for an RMN conversation. The supply side of loyalty-grade signal is about to widen considerably, and the buyers who have built the activation and measurement muscle will price it correctly. The ones who have not will overpay for the next three quarters.
There is a quieter point underneath all of this. Loyalty data is defensible precisely because it is consented, first-party, and tied to a real transaction. That is the same property that makes it durable through whatever the cookie, identifier, and agentic-browser landscape does next. The signals that survive an AI-shopping reset are the signals that were never borrowed in the first place.
The work for the next planning cycle is unglamorous. Audit which RMN partners expose loyalty signal at the bid, not in the deck. Rewrite the measurement clause to incrementality. Staff the creative operation to ship the variant count the signal can carry. Reprice the line items accordingly.
Audience and contextual buys will keep a seat at the table. They should. But the layer that holds up when the shopper stops typing and starts delegating is the one written in basket data. Build the stack around it.