PubMatic Touts 80+ Agentic Ad Campaigns, Predicts 25% Agentic Trading by 2028

Full analysis

Rajeev Goel, PubMatic's CEO, says the company has run more than 80 "agentic" ad campaigns, where AI agents make the media buying decisions on their own, and that by the end of 2028 roughly a quarter of PubMatic's business will trade that way. The campaigns touched all five holding companies: Omnicom, WPP, Publicis, Dentsu, and Interpublic. The same thesis props up the hire of Megan Ramm, built to scale those relationships, and PubMatic's Activate product, which lets buyers hit inventory directly.

What's actually being decided here isn't PubMatic's roadmap. It's whether the DSP fee layer survives when an AI agent can buy straight from the supply side. That's the question every operator in this chain has to answer.

This is easy to undo for any one operator. Running an agentic pilot costs little and can be shut off Monday. The thing that's hard to undo is the stack decision underneath it: if a holdco rebuilds its trading workflow around agents hitting SSPs directly, the DSP relationship doesn't come back cheaply. No hard deadline. The 2028 date is a forecast, not a clock.

The Skeptic. Eighty campaigns is a count, not a result. Goel gave us no average spend, no threshold crossed, no definition of "agentic" that separates a real autonomous bidder from a GPT wrapper firing a pre-negotiated private deal. "Agentic campaign with all five holdcos" could mean five innovation-team experiments with a human trader approving every move. For a non-specialist: the headline says robots are buying ads, but we can't tell if the robot decided anything or just clicked a button a human already set up. The 25%-by-2028 call has no accountability. Goel will be three forecasts past it before anyone checks. Every SSP is saying "AI-native" this quarter. The word is free.

The Strategist. Strip the buzzwords and this is an attack on the DSP's reason to exist. A demand-side platform earns its margin on audience targeting, bidding logic, and reporting. If an agent carries that logic itself and PubMatic's Activate supplies the inventory, the DSP becomes an optional toll. The 2028 window is believable for one reason: three to four years is about how long it takes a holdco trading desk to rebuild its daily workflow around new tooling. For a non-specialist: the company that sells ad space is trying to make the company that places the buys unnecessary. The risk in this read is that it's too clean. Infrastructure transitions in ad-tech almost never arrive on the tidy timeline the slide promises.

The Operator. The pressure lands on The Trade Desk and Google's DV360 first. If Activate is quietly routing agent buys SSP-direct, a programmatic director at a holdco gets an RFP from her own innovation team asking why she's paying DSP fees when the agent can hit the SSP. That question is uncomfortable and it won't go away. For a non-specialist: the middleman's cut starts looking optional in the test budget. But here's what breaks at 90 days: brand safety sign-off, legal review, and budget authorization are still human workflows. An agent that clears auction logic in milliseconds still waits two weeks for a brand safety team to approve the inventory list. The plumbing already works. The approvals are the bottleneck.

The Market Analyst. PubMatic's stock has been marked down on the fear that open-web programmatic is losing relevance. The agentic story is a re-rating pitch: it hands sell-side analysts a new revenue line to model instead of declining CPM yields. "Multiple compression" just means investors stopped paying up for the growth story, and Goel is trying to give them one back. Magnite is the direct comparison, the other large independent sell-side platform. If PubMatic lands one credible public case study showing an agentic direct-buy displacing DSP spend, Magnite has to answer within a quarter or cede the narrative. The wildcard is Amazon, which owns both a DSP and the agent infrastructure and could run both sides of this trade without choosing.

The CFO. Count the real economics. The DSP fee that an agent routes around is somebody's revenue, and that somebody is a customer or partner of nearly everyone reading this. If 25% of PubMatic's flow goes agent-direct, that's margin lifted out of the DSP layer, but it's also a bet PubMatic absorbs the cost of building and supporting agent-accessible infrastructure for buyers who currently pay nothing for it. The payback question: does direct agentic flow carry a better take rate than DSP-mediated flow, or is PubMatic trading a known fee stream for a cheaper, higher-volume one? Goel hasn't said. Until he does, 25% is a volume forecast with no margin attached.

Where the council splits

Three real disagreements.

The Skeptic says 80 campaigns is noise until we see spend and a definition. The Strategist says the number barely matters because the direction is right regardless. They're arguing about whether to trust the evidence or the thesis. Both can be correct: the campaigns can be small and theatrical today and the DSP disintermediation still real by 2028.

The Operator and the Strategist disagree on what blocks this. The Strategist thinks workflow rebuild is the gate, three to four years of holdco retooling. The Operator thinks the gate is approvals: brand safety, legal, budget sign-off, none of which agents touch. If the Operator is right, the 25% number stalls on the compliance desk, not on the technology.

The Market Analyst and the CFO disagree on what success looks like. The Analyst wants one public case study to move the story. The CFO wants the take rate on agentic flow versus DSP-mediated flow. A win on the narrative and a loss on the margin are both possible at once.

What it hinges on

Two things decide this, and both are missing from Goel's disclosure. First, whether "agentic" means genuine autonomous decisioning or a human-approved private deal with an AI label on it. Second, whether agent-direct flow pays PubMatic a better take rate than routing through a DSP. Without the first, there's no disruption. Without the second, there's no reason for PubMatic to want it.

The council leans skeptical on the 25% figure and credible on the direction. DSP disintermediation by agents is a real threat worth preparing for. PubMatic hitting a quarter of its volume on it by 2028 is a sales forecast, not a plan.

Before anyone reorganizes a stack around this: demand a case study with named spend and a clear line between DSP-mediated and SSP-direct agentic buys. That distinction tells you whether this is incremental or cannibalistic to your current relationships.

Prediction: PubMatic's full-year 2028 revenue, reported in February 2029, will not show 25% of its business trading agentically as a disclosed, audited line, and no independent measurement will confirm the figure by then.

Confidence: Medium. The forecast has no accountability mechanism and the approval workflows agents must clear are unsolved.

Why: Goel's own disclosure gives no spend figures, no definition separating autonomous decisioning from human-approved private deals, and no take-rate math, which is the pattern of a narrative built for analysts rather than a tracked operating metric. Brand safety sign-off, legal review, and budget authorization remain human steps that no agent currently clears at scale, and those approval workflows are the real gate on agentic trading reaching 25% of volume. For 25% to be real and provable, PubMatic would have to define the metric precisely and let it be audited, and a company that floated the number this loosely has every incentive to keep the definition loose so a smaller real figure still reads as validation. The opposite outcome, a cleanly audited 25%, would require both the compliance workflows to collapse and PubMatic to volunteer a hard definition it has so far avoided.

Revisit by 2029-02-28: We're right if PubMatic's 2028 results and any independent ad-tech measurement show no audited 25% agentic-trading figure. We're wrong if PubMatic reports, or a third party confirms, that roughly a quarter of its flow traded agentically under a stated definition.

The 25% figure is unfalsifiable by design. The direction is real. It becomes a metric worth tracking the day Goel attaches a definition and a take rate to it.