Curation was supposed to be a big f*cking deal. So what went wrong?

Curation promised a big f*cking deal and delivered a big f*cking disappointment. Which is a shame, because another round of ad network reincarnation is burying the very innovation that could deliver on the promise.  

It arrived at exactly the right moment. Made-for-advertising sites had turned the open web into a reputational and performance minefield, and advertisers were waking up to it. Meanwhile, CTV was elevating the open internet, bringing professionally produced, premium content into an ecosystem that had too often been treated as the internet’s bargain bin.  

Curation promised to sort through all of that. The word itself implied judgment. Someone, or something, was selecting the good stuff. 

And that mattered because programmatic has fundamentally been about getting a good deal. Not merely buying cheap inventory, but finding inventory the market has undervalued.  

That version of curation promised to be a BFD. 

But then it got co-opted.  

The reason it was co-opted is economic. I started to understand this while interviewing dozens of candidates coming from audience, curation and managed service companies. What I heard, over and over: these companies are taking 40–70% margin (aka their take rate of the total “media dollar”). Not as exceptional deals or embarrassing edge cases, but as. a. rule. 

And a lightbulb went off for me: certain ad tech business models never really die. Like cockroaches, they find a way to keep living in the new ecosystem. 

A company built around an opaque 70% take rate cannot easily wake up one morning and become a transparent software company or change its pricing model. Cutting the take rate from 70% to 30% would mean needing roughly 2.3 times the client spend just for flat annual growth. 

So the pressure to repackage is enormous. Curation offers a fresh story to businesses still attached to some very old economics: aggregate inventory, sell the package and retain a spread the buyer can’t easily inspect. You don’t have to have started as an ad network to end up behaving like one.  

The original promise was selectivity. But once everyone rushed in, almost anything that could be pushed into a deal ID could be called curated.  

Curation is especially seductive because the technology can feel free. The fee is baked into the media, which makes it easy for a vendor to say the product costs nothing. Of course it costs something. The buyer simply cannot see where the media ends and the fee begins. 

This creates one of the industry’s most perverse dynamics: the transparent fee is the fee that gets negotiated. The hidden fee is considered “free.” 

And then there is the actual product “build.” The same interview candidates described curation offerings as mostly built from hand-picked site lists or a one-time model that categorized inventory into static deals. So advertisers end up buying the same thing as their competitors. Just as AI was making advertiser-specific valuation possible, these offerings were selling everyone the same answer. 

Meanwhile, a genuinely useful technology has found an application in the same supply-side pipes: containerization. The name is less exciting than “curation.” The possibilities are far more interesting. 

A software container packages code and the dependencies it needs to run in a compatible environment. The IAB TechLab’s Agentic Realtime Framework (ARTF) makes it possible for independent companies to deploy their own decisioning inside an SSP. Other paths to leverage this technology for both sell-side and buy-side decisioning are emerging.  

Decisioning directly inside the SSP or potentially a DSP exposes page-level signals: a model can evaluate the individual URL instead of relying on a domain roll-up. Many more signals of page quality and relevance can be scored in ways that give brands more precision and visibility. 

That is materially different from assembling a list of publishers and calling it intelligence. A model can evaluate individual opportunities across a much larger pool of supply, apply advertiser-specific logic in real time and return its decision through the pipes buyers already use. 

The container does not make the model smart. It does not guarantee quality, transparency or fair pricing. It simply makes it possible to bring differentiated intelligence into infrastructure that used to run primarily on the platform owner’s intelligence. 

That may be the most exciting thing happening under the curation label. It is also why buyers need to stop treating the label as the answer. 

For buyers who want to extract real value from curation, here’s how I’d vet it: 

• Ask what’s built for you. Is a custom model being built and deployed for your brand, based on your goals?  

• Inspect what it selects. A partner should only make the plan if they can show you what you’re buying, explain why it fits your brand and report where your ads actually ran. 

• Expect ongoing optimization. Your model should learn from results and adjust selection and valuation over time. 

• Ask what they’re taking. Get the curator’s take rate disclosed, understand what it covers and ask for a cap. 

The technology is finally catching up to the original promise. We can make more specific decisions, for individual advertisers, across far more opportunities. And we’ve managed to bury that progress under a label that can also mean “a site list with a very ambitious markup.” 

I still think curation can be a big f*cking deal. But the companies that deserve to win should be able to finish the sentence, “We selected this impression, for this advertiser, for this reason.” 

“And here’s what we charged” would be a nice touch, too. 

This op-ed represents the views and opinions of the author and not of The Current, a division of The Trade Desk, or The Trade Desk. The appearance of the op-ed on The Current does not constitute an endorsement by The Current or The Trade Desk.