Podcast episode

Is It Recipe Sites? Or Is It Cake?

by Ken Rona ·July 27, 2026 ·AdTalk

attribution measurement programmatic publisher-economics ssp

TL;DR

Hosts Adam (buy-side, Chalice) and Gareth (sell-side, Gamera) run through a wide-ranging practitioner session covering incrementality measurement, the bull/bear cases for programmatic economics, SSP-driven format innovation, and publisher distress. The episode is dense with real operational detail — useful for anyone thinking about programmatic efficiency, retail media limitations, or the mid-market opportunity — but light on breaking news.

What was covered

User IDs as correlation fields, not buckets. Adam explained how Chalice uses LiveRamp and other identity systems (RampIDs, The Trade Desk's UID, Experian IDs) not as rigid audience segments but as inputs into per-advertiser propensity models that score for incremental customer likelihood. The platform approach of treating IDs as fixed segments is contrasted as fundamentally inferior.
Retail media networks and the conquesting dilemma. Adam argued that retail media network (RMN) data is predominantly loyal-repeat-buyer data, making it most valuable for competitor conquest — a brand like Hyundai buying Toyota-buyer data — rather than for the incumbent brand's own retention campaigns. Top 2,000 advertisers account for roughly 40% of digital spend, framing this as an enterprise-scale problem, not a niche.
PSA control groups vs. ghost bidding. Adam and Gareth debated the best way to measure incrementality. Running a clean PSA (Public Service Announcement — a blank placeholder ad shown to the unexposed control group) produces clean exposed/unexposed data; ghost bidding (simulating a bid without actually entering the auction, to build a synthetic control) produces messy, hard-to-trust data. Despite PSA being simpler and cleaner, most advertisers resist it to avoid learning that a large share of conversions would have happened anyway.
CFO power and MMM as a forcing function. Adam noted a structural shift: CFOs increasingly have access to Marketing Mix Modeling (MMM — statistical modeling that attributes sales to media channels) and are using it to challenge CMO-reported media metrics. He predicted a recession would accelerate this tension and drive more rigorous incrementality measurement.
Bull/bear cases for programmatic economics. Gareth's bear case: new AI-middleware vendors re-complicate the stack the way old middleware did, and Google Cloud locks enterprise buyers into its closed ecosystem (already Campaign Manager 360 + GA4 + YouTube + search). Adam's bull case: programmatic tech fees compress from ~25–30% today to 15–18% (citing Andrew Casale of Index Exchange on record saying fees could go to 6% at the SSP layer), making programmatic competitive enough to crack mid-market performance advertisers currently spending exclusively on Meta — described as "the white whale."
SSP format innovation and native ad mechanics. Gareth explained that true programmatic native (as defined in the OpenRTB spec) delivers a structured array of components — image, title, subtitle, description — rather than a pre-built banner tag; the publisher's page assembles them. SSPs with code-on-page (like Media.net, Triplelift, Magnite/former RTK) are best positioned to drive format creation because they already have direct publisher relationships and early page load presence. Gareth noted a distinction between true native and "fake native" — a creative styled to look native but served in a standard 300×250 slot.
Publisher Q2 traffic squeeze and irrational brand-safety blocking. Adam reported that publisher ad supply reportedly fell 40% in Q2, attributed to Google retaining users inside search. He questioned the methodology (noting bid-stream request counts are an unreliable proxy for actual impressions). A specific case study: a luxury brand running a campaign that was performing well on news publisher inventory — above-average click-through and sell-through rates — killed the news URLs anyway on brand-safety grounds, with no data-based rationale.
Sam Altman's rumored ad network. The hosts closed with a brief critique of a reported claim that Sam Altman's AI ad network would generate $100 billion in revenue, noting the current market for that category is approximately $5 billion annually and his network is reportedly making roughly $1 billion — calling the framing a significant expectation-setting misstep.

Notable claims & predictions

Adam: "N-state could be 15 [percent total programmatic fees] — that's possible. The SSP going to six and you just need algorithm six and buy-side tech six." A compression from today's estimated 25–30% stack cost to 15–18% total would meaningfully change the mid-market economics of programmatic.
Adam: "The top 2,000 advertisers spend 40% of the digital money. So it's not a small segment. If you get to the top 5,000 advertisers, it's already a majority." Framing enterprise incrementality problems as the bulk of the market, not a niche.
Adam (paraphrasing Andrew Casale of Index Exchange): "Casale is on record — Casale said we're going to get chased down to 6% [SSP take rate]." Attributed to Casale directly.
Gareth (bear case): "Google Cloud is a very aggressive third place in cloud and their advantage is all the enterprises already spend a billion dollars with Google... it almost forces [agencies] to build everything on Google Cloud." The lock-in risk from Google's bundled ecosystem (GA4 + CM360 + YouTube + Cloud) is the central bear case for independent programmatic.
Adam: "Remarketing and lookalike targeting are the least incremental things we do and it's like 80% of the budget — and prospecting has the highest CPA and is always questioned — and that's the most incremental." A structural indictment of how most programmatic budgets are allocated.

Fact check

Gareth attributed to Adam/show: "We've probably used around 30% [total programmatic fees], 25 to 30% going to your various media vendors." This is a commonly cited industry range and is broadly consistent with third-party analyses of the programmatic supply chain. Unverified as a precise current figure — estimates vary widely by format, deal type, and vendor mix — but not contradicted by well-established fact.
On media.net: Adam described media.net as originally a "big search arbitrage business," a Bing and Yahoo partner, sold for "$800 million the first time." The $800 million sale figure (to a Chinese consortium, circa 2016) is consistent with widely reported public accounts of that transaction. No basis to challenge. The claim that media.net was "recently bought back by the old founder" is stated as current fact — this is unverified in the transcript with no date given; the show does not establish when this occurred.
Adam on Sam Altman's ad network: "The current market is five billion [dollars] and he was going to make a hundred billion." The $5 billion market figure and ~$1 billion current revenue figure are stated without sourcing in the transcript. These are unverified — the hosts appear to be summarizing press coverage they recall but cite no specific report. Listeners should treat both numbers as approximate recollections, not cited data.
Adam, on Google retaining search users reducing publisher traffic: The claim that publisher ad supply "fell 40% in Q2" is attributed to an unnamed article and Gareth himself questions the methodology, noting bid-stream request counts are unreliable. Contested internally by the hosts themselves — treat as an unverified industry data point, not a confirmed figure.
Incentive flag — Adam on Chalice: Adam is co-host and represents Chalice, whose core value proposition is precisely the incrementality-based, propensity-scoring approach he describes as superior to platform targeting. His characterization of platforms as unable to distinguish new from repeat customers, and of PSA-based control groups as clearly superior to ghost bidding, directly supports Chalice's commercial positioning. The underlying logic is sound and widely shared by practitioners, but listeners should note he is talking his own book throughout the incrementality discussion.

Full analysis

Adam Greco of Chalice and Gareth Holmes of Gamera spent an hour on the practitioner plumbing of programmatic: whether tech fees can compress enough to make the open web competitive with Meta, whether anyone actually wants clean incrementality data, and where publishers stand as Google keeps users inside search. No breaking news. But the through-line matters to every operator: the economics of the open web and the honesty of measurement are on a collision course with the CFO.

Reversibility: Mostly Type 2 for any single operator (you can change vendors, change measurement approach). The industry-level shift toward incrementality and fee compression is Type 1 and slow-moving. What's actually being decided: whether the open web can win back mid-market performance budgets that live on Meta today, and whether buyers will tolerate measurement that tells them the truth. Forcing function: the next recession, and CFOs armed with marketing mix models.

The Market Analyst — The bull case here is a fee story. Adam argues total programmatic take drops from roughly 25 to 30 percent today to 15 to 18 percent, and cites Andrew Casale of Index Exchange saying SSP take rates get "chased down to 6 percent." If that happens, the open web gets cheap enough to compete with Meta for the mid-market performance buyer, the "white whale." That's the whole independent-adtech thesis in one number. For a non-specialist: the middlemen between advertiser and publisher get paid a lot less, which makes buying ads on regular websites worth doing again. The bear case is Gareth's: Google Cloud already collects a billion dollars a year from big enterprises and bundles GA4, Campaign Manager, YouTube, and search, so agencies get pulled into building everything on Google anyway. Fee compression helps the independents; bundling lock-in helps Google. Both can be true.

The Skeptic — The load-bearing assumption is that anyone wants clean incrementality. They don't. Adam says it himself: remarketing and lookalikes are the least incremental things we do, and they eat 80 percent of the budget, while prospecting, the most incremental, gets questioned constantly. A clean PSA control group (a blank placeholder shown to the unexposed group) is simpler and cleaner than ghost bidding, yet advertisers resist it because it reveals how many conversions would have happened anyway. So the demand for honesty is theoretical. In plain terms: a lot of ad budgets survive precisely because nobody measures them properly. Fee compression to 15 percent won't matter if buyers keep spending to protect their own numbers.

The Operator — Two things here are usable Tuesday morning. First, the native format point: true programmatic native ships a structured set of parts (image, title, description) that the page assembles, and the SSPs with code on the page (Gareth names Media.net, Triplelift, and Magnite via the old RTK tech) are the ones positioned to build new formats, because they're already on the page early. If you run a publisher or an SSP, that's a real advantage to press. Second, the luxury brand that killed news URLs despite above-average click-through and sell-through, on brand-safety grounds, with no data behind it. That happens every day. Somebody's brand-safety policy is quietly torching your best-performing inventory, and nobody's checking the numbers.

The CFO — Adam's structural call is the one I'd bet on: CFOs now have marketing mix models, and they're using them to challenge what the CMO reports. That's a power shift inside the building. When the CFO can independently model what media actually drove sales, the CMO's attribution deck stops being the last word. A recession accelerates it. For the ecosystem, that means the measurement vendors who can survive CFO scrutiny win, and the ones selling flattering last-touch numbers get exposed. In plain terms: the person who controls the money is getting their own scoreboard, and it doesn't match the marketing team's.

The Customer / End User — The mid-market performance advertiser is the customer everyone's circling, and the honest question is whether they care about any of this. They're on Meta because it works and it's one login. Telling them the open web's fees dropped to 15 percent is not a pitch they've asked for. Retail media has the same gap: Adam's point that retail data is mostly loyal-repeat-buyer data, best for conquesting a competitor's customers rather than retaining your own, is sharp and mostly unheard by the brands buying it. The buyer isn't demanding incrementality. Someone has to make them want it.

Where they part ways. The Market Analyst thinks cheaper fees crack the mid-market open. The Skeptic says price was never the blocker; buyers protecting their own numbers is. That's the real tension: is Meta's grip about economics or about accountability? If it's economics, compression wins budgets back. If it's that Meta lets buyers avoid hard questions, cheaper open-web fees change nothing.

Second tension: the CFO and the Customer disagree about who's driving. The CFO says finance forces rigor from above. The Customer says the marketer still picks the channel and still prefers the easy Meta login. Whether incrementality wins depends on which one actually holds the pen.

What it hinges on. One belief: does honest measurement create demand, or destroy it? If a recession makes CFOs demand incrementality, the independents with real propensity and control-group methods win, and fee compression gives them room to compete. If buyers keep hiding in remarketing to protect their dashboards, the whole open-web-revival thesis stalls regardless of fees. Before betting a roadmap on it, watch two things: whether SSP take rates actually move toward Casale's 6 percent in published rate cards, and whether any large advertiser publicly adopts PSA control groups instead of quietly avoiding them.

Prediction: No independent SSP will publicly cut its standard take rate to 6 percent, or near it, by the time of Index Exchange's and Magnite's Q2 2027 earnings and rate-card updates; published SSP take rates stay in the mid-to-high teens or higher.

Confidence: Medium — Fee cuts get talked up in interviews, not printed on rate cards.

Why: Casale floated 6 percent as a direction of travel, and Adam repeated it as the bull case, but talking down your own take rate in a podcast costs nothing while actually printing it costs revenue. SSPs make money on that spread, and there's no competitive force yet strong enough to make one voluntarily lead the whole category to 6 percent. The mechanism that would force it, mass buyer defection to a cheaper path, hasn't happened, because as the Skeptic notes buyers aren't optimizing on fees anyway. The opposite outcome, a real published cut to single digits within a year, would require a price war nobody has started.

Revisit by 2027-08-15: We're right if the major independent SSPs' disclosed or reported standard take rates remain in the mid-teens or higher through their Q2 2027 reporting. We're wrong if any top-tier independent SSP publishes or confirms a standard take rate at or near 6 percent in that window.

Listen to the episode →