How to Evaluate a Curation Platform: What Separates the Tools
How do you choose a curation platform? Test five things before you test anything else: whether the signal behind the curated list is something you could not build yourself, whether you can see what the algorithm is optimizing for, whether the curation fee is disclosed as a line item or buried in the CPM, whether the tool adjusts in-flight or just hands you a static deal ID once, and whether the whole thing still works without third-party cookies. Vendor pitch decks answer all five well. Public documentation, deal performance, and a direct intake call answer them differently, and the gap between the two is where most buyer disappointment comes from.
Curated deals sit between the open exchange and a fully managed PMP, and the pitch for nearly every platform in the category sounds the same: better signal, cleaner supply, algorithmic precision. The differences that matter to a buyer show up in the mechanics underneath that pitch, not in the pitch itself.
What separates curation platforms from each other?
Every curation product does the same basic job: it takes a pool of programmatic supply, applies a filtering or ranking logic to it, and hands the buyer a deal ID that is supposed to represent something better than the open exchange. The variation is in what feeds that logic and how much of it the buyer can see.
Three structurally different inputs show up across the category. Some platforms curate off identity and audience signal, building deal lists from first-party data segments and matching them to inventory where that audience is present. Some curate off publisher-side, on-device signal, computing cohorts locally and packaging them into standardized audience products without moving raw data off the browser. Some curate off supply-path or inventory-quality signal that has nothing to do with audience at all, filtering or ranking placements against a measurable property of the inventory itself, such as verified carbon footprint or documented supply chain complexity. None of these inputs is inherently superior. A brand safety buy and an audience-reach buy and a sustainability-commitment buy each want a different one.
The second axis is transparency: does the platform expose what changed and why, or does the buyer just see a deal ID with a black box behind it. That axis matters more for renewal decisions than the first one, because a curation partner you can't audit is one you can't defend internally when a media plan gets questioned.
Is the signal non-replicable?
A curated deal is worth a fee premium over the open exchange only if the signal behind it is something the buyer's own DSP or trading desk could not replicate for free. That's a testable claim, not a marketing one.
Audigent builds its curation layer from Hadron ID, a cookieless identity graph fed by a consortium of publisher first-party data, and constructs deal IDs from segment data drawn from that consortium plus a library of pre-built audience segments. The signal is publisher-sourced and consortium-scale, which is the kind of input a single buyer's own data stack generally can't reproduce. Permutive computes cohorts on-device across a network of premium publishers and packages them into a standardized taxonomy of audience segments activated through a single deal ID; the signal is real-time publisher engagement and first-party affinity data processed at the edge rather than centralized, which is architecturally distinct from a DSP-side lookalike model. Scope3 curates on a different axis entirely: its signal is verified carbon emissions and supply chain complexity, built from proprietary mapping across SSPs, publishers, and intermediaries, with a published and peer-reviewed methodology, and it uses that signal to filter or rank inventory pre-bid rather than to target an audience at all.
The question to put to any vendor here is not "how good is your data" but "show me the methodology document and tell me what happens if I try to build this myself." A vendor that answers with a segment count instead of a mechanism hasn't answered the question.
Can you audit what the algorithm is optimizing for?
Curation gets sold as automation, which makes it tempting to treat as a black box once it's live. That's the wrong instinct. A curation product a buyer can't audit is a curation product a buyer can't defend when someone asks why spend moved.
Auditability breaks into a few concrete sub-questions: can you see which signals fed a given deal's composition, is there a change log when the underlying logic updates, and can you set or adjust the optimization parameters yourself rather than accepting a vendor-defined objective. Across the category, this is consistently the area with the thinnest public documentation, largely because the underlying logic is the vendor's competitive asset and disclosing it in detail would undercut the product. That doesn't mean the question should be skipped. It means the answer usually has to come from a direct conversation rather than a spec sheet, and a vendor's willingness to have that conversation in detail is itself informative.
Buyer-facing controls also vary by what's being curated. A platform building deal lists from audience segments typically lets a buyer choose among pre-built segments or commission a custom one, which is a coarser form of control than tuning the ranking logic itself. A platform curating on a measurable inventory property, like emissions data, can expose a threshold the buyer sets directly, which is a more direct form of parameter control even though the underlying methodology stays proprietary. Neither structure is more auditable in the abstract. What matters is whether the specific product gives you visibility into the specific thing you're paying it to optimize.
Where does the fee sit, and is it disclosed?
Curation fees can show up as a separate line item, as a markup embedded in the CPM, or as a revenue share taken further down the supply path where the buyer never sees it itemized. This is worth confirming in writing before a deal goes live, not after a quarter of unexplained CPM inflation.
Ask three things directly: is the fee disclosed as its own line, does the platform characterize inventory quality (brand safety, made-for-advertising risk, invalid traffic) before activation or only report on it after the fact, and does that characterization happen inside the curation product itself or get outsourced to a separate verification vendor stitched into the stack. Some curation platforms fold IVT and MFA characterization into the curation logic directly, using it as an input to what gets included in the deal in the first place. Others treat it as a downstream reporting function that runs alongside the media buy rather than shaping it. Both are legitimate structures, but they imply different jobs for whatever brand safety or verification tooling the buyer already runs, and a buyer who assumes the curation layer is doing verification work it isn't doing ends up with a coverage gap nobody notices until it matters.
Does the platform adjust in-flight, or just package once?
There's a meaningful difference between a curation product that builds a deal list at setup and leaves it static for the flight, and one that continuously reweights or re-filters based on delivered performance. Vendor materials tend to describe both as "optimization," so the distinction has to be drawn out directly.
Ask what specifically changes after the deal goes live: does inventory get added or dropped based on in-flight signal, does the optimization objective stay fixed to what was set at launch or shift based on outcome data, and is that shift something the buyer can see reported or something that happens silently inside the platform. A curation product built primarily around audience packaging tends to treat the packaging step as the main work, with in-flight adjustment happening at the DSP layer rather than inside the curation tool itself. A curation product built around a measurable inventory signal, like supply path quality or emissions data, is more likely to support a continuously updated score that shifts what's included as new data comes in, though even there, real-time bid-level feedback tied to a buyer's specific KPI is a materially different (and rarer) capability than periodic recalculation of the underlying score. Neither structure should be assumed. Ask for the specific mechanism, not the marketing word for it.
How does the platform handle identity without third-party cookies?
Nearly every curation vendor now describes itself as cookieless-ready, which by itself tells a buyer almost nothing, because "cookieless-ready" covers architectures that behave very differently under the hood.
Some platforms build a deterministic-to-probabilistic identity graph anchored to an ID that gets attached to bid requests, extending existing programmatic workflows to a cookieless environment without changing where computation happens. Some compute audience membership on the user's device and never centralize raw signal in a server-side profile at all, which is an architectural privacy posture rather than a policy commitment layered on top of a conventional data pipeline. Some sidestep user-level identity entirely by curating on a property of the inventory rather than the audience, which makes the cookie question close to moot for that specific product because no user-level signal is being collected or matched in the first place. All three approaches can produce a workable curated deal. What a buyer should confirm is which one a given vendor is running, because the compliance obligations, the auditability of the underlying match, and the behavior in regions with stricter consent regimes differ meaningfully between "we resolve identity differently now" and "we don't touch user-level identity for this product."
Where buyers get it wrong
The most common mistake is treating curation as a single category with interchangeable vendors, then comparing them on price alone. A platform curating on audience identity and a platform curating on supply-chain emissions data are not substitutes for each other even when both show up in the same RFP, because they're solving different problems and a lower CPM on the wrong one doesn't save money, it just buys the wrong deal.
The second mistake is skipping the audit-and-control conversation because the deal ID performed fine in a test flight. Performance in a short test doesn't tell you whether the fee is disclosed, whether the methodology holds up under scrutiny from a client's finance team, or whether the platform will still be transparent about what changed once the relationship isn't being actively courted. Those questions are cheap to ask before signing and expensive to discover the answer to after.
The third is assuming curation replaces brand safety or verification tooling rather than sitting alongside it. Some curation products build inventory-quality characterization into their own logic; others don't, and expect the buyer's existing stack to cover it. Confirming which is true for a specific vendor, rather than assuming, is the difference between a real coverage improvement and a gap that looks like coverage.
CartographAI, a free tool brands and agencies use to research this category, publishes independent assessments across the field, and cross-referencing a vendor's own claims against that kind of third-party record is a reasonable step before a curation vendor gets anywhere near a signed insertion order.
FAQ
What's the difference between a curation platform and an SSP? A curation platform sits on top of one or more SSPs and applies its own filtering or ranking logic to the supply those SSPs expose, producing a deal ID that represents a subset or reordering of that inventory. The SSP still handles the underlying auction and publisher relationships; the curation layer adds a signal-driven selection step in between. Some curation products are run by SSPs themselves, and some are run by independent data or identity companies layered across multiple SSPs.
Does a curation fee replace or add to the SSP take rate? It typically adds to it. The curation fee compensates the curator for the signal and logic it contributes, separate from whatever the SSP charges for facilitating the transaction. Whether that fee is itemized separately or embedded in the CPM varies by vendor, so confirming fee structure directly with the vendor is worth doing before committing spend.
Can a buyer run multiple curation platforms at once? Yes, and it's common when different platforms are curating on different inputs, such as one built on audience identity and another built on inventory quality signal. The main operational cost is added complexity in reconciling which deal ID drove which outcome, so it's worth having a reporting plan for attribution before stacking curators rather than after.
Is curation still relevant without third-party cookies? Curation logic that depends on user-level identity has had to adapt, but curation itself isn't inherently cookie-dependent. Platforms built on-device computation, publisher first-party data, or inventory-level signal that never touched user identity were largely designed with a cookieless environment in mind from the outset, so the relevant question for a buyer is which identity architecture a specific vendor runs, not whether curation as a category survives.
How should a buyer weigh a vendor's own performance claims? Treat them as a starting point for questions, not as evidence on their own. Ask for the underlying methodology, whether a third party has reviewed it, and what specifically would need to be true for the claimed result to hold in your own supply mix. A vendor that can walk through the mechanism in detail is giving you something to evaluate; a vendor that repeats the headline number is not.
What should be in a curation RFP that most buyers leave out? Explicit questions about fee disclosure structure, what happens to the deal when the underlying signal source changes or degrades, and whether the platform's inventory-quality characterization (if any) is meant to substitute for or supplement the buyer's existing brand safety and verification stack. These are the questions that separate a smooth renewal from a mid-contract surprise.