Cookieless Audience Intelligence · Workflow Guide
Ad Inventory Curation: From Audience Thesis to Deal ID
Ad inventory curation is the practice of selecting programmatic inventory against an audience or quality thesis, packaging it as a Deal ID, and distributing it to buyers' DSP seats. It has become the fastest-growing layer of the programmatic supply chain because it lets sell-side data do the targeting work that third-party cookies no longer can. This guide walks through the full curation workflow — who curates, how packages are built, and how domain-level cookieless audience segmentation data selects the inventory that goes into them.
102MDomains with audience data
285Interest sub-segments (INT.*)
283Purchase-intent segments (PI.*)
0User IDs required
Definition
What is ad inventory curation?
In programmatic advertising, curation is a middle layer between supply and demand. A curator takes the enormous, undifferentiated pool of open-exchange inventory — millions of domains and apps available through SSPs — and applies selection criteria to it: which sites, which page contexts, which audiences, which quality thresholds. The result is a bounded, described package of inventory that is exposed to buyers as a Deal ID (a private marketplace deal, or PMP, in most implementations). A buyer activates the deal from their existing DSP seat exactly as they would any other deal: same auction mechanics, same bidder, same reporting — but the supply inside the deal has been pre-filtered by the curator's data.
The important conceptual shift is where the targeting data is applied. In the cookie-era model, audience data lived on the buy side: a DSP matched user IDs against licensed third-party segments at bid time. In the curation model, data is applied on the supply side, before the bid request ever reaches the buyer. The curator decides which impressions qualify for the package; the buyer simply bids into a stream that already embodies the audience and quality definition. Because the selection logic runs against pages, domains and sellers rather than against user identifiers, it works identically on Safari, Firefox, iOS and every other environment where third-party cookies are blocked.
Curation is not a new ad format and not a new auction type. It is best understood as merchandising for programmatic supply: someone with a data advantage or a point of view assembles inventory into a product a media buyer can purchase in one line item, instead of the buyer reconstructing that selection themselves across thousands of domains.
Market context
Why curation has grown so fast
Three structural forces pushed curation from a niche SSP feature to a core programmatic workflow. None of them is a fashion; all three are consequences of how the supply chain is evolving.
The signal moved to the supply side
Safari, Firefox and iOS environments already block third-party cookies by default, which puts roughly 40% or more of web traffic beyond the reach of buy-side audience matching — and privacy regulation shrinks the consented pool further. Cookies remain available on Chrome, but a buyer can no longer assemble a cross-browser audience from user IDs alone. Page-level and domain-level signals — what the content is about and who it is therefore written for — are observable on 100% of impressions, and they are naturally applied where curation happens: on the supply path, before the bid request. The full landscape of alternatives is covered in our guide to cookieless targeting.
Supply-path optimization (SPO)
Buyers spent years discovering that the same impression reaches them through many resold paths with different fees and different transparency. SPO is the response: concentrate spend on fewer, cleaner, better-understood paths. A curated deal is SPO in product form — a named counterparty, a known domain list, a defined selection rule and a single path into the auction. For many agencies, moving open-exchange spend into curated deals is now the default way to enforce supply-path discipline at scale.
Buyers want audience and quality — without IDs
Demand did not get simpler when cookies got scarcer. Buyers still brief on audiences ("home improvers with high household income"), and they simultaneously demand protection from made-for-advertising (MFA) sites, low-attention placements and misrepresented supply. Curation is the one layer where both requirements can be satisfied together: audience definition from content-inferred data, plus quality screening such as MFA detection, applied to every domain before it enters the package.
The players
Who does the curating?
Curation is a role, not a company type. Four categories of organizations run curation businesses today, each bringing a different asset to the selection step.
Independent curation platforms
Companies whose entire product is the curation layer: tooling to build inventory packages from data, mint Deal IDs across multiple SSPs, syndicate them to any DSP seat, and report on delivery. They typically bring workflow and distribution breadth rather than proprietary media, and they monetize through a fee on curated spend.
SSPs with curation products
Most major exchanges now expose self-serve curation: any approved party can define a package over the SSP's supply, attach data, and generate deals. For SSPs, curation converts their position in the bidstream into a data-activation surface — and gives buyers a reason to transact more spend through that exchange specifically.
Agencies and holding companies
Large buyers curate for themselves. Agency groups assemble preferred-supply marketplaces — vetted domains, negotiated paths, quality floors — and route client spend through them. Here curation is less a product than a procurement strategy: the agency captures the selection value instead of paying a third party for it.
Data and measurement companies
Companies whose asset is a dataset — audience attributes, contextual classifications, attention metrics, sustainability scores, quality ratings — increasingly activate it as curated deals rather than (or alongside) licensing it into DSPs. Packaging data as a Deal ID makes it buyable by any seat with no integration work on the buyer's side.
Step by step
The inventory curation workflow
Every curated package, whatever its theme, moves through the same five stages. The craft is almost entirely in stages one and two — the thesis and the selection data; the rest is programmatic plumbing.
Define the audience or quality thesis
A package needs a reason to exist that a planner can repeat in one sentence: "affluent home improvers", "B2B IT decision makers at enterprise companies", "premium news, MFA-screened", "parenting content, new-parent life stage". The thesis determines everything downstream — which attributes select inventory, what the deal is named, which advertisers it is pitched to, and what success looks like. Weak packages are almost always weak here: a list of "good sites" with no definable audience or quality claim is not a thesis, it is a preference.
Select domains and inventory with data
The thesis is translated into filter criteria and run against a dataset that describes the candidate supply. This is where domain-level audience intelligence does the heavy lifting: instead of hand-picking sites from memory, the curator filters a corpus — in our case 102M domains carrying content-inferred audience attributes — for the demographics, interests, purchase intent or firmographics the thesis requires, then intersects the result with quality signals (MFA risk, content categories to exclude, brand-safety flags). The output is a defensible domain list with a documented derivation: every included domain can show why it qualified. The audience segmentation API adds per-URL resolution on top for packages that need page-level rather than site-level precision.
Package as a Deal ID / PMP
The domain (or URL) list becomes a deal in one or more SSPs or curation platforms: a Deal ID with a floor price, auction mechanics (usually a standard second- or first-price auction within the deal), and the curator's fee attached. Larger curators mint parallel deals across several exchanges so the same package is reachable whichever supply path a buyer prefers. Metadata matters at this stage — a clear name, the audience definition, and the included-domain transparency that lets a buy-side trader evaluate the package before spending.
Distribute to DSP seats
A Deal ID is inert until it is in a buyer's platform. Distribution is part sales, part operations: the deal is pushed or synced into the DSP, associated with the advertiser's seat, and attached to a line item like any other supply source. Because activation happens in the buyer's existing stack, curated packages require no SDK, no integration and no new contract with each publisher — which is precisely why data companies find curation a lower-friction route to activation than buy-side integrations.
Measure, prune and refresh
Post-launch, the package is a living product. Delivery reporting shows which domains actually clear and at what CPMs; performance data shows which contribute outcomes; and the underlying audience data itself shifts as sites change their content. Mature curators re-run the selection on a schedule — dropping domains whose audience profile or quality score has degraded, admitting new domains that now qualify, and tightening filters where delivery concentrated on the weakest members. A package that is never refreshed decays into exactly the static site list curation was meant to replace.
The selection engine
How domain-level audience data selects inventory
Step two of the workflow is where curation quality is decided, so it is worth being concrete about how data-driven selection works. Our approach infers the likely audience of a domain from its content — what the site is about, who it is written for — rather than from tracking its visitors. No cookies, IDs or user-level data are involved at any point, which makes the resulting packages privacy-safe by construction. Persona assignments are deterministic (a fixed IAB-category-to-persona mapping across a 1,667-persona taxonomy), while demographic, interest, intent and firmographic attributes are model-inferred from content and carry a banded confidence value — low, medium or high — that curators can threshold to trade reach against precision.
Every attribute is expressed in a controlled, versioned vocabulary (v1.0, aligned with IAB Audience Taxonomy 1.1), so a filter written today returns the same kind of answer next quarter. A curator building a package queries the 102M-domain corpus along four audience dimensions and one quality dimension:
Demographics8 age brackets, 5-point gender skew, 6 income bands, 7 education levels, 14 life stages
Interests29 interest groups, 285 sub-interests — coded as INT.* (e.g. INT.home_garden)
Purchase intent34 intent groups, 283 in-market segments — coded as PI.* (e.g. PI.travel.hotels_and_resorts)
B2B firmographicsIndustry, company-size and seniority bands on LinkedIn-standard scales
The quality dimension runs alongside: content classifications to exclude unsuitable categories, and inventory-quality signals such as MFA scoring to keep made-for-advertising domains out of packages that sell on quality. Combining the two is what distinguishes a curated deal from a bare contextual segment — the package asserts both who reads this inventory and that the inventory is worth buying. The complete coded structure of every vocabulary — all brackets, bands, INT.* and PI.* codes — is browsable on the audience segmentation taxonomy page.
Worked example
Curating a “High-Income Home Improvers” package
A home-improvement retailer briefs its agency: reach households likely to undertake renovation projects, skewing 35–54 and higher-income, on quality inventory. Here is how that brief becomes filter criteria against the domain corpus, and what a qualifying domain's profile looks like. (Domain shown is illustrative.)
// Selection filters run against the 102M-domain corpus { "audience_filters": { "age_bracket": ["35_44", "45_54"], "income_band": ["upper_middle", "high"], "interests": ["INT.home_garden"], "purchase_intent": ["PI.home_garden_services.home_improvement_and_repair", "PI.home_garden_services.landscaping_services"], "min_confidence": "medium" }, "quality_filters": { "mfa_risk": "low", "exclude_categories": ["sensitive_topics"] } } // One qualifying domain from the result set { "domain": "renovation-monthly.example", "age_bracket": [ {"code": "35_44", "confidence": "high"}, {"code": "45_54", "confidence": "medium"} ], "income_band": {"code": "upper_middle", "confidence": "medium"}, "interests": ["INT.home_garden"], "purchase_intent": [ {"code": "PI.home_garden_services.home_improvement_and_repair", "confidence": "high"} ], "mfa_risk": "low" }
Package definition, rendered
Age brackets
35–4445–54
Income bands
Upper-middleHigh
Interest group
Home & Garden
In-market segments
Home ImprovementOutdoor Living
Quality gate
MFA risk: LowConfidence ≥ Medium
The filtered domain list becomes the deal's inclusion list. Each domain carries its own attribute profile, so the curator can show a buyer — per domain — exactly why it is in the package: coded values (age_bracket 35_44, PI.home_garden_services.home_improvement_and_repair) render as the plain-language labels above. Tightening min_confidence to "high" shrinks the list toward precision; relaxing it extends reach.
Buying models compared
Open exchange vs curated deal vs direct
Curation sits deliberately between the two older buying models — more defined than the open exchange, more scalable than direct deals. The comparison below shows where each model gets its targeting data and what the buyer controls.
| Dimension | Open exchange | Curated deal (Deal ID) | Direct / programmatic guaranteed |
|---|---|---|---|
| Targeting data | Buy-side only: DSP segments and bid-request signals. Audience matching depends on IDs, so coverage collapses on cookieless traffic. | Supply-side: curator applies audience and quality data (content-inferred, ID-independent) before the bid request — works on 100% of traffic. | Publisher first-party data and declared context, negotiated per deal; strongest on the publisher's logged-in audience, scoped to that publisher. |
| Transparency | Low–variable Long, resold supply paths; domain lists discovered after the fact in delivery reports. |
High by design Named curator, disclosed selection logic and inclusion list; one known path per deal. Curation fees should be disclosed — buyers should ask. |
Full Contractual placements, direct counterparty, no intermediary selection. |
| Buyer control | Broad reach with blunt instruments: block lists, pre-bid filters, whatever the DSP exposes. | Dual-layer: the package defines the supply, and the buyer keeps every DSP-side control (frequency, brand safety, bidding) on top of it. | Maximum control over placement and price, minimum flexibility mid-flight; changes require renegotiation. |
| Scale & effort | Effectively unlimited scale, minimal setup, highest variance in what you actually buy. | Bounded but large — a data-driven filter over millions of domains; one line item to activate, refreshable without renegotiation. | Limited to negotiated publishers; highest per-deal effort in sales cycles and trafficking. |
| Typical pricing | Open auction at floor prices; lowest CPMs, fees embedded along the path. | Auction within the deal above a negotiated floor; curation fee added to the supply path in exchange for the selection work. | Fixed or guaranteed CPMs at a premium, often with guaranteed volume. |
Adjacent standard
Curation and seller-defined audiences
Curation is closely related to — but distinct from — the IAB Tech Lab's seller-defined audiences framework. Under seller-defined audiences (SDA), the publisher declares who its audience is, expressed in standardized IAB Audience Taxonomy codes and passed in the bid request, so buyers can target audience attributes without any user identifier changing hands. Curation and SDA answer the same buyer question — "whose attention am I buying?" — from two different positions in the chain: SDA is the publisher's first-party declaration; curation is a third party's selection across many publishers.
In practice the two are complementary, and curators are often the main consumers of SDA signals. A curator can build packages from publishers' seller-defined audience declarations, aggregating consistent segments across dozens of sellers into a single Deal ID a buyer can activate in one step. And where publishers do not (or cannot) publish SDA signals — which is still most of the long tail — content-inferred, domain-level audience data fills the gap with an independent, consistently-defined estimate of each domain's audience. Because both our vocabularies and SDA are aligned with IAB Audience Taxonomy 1.1, a curator can express a package thesis once and satisfy it from either source: declared where available, inferred everywhere else. It also gives curators a validation layer — an inferred profile that broadly agrees with a publisher's declaration is evidence the declaration is honest, which addresses the trust question that has followed SDA since its introduction.
For buyers, the takeaway is simple: curation is the distribution mechanism, and audience definitions — seller-declared or content-inferred — are the raw material. The packages worth paying a curation fee for are the ones that can show their raw material.
FAQ
Ad inventory curation, frequently asked
What is ad inventory curation in programmatic advertising?
Ad inventory curation is the practice of selecting programmatic inventory against a defined audience or quality thesis, packaging the selection as a Deal ID (private marketplace deal), and distributing it to buyers' DSP seats. The curator applies data — audience attributes, contextual classifications, quality scores — on the supply side, before the bid request reaches the buyer, so the deal delivers a pre-filtered stream of impressions that the buyer activates like any other deal from their existing platform.
Who are the curators in programmatic advertising?
Four categories of organizations curate inventory: independent curation platforms whose product is the packaging and distribution workflow itself; SSPs and exchanges that offer self-serve curation over their own supply; agencies and holding companies that build preferred-supply marketplaces for their clients; and data or measurement companies that activate their datasets as curated deals instead of licensing them into DSPs. Anything with a data advantage and access to supply can, in principle, curate.
What is the difference between a curated deal and a PMP?
Technically a curated deal usually is a PMP — it is transacted as a Deal ID with a floor price and auction mechanics. The difference is who assembles it and from what. A classic PMP is a single publisher packaging its own inventory for a buyer. A curated deal is assembled by a third party (a curation platform, SSP, agency or data company) across many publishers, selected by data against an audience or quality definition rather than by a direct publisher relationship.
How do curators target audiences without third-party cookies?
By applying audience data to inventory rather than to users. Content-inferred audience intelligence estimates who a domain's or page's readers are likely to be — age brackets, income bands, interests, purchase intent, B2B firmographics — from what the content is about, with no cookies or user identifiers involved. Since the signal is attached to the inventory itself, it works on Safari, Firefox and iOS, where third-party cookies are blocked by default, exactly as it does on Chrome, where they remain available. Publishers' seller-defined audience declarations serve the same goal from the publisher's side, and curators frequently combine both.
How is inventory actually selected for a curated deal?
By filtering a described supply corpus against the package thesis. A curator working with domain-level audience data queries for domains whose inferred audience matches the target — for example age brackets 35–44 and 45–54, upper-middle or high income bands, Home & Garden interests and Home Improvement purchase-intent segments — at or above a chosen confidence band, then intersects the result with quality filters such as low MFA risk and content-category exclusions. The filtered domain list becomes the deal's inclusion list, with a documented reason for every domain's membership.
Do curated deals cost more than the open exchange?
Usually yes, on a CPM basis: curated deals carry negotiated floors and a curation fee that compensates the selection work. The relevant comparison, however, is effective cost against outcomes. Open-exchange buys include impressions on made-for-advertising and mismatched-audience inventory that deliver little value; a well-built curated deal removes them before the auction. Whether the fee is justified depends on the curator's data actually being predictive — which is why buyers should ask curators to show the selection criteria and per-domain rationale behind a package.
Build your next package on audience data you can inspect
Open the live demo and pull the content-inferred audience profile for any domain or URL — demographics, interests, purchase intent and personas with banded confidence — the same attributes a curation filter runs on.