Butler/Till and PubMatic run fully autonomous agentic CTV campaign

Butler/Till, PubMatic, and the DSP-Free CTV Buy

PubMatic (an SSP, or sell-side ad platform) and independent agency Butler/Till ran a CTV campaign for Geloso Beverage Group with no DSP in the loop. Butler/Till fed a brief into Claude, PubMatic's Agentic OS built and executed the campaign, and the two parties report supply-chain and tech costs down roughly 80%, 40% more impressions on the same budget, 98% video completion, and under 1% waste. The SSP just tried to eat the DSP's job.

Read the numbers carefully. PubMatic controlled the agent and the supply, then reported the results. When one platform routes the impressions and grades them, the highest-completion placements get the budget. That 98% completion rate is a measure of how PubMatic routed inventory, which tells you less than it seems about the campaign's actual performance. And the "fully autonomous" headline doesn't hold up either. Butler/Till Chief Strategy Officer Scott Ensign called it "heavily augmented by agentic technology" but "still directed by human media buyers." The agency kept its hand on the wheel, which is exactly right given that they own the client liability. So this is a strong proof of concept, not a lights-out autonomous system.

That said, the 80% cost number is real enough to do damage. The programmatic tax between brief and impression is genuinely fat, and that figure will sit on every holdco budget call this quarter. If agencies can match outcomes going direct to the SSP, the managed-service margin that trading desks and DSPs skim becomes very hard to defend at renewal. The first thing that breaks at scale: brand safety, frequency control across a portfolio, competitive separation, and the audit trail a compliance team actually requires. One clean campaign for one alcohol brand is not fifty concurrent campaigns with shared budgets and a legal team watching.

For The Trade Desk (the largest independent ad-buying platform), this is a direct poke at its mid-market CTV franchise. If independent agencies can replicate TTD outcomes without a DSP license, TTD's value proposition for non-premium buyers softens, and its own agent tooling starts looking reactive. For rival SSPs like Magnite and Index Exchange, the competitive logic is identical: PubMatic just claimed the framing of "the place you buy media" backed by an 80% cost headline and a marquee agency name. The natural-language layer calling existing supply APIs is not proprietary to PubMatic. Rivals cannot afford to let one platform own that pitch heading into upfront season.

There is also a quieter winner here: Anthropic, whose Claude model is the activation layer nobody voted for.

Before committing budget: run one campaign where a third party you pay holds the measurement and the inventory logs, with PubMatic having no role in either. If the completion rate holds when the platform isn't grading its own routing, the story is real.

Our call: By the end of Q1 2027, at least two of Magnite, Index Exchange, OpenX, or Yieldmo will publicly announce their own agentic direct-buy interface, letting advertisers or agencies brief campaigns without a DSP. Confidence is medium. The competitive incentive is obvious; the timing and the branding are not. The call is about who ships the pitch, not who proves the 80%.

PubMatic and independent agency Butler/Till ran a CTV campaign for Geloso Beverage Group with no DSP in the loop. Butler/Till typed a brief into Claude, PubMatic's Agentic OS built and ran the campaign, and the parties report supply-chain and tech costs down ~80%, 40% more impressions on the same budget, 98% video completion, and under 1% waste. What it means for operators: the SSP just tried to eat the DSP's job, and used an AI agent as the can opener.

Reversibility: Type 2 for any one operator testing it. Type 1 for the category if the pattern holds and DSP trading margin structurally re-rates.

What's actually being decided: Not "is agentic buying real." It's whether the buy-side is willing to hand campaign setup, strategy, and optimization to a platform that also owns the supply. That's a control-and-transparency question wearing an efficiency headline.

Forcing function: Q3 upfront season. Every holdco budget meeting this quarter now has an 80% number sitting on the table.

---

The Market Analyst: For a public-company reader, this reframes PubMatic's story. An SSP (the seller's plumbing) selling itself as "infrastructure for agentic media buying" is a different, richer pitch than "SSP that survived the auction wars." But one campaign, one brand, self-reported, does not move a stock's fundamentals. This is a direct poke at The Trade Desk's mid-market CTV franchise. If independents can match TTD's outcomes going direct to the SSP, TTD's managed value for non-premium buyers softens, and Kokai's agent gestures start looking defensive. Magnite and Index Exchange will not let PubMatic own this framing alone through the fall.

The Skeptic: Steelman the doubt, because it steelmans easily. This is a press release with a case-study costume. PubMatic controls the agent AND the supply, then reports a 98% completion rate. Of course it does. When one party routes the impressions and grades the impressions, the highest-completion placements get the budget. That figure reflects the routing. The campaign performance was never independently verified. And the "no DSP" line oversells: the agent is running DSP-equivalent logic inside PubMatic's stack. The demand-side function did not vanish. The SSP absorbed it, and now sees the advertiser's strategy at a depth no DSP ever gave it. For a non-specialist: the seller just became the buyer's shopping assistant, and it kept the receipts.

The Operator: The 80% cost cut is what circulates, and it's aimed straight at trading-desk headcount. If a Claude brief stands up a CTV campaign without a trader touching a DSP console, the managed-service margin agencies and DSPs skim gets very hard to defend in a renewal conversation. What breaks first Tuesday morning: brand safety and frequency control across a portfolio. One autonomous campaign for one alcohol brand is clean. Fifty concurrent campaigns, shared budgets, competitive separation, dayparting rules, and a compliance team that wants an audit trail is a different animal. The 90-day second-order effect: DSP account teams start missing RFP invitations from Butler/Till-tier independents, and every rival SSP ships an agentic-layer announcement before the numbers are reproduced.

The Customer / End User (the advertiser and its agency): Read Scott Ensign's own framing, because it's more revealing than the headline. Butler/Till's chief strategy officer called this "heavily augmented by agentic technology" but "still directed by human media buyers." That is not "fully autonomous," whatever the press summary says. The agency kept its hand on the wheel, which is exactly right, because the agency owns the client relationship and the liability. What the advertiser actually wants: cheaper activation without surrendering the ability to say why a dollar went where it went. Geloso got a great demo. The open question for any brand is whether it will let a supply-owning platform pick its media with no independent referee on the outcomes.

The CFO: The 80% is a supply-chain-and-tech-cost figure, and it matters, because the programmatic tax between brief and impression is genuinely fat. But the payback math has a hidden line: what does it cost to lose independent measurement and negotiating leverage? When the SSP sets strategy, runs the auction, and reports the result, you've traded a fee for a conflict of interest. That can absolutely be worth it at pilot scale. At portfolio scale, the CFO question is whether the savings survive once you re-add the third-party verification, the audit tooling, and the human oversight that a one-off proof-of-concept skipped.

---

The tensions

1. Efficiency vs. the fox guarding the henhouse. The Operator and CFO see a real cost cut worth chasing. The Skeptic sees a platform grading its own homework. Both are right, and the reader's own risk tolerance decides which dominates.

2. "Fully autonomous" vs. what actually happened. The press summary says no human intervention. Ensign says human-directed. That gap is the whole story. The buy-side kept control on purpose, which shows where the real ceiling on autonomy sits. It is accountability, not a technical barrier.

3. Who owns the moat now. The win goes to whoever has the best agent interface, and if that framing holds, PubMatic's edge is API reliability and being inside Claude. Anthropic quietly becomes the activation layer nobody voted for.

What this hinges on

Three beliefs. First, whether the 80% and 98% survive an independent, third-party-measured campaign the platform did not route end to end. Second, whether agencies will cede strategy to a supply-owning platform once real client liability is attached, or keep the human in the loop like Butler/Till did. Third, whether rival SSPs and TTD move fast enough that "agentic buying" becomes table stakes rather than PubMatic's differentiator.

The council leans skeptical on the specific numbers and bullish on the direction. The disintermediation pressure is real. The clean version reported here is not what portfolio-scale execution looks like.

Before committing: run one campaign where a third party you pay holds the measurement and the inventory logs, with the platform having no role in either. If the completion rate holds when PubMatic isn't grading its own routing, the story is real. If it doesn't, you learned what the demo was hiding for the cost of one flight.

---

Prediction: By the end of Q1 2027 earnings season (roughly March 2027), at least two of PubMatic's SSP peers among Magnite, Index Exchange, OpenX, and Yieldmo will publicly announce their own agentic buy-side interface that lets advertisers or agencies brief campaigns directly, cutting the DSP out of the workflow.

Confidence: Medium. The competitive incentive is obvious, but timing and how loudly they brand it are not certain.

Why: PubMatic just claimed a narrative that reframes the SSP from "auction plumbing" to "the place you buy media," and it landed a marquee independent agency and an 80% cost headline to sell it. That is exactly the kind of positioning rival SSPs cannot afford to let one player own heading into upfront season, because the pitch to mid-market buyers is identical for all of them and the tech (a natural-language layer calling their existing supply APIs) is not proprietary to PubMatic. The opposite outcome, silence from the rest of the SSP field, would require every competitor to concede the framing while their sales teams get asked about it in every RFP, which is not how this category has ever behaved. The gap between announcement and reproducible results at scale is wide, which is why the call is about who ships the pitch, not who proves the 80%.

Revisit by 2027-03-31: We're right if two or more of Magnite, Index Exchange, OpenX, or Yieldmo announce an agentic direct-buy interface by then. We're wrong if at most one does, or if the category stays framed around DSP-side agent tooling.