Retail media is not slowing down, but the layer that has been selling it is. The pattern in the last four weeks of second-quarter results points to a specific outcome: the bundled managed-service layer of retail media likely keeps contracting through the first quarter of 2027, while budgets migrate to retailer-controlled self-serve platforms and to licensed ad infrastructure. Expect at least two more top-100 US or European retailers to disclose moving onsite monetisation off a bundled managed-service network before the end of Q1 2027, and expect Criteo’s Retail Media contribution ex-TAC to remain below the prior-year level when it reports fourth-quarter numbers in February 2027.
That is a narrower claim than the one the market appeared to price in early August, when one set of ad tech results was read as evidence that retail media itself had rolled over. The underlying data does not support that reading. It supports something more specific and, for retailers, more consequential.
In short
- The prediction: the managed-service tier of retail media likely keeps shrinking through Q1 2027, with at least two more top-100 US or European retailers disclosing a move off a bundled managed-service network toward in-house tooling or licensed ad infrastructure.
- Signal 1: Criteo’s Q2 2026 print showed Retail Media revenue of $47.9m, down 22% at constant currency, driven by two clients reducing scope (a $75m full-year impact), while contribution ex-TAC excluding those two accounts grew roughly 20%.
- Signal 2: Perion reported on 10 August 2026 that Retail Media vertical spend rose 60% year over year, alongside a new in-store retail media technology mandate from Best Buy Canada.
- Signal 3: Fluent reported the same day that Commerce Media Solutions revenue grew 90% to $30.5m, or 63% of consolidated revenue, with an annual run rate above $125m.
- The counter-signal: Criteo’s same-retailer retention was 113% excluding its largest client, and it added Lidl to its onsite roster in May 2026, so the intermediary tier is repricing rather than disappearing.
Why this matters now
Retail media has spent five years being described as a single market. It is not. It is at least three markets stacked on top of each other, and they are now moving in opposite directions.
The first is demand: brand and agency money chasing shopper data at the point of purchase. That is still growing, and the second-quarter numbers say so. The second is infrastructure: the ad server, the auction, the measurement layer, the identity plumbing. The third is service: the humans who sell the inventory, build the campaign, and manage the account on the retailer’s behalf.
For most of the last decade those three were sold as one product, because most retailers could not staff any of them. A network partner arrived with technology, demand, and a sales team, took a share of the revenue, and the retailer booked high-margin income it had not previously had. The economics were attractive enough that few retailers examined the split closely, a dynamic that shows up in marketplace advertising too, where the true cost of promoted placement is often buried inside a blended commission line rather than priced separately.
What is changing is that the largest retailers now have the scale to unbundle. They have engineering teams, first-party data infrastructure, and enough advertiser demand arriving unprompted that the sales layer is no longer the scarce input. When that happens, the managed-service margin becomes the most expensive thing on the invoice, and the first thing to be cut.
Signal 1: Criteo’s retail media line splits in two
Criteo’s second-quarter 2026 results, published in early August, are the clearest single data point available. Group revenue came in at $428m, down 11% year over year, with contribution ex-TAC of $255m, down 12% at constant currency. Media spend on the platform rose 9% at constant currency to about $1.1bn, which is itself worth pausing on: more money flowed through, less of it stayed.
Inside that, Retail Media revenue was $47.9m, down 22% at constant currency, with segment contribution ex-TAC of $47.2m against $60.0m a year earlier. The company attributed the decline to two Retail Media clients reducing the scope of their engagements, with a full-year 2026 impact of roughly $75m, phased as approximately $27m in Q1, $21m in Q2, $20m in Q3 and $7m in Q4.
The detail that matters is what the largest of those clients actually did. Per the disclosure, first flagged in May 2025, it discontinued managed services and curtailed brand demand sales. It did not stop running retail media. It stopped buying the service wrapper and the outsourced sell-side function.
Two further figures make the split explicit. Same-retailer retention was 84% including the largest client, and 113% excluding it. And excluding the two affected accounts, Retail Media contribution ex-TAC grew roughly 20% year over year. The demand base is expanding; the bundled service relationship at the top of the client list is contracting.
| Metric | Reported | What it measures |
|---|---|---|
| Retail Media revenue | $47.9m, down 22% cc | Headline segment contraction |
| Retail Media contribution ex-TAC | $47.2m vs $60.0m | Economics after traffic acquisition cost |
| Same-retailer retention (incl. largest client) | 84% | Effect of one account unbundling |
| Same-retailer retention (excl. largest client) | 113% | Underlying expansion in the rest of the book |
| Contribution ex-TAC excl. two affected clients | Approximately +20% | Growth of the non-unbundled base |
| Platform media spend | Approximately $1.1bn, up 9% cc | Volume still rising through the pipes |
The guidance path over 2026 tells its own story. February guidance pointed to contribution ex-TAC of flat to up 2%. May guidance moved to a low-single-digit decline. August guidance moved again, to a decline of 10–12% at constant currency, with management assuming no recovery in large-client spending. Guidance that steps down three times in six months is usually describing a structural change, not a soft quarter.
Management was careful to separate the two effects. The chief financial officer attributed the quarterly shortfall primarily to Performance Media dynamics rather than the retail client headwinds, which had been forecast and were delivered within expectations. Performance Media revenue of $380.1m was down 9% at constant currency, with fashion media spend down 21%. The chief executive described several large enterprise clients further reducing spending, driven by client-specific decisions and softer demand in specific verticals.
That is an honest split, and it cuts both ways. It means the retail media contraction was known and planned for. It also means the retail media contraction is not a cyclical artefact that a better ad market fixes.
Two adjacent moves belong in the same signal. The company expanded its GO platform to full self-service access in March 2026, and reported that more than half of its small clients globally have adopted it, faster than anticipated. And it named its chief strategy officer as chief financial officer effective 10 August 2026, with the outgoing finance chief departing at the end of September. A strategy-to-finance handoff during a guidance reset generally signals a repositioning rather than a cost exercise.
Signal 2: Perion’s retail media spend grows 60% in the same quarter
On 10 August 2026, Perion Network reported second-quarter results that ran in the opposite direction on the same market. Spend through its Perion One platform rose 15% year over year to $156.7m, with retail media vertical spend up 60%, connected TV up 56%, digital out-of-home up 45%, and its AI agent product up 136%.
Group revenue of $98.2m and contribution ex-TAC of $42.3m at a 43% margin describe a company that is smaller than Criteo and structurally different. Perion sells into the retail media opportunity as technology and channel access, not as an outsourced commercial function for a retailer’s ad business.
The most instructive line in that release is not a number. Perion cited Best Buy Canada selecting it as the in-store retail media digital out-of-home technology partner. That is a retailer buying a defined technical capability for a defined surface, on its own network, rather than handing a whole monetisation stream to a network partner.
This is what a signal looks like when it is independent rather than corroborating. Perion is a separate company with separate clients and a separate filing. It is not reporting the same event from a different angle. It is reporting that the money Criteo lost at the service layer did not leave retail media.
Signal 3: Fluent’s commerce media becomes the whole company
Also on 10 August 2026, Fluent reported that Commerce Media Solutions revenue grew 90% year over year to $30.5m, representing 63% of consolidated revenue, against $16.1m and 36% in the same quarter of 2025. The annual revenue run rate for that business now exceeds $125m, at a 27% gross margin. Consolidated revenue was $48.4m, up 8%, with revenue from continuing businesses in aggregate up 25%.
The chief executive described the quarter as an inflection point the company had been building toward. Read alongside the margin figure, the more interesting characteristic is the shape: a 27% gross margin is an infrastructure and placement business, not a managed-service business. The company is growing by supplying commerce media capability into other people’s transaction flows.
Three companies reporting within roughly a week of each other, filing separately, describing the same market from three positions, is the minimum bar for treating a pattern as real rather than as one company’s execution problem.
| Signal | Source and date | Direction | What it is evidence of |
|---|---|---|---|
| Retail Media revenue down 22% cc; two clients cut scope; largest discontinued managed services | Criteo Q2 2026 results, early August 2026 | Down at the service layer, up approximately 20% excluding affected accounts | Large retailers shedding the bundled service wrapper, not the channel |
| Retail media vertical spend up 60%; Best Buy Canada in-store DOOH mandate | Perion Q2 2026 release, 10 August 2026 | Up | Demand routing to technology-only vendors for defined surfaces |
| Commerce Media Solutions up 90% to $30.5m, 63% of revenue, run rate above $125m | Fluent Q2 2026 release, 10 August 2026 | Up | Infrastructure-shaped commerce media scaling at infrastructure-shaped margins |
What the pattern suggests
Put the three together and a fairly precise mechanism emerges. Retail media budgets are not contracting. The share of those budgets captured by a bundled intermediary is contracting, and it is contracting fastest at the largest retailers, because they are the only ones who can currently replace the function.
The sequencing appears consistent. A retailer starts with a full-service network partner. Advertiser demand becomes inbound rather than solicited. The retailer hires a commercial team, then a product team, then either licenses an ad server or builds one. The network partner keeps the long tail of advertisers and the international demand it can uniquely source, and loses the managed-service margin on the top accounts.
If that sequence is right, the next disclosures should come from the tier immediately below the largest retailers: national chains with meaningful digital traffic, an existing media business in the tens of millions of dollars, and a technology organisation already building customer-facing systems. That is the same cohort now building selective, invitation-only marketplaces rather than open ones, and for a similar reason: at a certain scale, control of the surface becomes worth more than the speed of outsourcing it.
The falsifiable version of the prediction is therefore this. Before the end of Q1 2027, at least two more top-100 US or European retailers likely disclose (through a vendor announcement, an earnings call, or trade press confirmation) that onsite retail media monetisation is moving off a bundled managed-service arrangement. A future observer can check that against vendor press releases and retailer disclosures.
The precedent chain: Kroger, Kohl’s and the PromoteIQ sunset
This pattern has a documented history, which is what separates a prediction from a guess. The precedents are consistent in direction and have been accelerating in frequency.
Kroger Precision Marketing announced in mid-2023 that it was taking its self-serve retail media ad technology in-house, moving off a platform previously powered by Microsoft Advertising. The initial build covered product listing ads and onsite display on owned properties, with beta running and general availability targeted for the fourth quarter of that year. Reporting at the time noted the unit had added close to 100 roles across engineering, product, data science and operations to support it.
That staffing number is the honest cost of in-housing, and it is the reason this pattern moves down the retailer size curve slowly rather than all at once. Kroger’s broader commercial ambitions have continued to widen since, including a marketplace push signalled by its e-commerce leadership, which is a related expression of the same instinct to own the surface.
Kohl’s took the intermediate route rather than the full build. It moved its media network’s onsite advertising from PromoteIQ to Koddi, with accounts and campaigns transitioning on 3 August 2025, following what the retailer described to advertisers as a thorough review and evaluation of the marketplace. Microsoft has been winding down the PromoteIQ onsite offering, which converted a vendor choice into a forced migration for several retailers.
| Retailer | Approximate date | Move | Route taken |
|---|---|---|---|
| Kroger Precision Marketing | Mid-2023 | Off Microsoft-powered platform to own self-serve stack | Full in-house build, roughly 100 roles added |
| Kohl’s Media Network | August 2025 | PromoteIQ to Koddi for onsite ads | Licensed ad infrastructure, retailer-controlled |
| Criteo’s largest retail media client | Disclosed May 2025, impact through 2026 | Discontinued managed services, curtailed brand demand sales | Retained platform relationship, removed service layer |
| Best Buy Canada | Cited August 2026 | Selected a technology partner for in-store retail media DOOH | Capability purchase for a defined surface |
| Lidl | May 2026 | Expanded onsite retail media with a third-party network | Counter-example: chose the bundled route |
Three of those five moved control toward the retailer. One removed only the service layer while keeping the platform. One went the other way. That ratio is roughly what the prediction assumes: a strong directional bias, not a one-way market.
Wider context: who pays for the ad server
The economics underneath this are not complicated. A managed-service retail media arrangement typically bundles technology, demand access and commercial operations into a revenue share. Once a retailer’s advertiser demand becomes largely inbound, the demand-access component loses value quickly, and the retailer is paying a variable percentage for what has become a fixed-cost software and staffing problem.
Licensed infrastructure inverts that. The retailer pays a platform fee that scales far more slowly than revenue, hires its own commercial team, and keeps the incremental margin. The break-even point sits somewhere in the low tens of millions of dollars of annual media revenue, which is why the pattern started at the very top and is working downward.
Capital is following the same logic. Investment activity in the infrastructure tier, including a strategic investment by one commerce media platform into a competing auction and sponsored-ads vendor, indicates that the durable position is thought to be the ad server rather than the sales desk. Meanwhile the largest owned-and-operated networks continue to demonstrate the ceiling: Walmart’s global advertising business reached roughly $6.4bn in its FY2026 with growth in the mid-forties percent, entirely on infrastructure it controls.
There is a regulatory dimension arriving in parallel. As sponsored placement moves closer to algorithmic recommendation, the question of what must be disclosed to a shopper becomes sharper, and the deception exposure around undisclosed commercial steering is easier to manage on a stack the retailer controls than on one it rents. Compliance is quietly becoming an argument for in-housing.
Implications for retailers, brands and investors
For retailers, the decision is no longer build versus buy. It is a three-way choice between full in-house, licensed infrastructure with an internal commercial team, and continued managed service, and the right answer depends almost entirely on annual media revenue and existing engineering capacity. Below roughly $20m of annual media revenue, the managed route likely remains rational; above it, the maths tightens each year.
The trap is partial commitment. Retailers that license infrastructure without funding a commercial team tend to see fill rates fall, because inbound demand covers endemic advertisers and little else. The Kroger precedent suggests the staffing requirement is measured in dozens of roles, not a handful.
For brands and agencies, the practical consequence is fragmentation. Each retailer that leaves a bundled network becomes another separate buying interface, another taxonomy, another reporting standard. Planning teams should expect the number of distinct retail media integrations to rise through 2027, and should budget for the operational cost of that rather than assuming consolidation will arrive to rescue them.
It also raises the value of anything that standardises access across surfaces. The same pressure is visible in agentic shopping, where the durable pattern so far has been that retailers insist on controlling the checkout surface even when they open discovery to third parties. Retail media is following the same instinct one layer up.
For investors, the read-through is that segment labels are now misleading. A company reporting retail media revenue may be selling service, infrastructure, or demand, and those three have materially different growth rates, margins and durability. The relevant question at the next round of results is not whether retail media grew, but which layer the revenue sits in and whether the largest client can replace it internally.
Caveats: what could go wrong
The strongest counter-argument comes from the same filing that anchors the prediction. Criteo’s same-retailer retention was 113% excluding its largest client, and retail media contribution ex-TAC excluding the two affected accounts grew roughly 20%. If the intermediary layer were structurally failing, the rest of the book would not be expanding at that rate. A reasonable alternative reading is that one very large customer made an idiosyncratic decision and the market has over-generalised from it.
Second, the incumbent is repositioning rather than standing still. More than half of the company’s small clients globally have adopted its self-service platform, faster than management anticipated. If the bundled network successfully reconstitutes itself as licensed infrastructure plus a self-serve interface, the layer survives with a different revenue shape, and the prediction is technically right about behaviour but wrong about who captures the value.
Third, there is a live counter-example. Criteo added Lidl to its onsite retail media roster in May 2026, which is a large European grocer choosing the third-party route at exactly the moment the in-housing thesis was gathering evidence. European retailers may follow a different path from US ones, given lower average media revenue per retailer and tighter data-protection constraints on building first-party ad stacks.
Fourth, the cyclical explanation has real support. The finance chief explicitly attributed the quarterly shortfall to Performance Media rather than to retail media, with fashion media spend down 21% and several large enterprise clients cutting budgets. If the broader ad market recovers into 2027, softer numbers across the sector may be revealed as cyclical, and retailers may find the case for expensive in-housing projects harder to fund.
Fifth, timing risk is genuine. In-housing projects take four to eight quarters from decision to general availability, and they are announced inconsistently. Two disclosures before the end of Q1 2027 is a demanding bar, not because the moves will not happen, but because retailers frequently make them quietly and only confirm them when a vendor issues a press release.
Finally, an exogenous shift could reorder the question entirely. If a material share of product discovery migrates to AI assistants and agentic surfaces, the value of onsite sponsored placement changes shape, and the argument over who operates the onsite ad server becomes less important than the argument over who is visible upstream of it.
| Scenario | What happens | Early tell | Assessment |
|---|---|---|---|
| Base case: continued unbundling | Two or more top-100 retailers disclose moves off bundled managed service; incumbent retail media revenue stays negative year over year through the Q4 2026 report | Vendor win announcements from infrastructure players in Q4 2026 | Most consistent with the three signals |
| Repositioning case | The incumbent converts the relationship to licensed self-serve; revenue mix shifts, headline decline moderates by Q1 2027 | Self-serve adoption disclosed as a share of revenue rather than client count | Plausible; would confirm behaviour, not the loser |
| Cyclical case | Ad demand recovers, enterprise budgets return, in-housing projects are deferred on cost grounds | Performance Media returning to growth before retail media does | Possible; would weaken but not void the thesis |
| Reordering case | Agentic and AI-assistant discovery pulls attention upstream, reducing the strategic weight of onsite placement | Retailers reporting agent-referred traffic as a named channel | Slower-moving; unlikely to resolve within the window |
Primary filings for the companies discussed are available through the SEC’s electronic disclosure system, including the quarterly reports and results exhibits cited above (EDGAR entity page).
FAQ
Is retail media growth actually slowing?
The evidence points to a mix shift rather than a slowdown. Media spend flowing through one major platform rose about 9% at constant currency in Q2 2026 even as its retail media revenue fell 22%, and two separate companies reported retail and commerce media growth of 60% and 90% in the same period. The contraction is concentrated in the bundled service layer.
What exactly does “unbundling” mean here?
It means separating the three things a retail media network historically sold together: the advertising technology, access to advertiser demand, and the commercial team that sells and runs campaigns. Retailers are increasingly buying the first, sourcing the second directly, and staffing the third internally.
Which retailers are most likely to move next?
The pattern suggests national chains one tier below the largest players: those with meaningful digital traffic, existing media revenue in the tens of millions of dollars, and an in-house technology organisation already shipping customer-facing systems. Grocery, home improvement and specialty apparel chains fit that profile most closely.
Could this just be one company’s execution problem?
That is the most serious counter-argument. Same-retailer retention of 113% excluding the largest client, and roughly 20% growth in contribution ex-TAC excluding the two affected accounts, both suggest the rest of the book is healthy. The reason to treat it as structural is that two unrelated companies reported the opposite trend in the same window, which is difficult to explain as a single execution failure.
Does in-housing actually save money?
Not immediately, and not below a certain scale. In-housing converts a variable revenue share into fixed software and headcount cost, so it pays off only above a break-even that appears to sit somewhere in the low tens of millions of dollars of annual media revenue. One documented precedent involved adding close to 100 roles to support the build.
What does this mean for brands buying retail media?
Expect more separate integrations, not fewer. Each retailer that leaves a bundled network becomes another interface, taxonomy and reporting standard to support, so planning teams should budget for rising operational overhead through 2027 rather than assuming consolidation is imminent.
How would we know the prediction was wrong?
Three markers would falsify it: no additional top-100 US or European retailer disclosing a move off bundled managed service by the end of Q1 2027; incumbent retail media revenue returning to year-over-year growth at the Q4 2026 report in February 2027; and a visible run of large retailers signing new full-service network agreements, as Lidl did in May 2026.
Where does in-store retail media fit?
In-store is currently following the technology-purchase pattern rather than the managed-service one. The Best Buy Canada mandate cited in August 2026 was a specific in-store digital out-of-home technology selection, which is closer to buying a capability for a defined surface than to outsourcing a revenue stream.
Does regulation push in either direction?
Tentatively toward in-housing. As sponsored placement blends into algorithmic recommendation, disclosure obligations around commercial steering become harder to evidence when the ranking logic sits inside a third-party system the retailer does not control.