The Great Repackaging

The week the middle got a name

This week Digiday reported that Stagwell is building its own AI media curation marketplace. Read the parts list slowly, because the whole essay is in it. Stagwell Curate pools pre-vetted CTV, online video, display, and audio inventory into a reservoir the holding company owns. AI agents built with Claude score domains for ad quality, supply-chain quality, and technical performance — using OpenSincera, the free quality-metadata API that belongs to The Trade Desk. Magnite and FreeWheel supply the publisher plumbing. And when the package is cut, The Trade Desk still executes the bids.

An agency building supply-side curation out of buyer-side agents, on the DSP’s own quality telemetry, partly to reduce its dependence on that DSP’s direct-path product — while still buying through it. “This isn’t just using an off-the-shelf package provided by an SSP,” Stagwell’s Matt Adams told Digiday. “We control which sites and apps are eligible to send bid requests into our deals.” Gartner’s Andrew Frank called agencies doing their own curation unusual, “but it’s also a sign of the times.”

It is. By mid-2026, everyone in programmatic is a curator: every major SSP has a named curation platform, the credit bureau owns the category’s pioneer, the holdcos are building marketplaces, the retailers are packaging CTV, and the loudest institutional critic of curation has quietly assembled the most complete curation stack in the industry. Something structural is happening, and it deserves a more precise reading than either the celebration (“finally, quality!”) or the dismissal (“the ad-network tax reborn”) gives it.

What a curated deal actually is

Strip the branding and curation is one move: the decision moves to the supply side. A curator — an SSP product, a data company, an agency desk — applies audience, contextual, or supply-quality data to inventory before the bid request reaches the DSP, and exposes the result as a deal ID any buyer can target from any DSP. The impression is usually the same open-auction impression it always was. What is new is where the selection logic runs, and who charges for it.

Two mechanics separate this from the private marketplaces of a decade ago. First, curated deals are overwhelmingly multi-seller: Jounce Media’s census found multi-seller curated deal IDs made up 73% of programmatic bidding opportunities by spring 2025, against 8.7% for traditional single-publisher deals and 18.6% for the pure open auction. The average bid request now travels with an open auction and three and a half deal IDs attached. Second, a curated deal carries no preferential access — no first look, no floor advantage by default. It is open exchange inventory wearing a claim: someone selected this, and the selection is worth paying for.

The lineage matters for what comes later. Xandr Curate, launched in April 2021, is widely credited with kicking off SSP curation as a product category. Audigent — the specialist that built SmartPMPs by marrying publisher first-party data to premium inventory inside a single deal ID, co-locating in SSP data centers to enrich bids in flight — proved a standalone company could live in that seam. The boom, though, arrived in the back half of 2024, and it arrived for reasons that had little to do with any curator’s ingenuity.

Why the middle thickened

Five forces, compounding:

Signal loss, then signal whiplash. Google spent 2024 retreating from cookie deprecation and, in April 2025, confirmed it would not even ship the planned user-choice prompt. The identity rebuild the industry had half-financed was suddenly optional — but the underlying decay (Safari, Firefox, regulation, consent) never reversed. Data application drifted to the one place that still sees page context and publisher first-party signals natively: the supply side.

The MFA reckoning. The ANA’s programmatic transparency work put a number on waste — roughly $13 billion a year on made-for-advertising inventory — and the industry’s reflex was smaller, vetted pools. Curation productized the reflex.

SPO logic, inverted. Supply-path optimization taught buyers to shorten paths. Curation is, as Ari Paparo put it in early 2024, the anti-SPO — SSPs escaping commoditization by owning differentiated demand. But there is a real distinction from the reseller era, and Jounce’s Chris Kane makes it: curators cannot mint new bid requests. A curator repackages the auction; a reseller multiplied it.

DSP fee pressure and distrust. As buyers turned auditors, the DSP take rate became a target — and the events of 2026 (more below) did not help the DSPs’ case.

The retail-media halo. Retailer purchase data is the scarcest signal in the market, and offsite activation needed a distribution rail. Curated deals were sitting right there.

The land grab, camp by camp

The SSPs built the storefronts. Index Exchange launched Index Marketplaces in January 2024 after a four-month closed beta that hit a $100 million run rate — the fastest-growing product in the company’s history. Andrew Casale’s pitch is infrastructure, not data: “If you wanted to do this on your own, you’d have to connect to 2,000 publishers and be able to handle half a trillion requests a day.” PubMatic’s Connect grew 140% year over year in Q4 2024 across roughly 190 curation and data partners, and CEO Rajeev Goel offered the category’s most honest defense of the fee — “curation is work.” Magnite folded curation and activation into ClearLine on SpringServe’s ad-serving rails so curated deals execute closer to the impression; by Q1 2026 CTV was more than half its business and Michael Barrett was describing AI as foundational “from agentic buying, to creative development, to inventory curation.” Equativ relaunched its curation platform as Maestro days after completing the Sharethrough integration. FreeWheel shipped a buy-side Curation Hub. OpenX made OpenXSelect generally available — on the same July day, as it happens, that TripleLift published a survey of 223 buyers finding 99% call curation important, 21% rate their own execution very effective, and 60% do not know what they pay for it. That is the category in three numbers.

The data owners got absorbed. Experian announced its acquisition of Audigent in December 2024 — $200–250 million per Digiday’s sources — converting a three-year partnership into ownership and giving a credit bureau sell-side distribution for its identity spine. Drew Stein’s stated logic deserves quoting because it is the whole thesis of the category: deterministic IDs hosted in matching tables will lose fidelity, so the data must move to the supply side. Three months later Publicis announced it would fold Lotame — 1.6 billion IDs — into Epsilon. The pattern is unmistakable: the independent data-curation layer did not consolidate into ad tech; it consolidated into data owners — a credit bureau and a holding company. The specialists who stayed independent survived by differentiation: Onetag with an AI curation co-pilot and a creative-tech acquisition, Multilocal as curation-as-a-service, Permutive syndicating publisher cohorts, Seedtag riding contextual into CTV. Scope3, the climate-flavored curator, pivoted to an agentic media platform — more on that below.

Retail × CTV is where curation’s money is compounding. On a single day — April 27, 2026 — Roku launched Roku Curate (pre-built CTV packages fusing Roku signals with purchase data from Instacart, Kroger, Best Buy, Criteo, Fetch, and Fandango) and Walmart announced Connect Select (a curated CTV marketplace inside Walmart DSP, with Magnite, PubMatic, FreeWheel, and Index as rails and VIZIO, Paramount, and Warner Bros. Discovery as supply). Retail purchase data is the scarcest signal in the market; curated deals are how it travels offsite. eMarketer made curation a center-stage CTV theme for 2026, with US CTV spend forecast up 14.5% to $37.95 billion — and the premium streamers transact almost entirely through PG and PMP structures. In CTV, there effectively is no open auction to curate against; curation is the market.

The holdcos are the newest camp and the oldest story. WPP rebranded GroupM as WPP Media in May 2025 and launched Open Intelligence on top of the Premium Marketplace it has run with Magnite and PubMatic since 2022. Omnicom closed the IPG merger in November 2025, adding Acxiom to Omni — and by spring 2026 was executing live agent-to-agent buys while John Wren called DSP and SSP fees a “toll” and direct publisher relationships “an objective.” Dentsu’s DMX curates sites-and-pipes inside Magnite’s CTV ad server with Merkury identity on top. Havas runs Converged. And now Stagwell Curate. The through-line is not subtle: agencies are no longer choosing supply paths; they are becoming the supply path.

Which is why the principal-media receipts belong in this essay. Litigation disclosures put GroupM’s principal revenue past $1 billion in 2023; AdExchanger pegged Omnicom’s 2025 principal take north of $2 billion; Publicis’ CFO has acknowledged principal is over half the equivalent line. The ANA’s March 2026 study found 58% of marketers now buy principal media, up from 47% — while 90% doubt the recommendations serve their interest, up from 79%. Curation marketplaces are principal-style economics with better optics: dashboards, client-visible buy data, a quality stamp. Sometimes that is genuinely better. It is still the packager grading the package.

The loudest skeptic built the biggest curation stack

No company has spent more rhetorical energy against curation than The Trade Desk, and no company has assembled more of its machinery. The official line, from inventory-development VP Will Doherty in January 2025: “Curation can be a vacuous term,” and some versions of it “butt heads with our desire to have a fully transparent supply chain.” Jeff Green’s framing at Marketecture Live this March compared SSPs to house flippers, against OpenPath’s agent-style 4.5% fee.

Now inventory the actions. OpenPath, launched February 2022 with Reuters, The Washington Post, Condé Nast, and Hearst — direct publisher paths, announced alongside TTD’s exit from Google’s Open Bidding. The Sincera acquisition, announced January 2025 — independent telemetry on ad quality, embedded across Kokai, with the front end open-sourced that May as OpenSincera. Deal Desk, mid-2025 — AI-scored management of the very deal IDs curation trades in. Kokai’s reclassification of SSP paths as “resellers” in late 2025, penalizing them against OpenPath and TTD’s own curated list of efficient supply routes — the product that drew Casale’s unusually blunt “quite frankly, ignorant” at ATS London. And OpenAds, announced in October 2025 after Prebid changed how transaction IDs propagate: a TTD-run fork of the auction stack itself, with a first publisher wave in January 2026. A direct path, a quality-score API, a deal manager, an inclusion list, and an auction. If an SSP shipped that stack, the trade press would call it curation.

The distinction TTD would draw — transparency of fees, no yield management, a published take rate — is real and worth stating fairly. But 2026 has been unkind to the buy side’s moral high ground. Digiday reported US buyers pausing OpenPath after finding a 10–15% premium; Adweek reported Dentsu and WPP quietly exiting it, one executive noting OpenPath “overlaps with agency services, such as curation and SPO.” And in March, Publicis pulled TTD from its recommended list after an audit alleged ad-tech fees stacked beyond contract — TTD’s stock fell roughly 13% — before the two settled in June without disclosing terms. Three of the five biggest holdcos at odds with the biggest independent DSP inside a year, each dispute at bottom about the same thing: who may package, and what the packaging costs.

The fee question, honestly

The critique writes itself — AdExchanger’s May 2025 verdict was “the same old ad tech tax” — but the audited picture is more interesting than the slogan. Jounce’s census: roughly half of curated deal IDs carry no fee beyond the exchange’s standard take. About a third carry margin-based fees with a median of 14%. A tenth charge fixed CPMs, and a small static-price tail runs to effective margins above 50%. Scott Messer’s summary is the fairest one-liner in the discourse: “the net amount to the publisher is roughly the same” — the tax did not grow so much as redistribute.

So the honest questions are not “is there a fee” but three sharper ones. Is the selection real? Kane’s worst case — a curator charging a supply-chain fee “for simply loading a site list into an SSP” — describes a nonzero share of the market, and Onetag’s co-CEO says the quiet part: the space is “crowded with non-differentiated middlemen.” Who verifies the lift? There is no third-party verification layer for curation quality; fees ride inside deal terms where procurement never sees them; buyers are left to A/B test deal against open market themselves — Vox Media’s revenue chief was blunt that “most publishers are not seeing any lift.” That is a familiar disease on this site: a claim without an instrument is a price, not a measurement. And does the moment survive the packaging? A curated segment is the past tense of intent, cut in advance, sold as a shelf unit. The premium it commands rests on the belief that yesterday’s selection is still tomorrow’s relevance.

The receipt arrives

In December 2025, IAB Tech Lab released the Deals API — a standard for syncing deals from sell side to buy side that, for the first time, requires a deal to declare the business entities involved in packaging it: seller, packager, and curator. It reads like plumbing. It is actually the most important sentence in the category, because it converts curation’s central claim from an assertion into an attributable one.

This is the same move this site’s trust season kept circling: value that cannot be self-declared must be conferred, and conferral requires a name that answers. A curated deal is a quality claim wearing a deal ID. Until the claim carries a signature — who selected this, from what, under what policy, taking what — it prices as suspicion, which is precisely the discount the whole category trades at today: 99% of buyers say curation matters; 60% cannot say what they pay. The Deals API is the beginning of curation’s accountability chain. What travels with the package is the difference between a curated deal and a signal container — the container was always just curation with the provenance, consent, and policy attached and portable. Curation is packaging that learned to charge. Containerization is packaging that learned to answer.

What agents change — the tense, not the fact

The obvious forward question: if a buyer’s agent can evaluate supply itself, why pay for a pre-cut package? The glib answer — agents kill curation — misreads what is actually being built, starting with the fact that both agentic standards camps made curation a first-class object. AdCP lists a Curation Protocol on its roadmap and names curators as ecosystem actors; IAB Tech Lab’s stack gave curation its disclosure rail in the Deals API. Scope3’s agentic platform is, in essence, curation made continuous — brand-standards agents evaluating every impression instead of a deal cut in advance. Magnite is developing agent-to-agent transactions inside SpringServe. And Stagwell built its marketplace out of agents. The disruptor is being used to industrialize the incumbent.

What agents actually change is curation’s tense. A curated deal is a pre-computed answer to “which supply, for whom” — computed yesterday, sold today, decaying tomorrow. An agent with a compute budget inside the auction can compute that answer per brief, per moment, at decision time. Ari Paparo’s “curation by wire” sketch has this right: scaled SSPs will let agents package inventory dynamically, and the static deal ID survives mainly where a buyer’s definition is tight, stable, and widely available. The deal ID persists — as the settlement rail. The shelf is what dies: the pre-cut catalog of someone else’s yesterday, priced as if it were your today.

Read through the context economy’s lens, the whole land grab clarifies. Curation is the audience economy reorganizing on the supply side — who-and-what packaging, moved to where the signals still live, defended by whoever owns the data. It is the last, best trade of the segment era, and it is a genuinely good trade for exactly as long as moments are too expensive to compute in real time. That window is closing at the speed of colocation. The curators who survive it will be the ones whose packages stop being lists and start being governed, portable claims — selection with provenance, policy, and a signature, assembled as close to the moment as the rails allow. The ones who don’t will discover that “we picked these sites last quarter” is not a product an agent will pay for.

For operators, then

If you sell media: curation is incremental demand exactly as long as you can see the packaging. Demand the fee disclosure, watch for your premium inventory being bundled into someone else’s halo, and treat the Deals API’s curator field as a contract term, not metadata.

If you own data: the Audigent and Lotame exits mark the window. Independent data with supply-side distribution is an acquisition target; independent data without it is a margin waiting to be disintermediated. Either way, the durable asset is not the segment — it is the governed, provenanced package the segment travels in.

If you buy: the three questions above are the whole audit — is the selection real, who verifies the lift, does the moment survive the packaging. Price curated deals against your own open-market A/B, insist on knowing all three named entities in every deal, and treat any curator who cannot answer “what do you take” as having answered.

And if you are building the agentic stack: curation is your adoption path, not your enemy. The market has already agreed that packaged, quality-claimed supply is worth paying for. What it has not built is packaging that answers for itself. That is the open seam — and as of this week, with a holding company assembling bespoke supply out of Claude agents and another company’s open quality API, everyone can see where it runs.