In Defense of Defensibility

Full analysis

Adam Heimlich and Gareth Glaser spend an hour arguing one thing: whether defensibility in ad tech comes from custom, per-advertiser models or from pooled SaaS platforms, and they use that argument to defend Jeff Green while the market kicks him. The story for operators is what the whole conversation implies about where the moat sits, and who is exposed if Heimlich and Glaser are right.

Reversibility: This is a reading, not a decision. But the underlying bet, whether to build per-advertiser architecture or lean on pooled data, is Type 1. Hard to unwind once your data science org is committed one way.

What's actually being argued: That the SaaS valuation model VCs love does not fit enterprise ad tech, because real clients demand customization and service, and that the winners will build on the joint (one model per advertiser) rather than average everyone together. If that holds, it reprices a lot of "platform" stories.

Forcing function: The Trade Desk's weak quarter and the pile-on around it. That's the live catalyst Heimlich and Glaser are reacting to.

The Market Analyst takes the hit The Trade Desk just absorbed and reads Heimlich's defense as a claim about why the multiple should hold: DSPs are AI, and objectivity plus per-advertiser data hygiene is the moat. Fine as conviction, thin as evidence, and the episode admits it's light on hard data. The more interesting tell is the Sorrell line Glaser relays, that knowledge aggregation is dead as an agency moat and compute aggregation, buying cloud and AI in bulk and reselling to clients, is the new buying power but temporary. That's a warning to every holdco and every "platform" whose margin rests on being the middleman.

The Skeptic makes the case against per-advertiser models: they're expensive, they don't compound across clients, and they look a lot like a services business wearing a software valuation. Glaser basically concedes this when he says pure SaaS is a myth in enterprise. Good, but that cuts both ways. If customization is unavoidable, your gross margins are services margins, and no amount of "defensibility" language fixes that on an earnings call. Also note who's talking their book. Heimlich's "Google will feed publishers just enough traffic" forecast conveniently keeps publishers alive and dependent, which is exactly the world Chalice's publisher products need. Cite it as a view, not a fact.

The Operator starts with ARTF. Heimlich says PubMatic and Magnite have gone live with Automated Real-Time Feedback, a standard that lets buy-side signals flow back to the sell-side in real time, and he's calling on DSPs to follow. That's the Tuesday-morning question for an SSP or DSP product lead: do you integrate, and against whose spec? Unverified go-live, so verify before you reprioritize a roadmap. The second-order effect is the one worth watching. If ARTF real feedback loops become table stakes, the audit-chain rationale that props up multi-hop protocols like ADCP gets weaker, because everyone can pull their own logs. Glaser is blunt: three intermediaries paid per transaction, justified by an audit trail you could get out of anything. For non-specialists: Glaser is saying the middle layer charges a toll for a service the road doesn't need.

The CFO follows the fees. Buyers are reportedly paying around 5% of media budgets to measurement vendors like DoubleVerify and Nielsen, and Heimlich and Glaser note the grumpiness. If per-advertiser modeling actually outperforms pooled data on real conversions, the budget question becomes why keep paying verification and pooled-audience tolls at all. But building custom models per advertiser is a headcount line, not a software line. The payback only works if you can charge for outcomes, not seats. That's the whole tension in Glaser's SaaS-is-a-myth point, and it's the number a CFO should stress-test before believing the moat is real.

The Customer / End User is the brand that wants Coke and Pepsi served without their data mingling. Per-advertiser architecture is a genuine answer to that, and it's a real buying criterion, not a projection. Grether's Uber ad business gets held up as the template: bring an ad-industry operator in early and build the business around the advertiser's data, not the platform's. That's what large brands are actually asking for. Whether they'll pay services prices for it is the open question.

The tensions

Custom versus pooled is the spine. Heimlich says the joint wins on performance; the Skeptic and the CFO say the joint wins on performance and loses on margin, which is exactly why the market is repricing The Trade Desk. Both can be true. A better product with a worse valuation is still a worse stock.

ARTF versus ADCP is the second fault line. If real-time feedback standards spread, the audit-chain justification for multi-hop protocols collapses, and the Samba TV sales leader disputing Glaser on LinkedIn tells you the incumbents in that chain know it.

Amazon Fabric is the quiet one. Heimlich praises it as an interoperability marketplace and then flags concentration risk in the same breath. Everyone wants the plumbing until one company owns it.

What it hinges on: whether per-advertiser modeling's performance edge is large enough that advertisers pay a services-grade price for it. If yes, the pooled-SaaS platforms get squeezed and the fee-taking middle layers get thinner. If no, this is a well-argued podcast that changes nothing. Verify the ARTF go-live before you touch a roadmap, and treat Heimlich's Google-traffic forecast as a Chalice-shaped hope.

Prediction: In its next earnings report on 2026-11-06, The Trade Desk will again post year-over-year revenue growth below 20%, extending the slowdown Heimlich and Glaser are defending against.

Confidence: Medium. Two consecutive weak quarters and no named catalyst before the print.

Why: The episode exists because The Trade Desk just delivered another weak quarter and drew a pile-on, so the deceleration is already established, not speculative. The company's own defense, per Heimlich reciting Green, is architectural (DSPs are AI, objectivity is the moat) rather than a claim of imminent reacceleration, which is the language of a business managing through a slow patch, not exiting one. The agentic prompt interface both Heimlich and Glaser flag as the next step is a product roadmap item, not something that moves a Q4 revenue line by November. The opposite outcome, a snap back above 20%, would need a catalyst nobody on the episode names.

Revisit by 2026-11-06: We're right if The Trade Desk's Q3 2026 revenue growth prints under 20% year over year. We're wrong if it comes in at or above 20%.