Is It Recipe Sites? Or Is It Cake?

Full analysis

Adam Heimlich and Gareth Glaser spent an hour arguing that programmatic advertising can compress its fees enough to fight Meta and Google for mid-market performance budgets, that incrementality measurement is knowable but nobody buys it, and that retail media data mostly helps challengers conquest rather than incumbents grow. The implication for operators: the take rate that funds most of this industry is under structural pressure, and the measurement that would justify it is a product almost nobody wants to pay for.

This is analysis, not news. No deal, no print, no regulator. So the question is whether the structural read holds, and who should reprice their roadmap if it does.

Reversibility: N/A. This is a briefing on a worldview, not a decision. But the fee-compression thesis, if right, is a Type 1 (hard to reverse) hit to SSP and DSP economics.

What's actually being decided: Whether independent programmatic has a viable path back to mid-market performance spend, and what an operator should build or defend against on the way there.

Forcing function: Heimlich names it himself. A recession is the event that pushes CFOs to demand incremental measurement over platform-reported metrics.

The Market Analyst. Andrew Casale is on record saying SSP fees go to 6%. Heimlich stacks buy-side tech at 6% on top and lands at a 15 to 18% total take rate, down from the 25 to 30% blended rate they both cite today. For a smart generalist: the "take rate" is the cut every middleman keeps out of a dollar an advertiser spends before it reaches the publisher. Cut that from 30 cents to 15 cents and two things happen at once. Programmatic becomes cheap enough to chase performance budgets it can't touch today, and every SSP and DSP living on the old cut loses roughly half its margin per dollar. You cannot have the first without the second. The vendors cheering fee compression as a growth story are describing their own margin funeral.

The Skeptic. The whole efficiency case rests on advertisers actually wanting incremental measurement. Heimlich admits the opposite in the same breath. Getting a client to pay for PSA control inventory, the clean way to measure whether an ad caused a sale, is "one of the hardest cells" to sell. If nobody will fund the measurement, the whole efficiency case is a slide, not a supply source. And notice the incentive: Heimlich runs Chalice, which sells propensity scoring into exactly this narrative. A world where independent programmatic out-measures the walled gardens is a world where Chalice wins. That does not make him wrong. It means the thesis is coming from a seller of the thesis.

The Operator. Try to run the clean measurement Tuesday morning. Ghost bidding, where the platform suppresses a winning bid inside its own pipes to build a control group, gives you dirty exposed-versus-unexposed data because you never see how the plumbing worked. PSA placeholders are clean but the client won't pay for the empty inventory. So the practitioner is stuck choosing between opaque-and-free and clean-and-unfundable. Meanwhile the real 90-day break shows up on the supply side: a luxury brand killed a campaign that was beating CTR and sell-through benchmarks purely on brand-safety grounds. The measurement said it worked. The brand-safety rule overrode the measurement. That is the actual daily reality, and no fee-compression math fixes it.

The Customer / End User. Here the customer split matters. The challenger brand wins from Heimlich's retail media read: most retail data is loyal, repeat-buyer data, so it's a conquesting weapon for the upstart cereal brand targeting Kellogg's buyers, not a loyalty tool for Kellogg. So the incumbent CPG advertiser paying retail-network premiums for "their own" shoppers is largely buying people who'd have bought anyway. For a generalist: they're paying to advertise to customers already walking to the register. The mid-market performance advertiser is the customer everyone claims to want, and the honest answer from this episode is that today they're on AppLovin and Meta because that's where short-form video performance lives.

The CFO. The CMO cites the platform's own dashboard. The CFO cites Marketing Mix Modeling, the statistical read on what ads actually moved sales. Those two numbers disagree, and in a downturn the CFO wins that fight. That's Heimlich's real forcing function, and it's the one durable claim in the hour. But note what it implies for a vendor P&L: if the CFO wins, spend flows to whoever can prove incrementality cheaply, and away from whoever's value was the reported-metric story. Google's rumored flat-10% buy-side product is terrifying precisely because it's media-efficient at a clean, legible price. Legible beats clever when the CFO holds the pen.

The tensions.

Heimlich against himself is the biggest one. Fee compression makes programmatic viable for the mid-market, but the measurement that proves the value is the exact product he admits clients won't fund. The efficiency case and the "nobody buys measurement" case cannot both be true at once.

The Market Analyst against the industry's cheerleaders. Everyone quotes Casale's 6% as good news. It's good news for advertisers and a margin halving for the vendors saying it out loud.

The Operator against everyone. Even perfect measurement loses to a brand-safety veto that kills a winning campaign. The bottleneck isn't math. It's who has authority to overrule the math.

Synthesis. This hinges on three beliefs. One: that fees actually compress to the mid-teens rather than staying sticky at 25 to 30% because incumbents defend margin. Two: that a recession forces CFOs to demand incrementality at scale. Three: that clean measurement becomes buyable rather than staying the hardest cell on the rate card. The council leans toward belief two being real and beliefs one and three being aspirational. CFO power in a downturn is a known pattern. Fee compression to 15% is a forecast from a party who profits from it.

What to de-risk before you build to this: whether your own clients will fund PSA control cells at any price. If they won't, the whole efficiency narrative is theory. Run one paid incrementality cell with a real budget this quarter. If you can't sell it, you have your answer about the mid-market thesis.

Prediction: Blended programmatic take rates will still sit at or above 20% at the end of Q1 2027 earnings season, not the 15 to 18% Heimlich forecasts, because no major SSP or DSP will publicly cut its fee schedule to 6% within that window.

Confidence: Medium. Incumbents defend margin, and the 6% call comes from sellers of the thesis.

Why: The compression case rests on Casale's on-record 6% SSP claim plus a matching 6% buy-side assumption, but a public claim about where fees are "going" is not a rate card, and the people forecasting it either benefit from the narrative or are describing their own margin cut. Vendors do not volunteer to halve their own take without a competitive gun to their head, and the only gun in the room, Google's rumored flat-10% product, is not yet shipping at scale. The opposite outcome, sudden compression to the mid-teens, would require multiple public companies to announce fee cuts against their own P&L inside two quarters, which nothing in this episode shows is underway.

Revisit by 2027-05-15: We're right if no top-five SSP or DSP has published a headline take rate at or below 6%, and blended programmatic fees remain at or above 20% per third-party estimates. We're wrong if two or more major platforms cut disclosed fees to the 6% range and blended take drops below 18%.