Ben Edelman Will Send You to Affiliate Jail

Google Builds a Side Door While the Judge Is Still Writing the Lock

Google shipped BuyerDirect inside Google Ad Manager (the tool publishers use to schedule and serve every impression), and Ari Paparo's read is that it quietly undermines the DOJ's antitrust remedy. The DOJ spent years fighting to spin out AdX (the auction where inventory clears) and left the ad server alone. BuyerDirect lets buyers transact through the ad server and skip the exchange entirely. If the judge does nothing, Google routes around whatever remedy lands.

That's a big "if." A defendant launching a bypass while final remedies are still being written is more likely to provoke a wider order than to sneak past one. Courts in the remedy phase are built to notice exactly this move. The argument that Google out-designed the DOJ assumes judicial passivity, and that's not the way to bet.

For Index Exchange, PubMatic, Magnite, and OpenX, the stakes are straightforward. They wanted a weakened AdX to open market space. They may get a technically divested but functionally irrelevant one instead, because spend follows the server, not the exchange.

The more important structural point sits underneath all of it. Owning both ends of the pipe is what lets you route around a remedy or assemble data nobody else can. FreeWheel is doing the same thing on the buy side: it's giving Beeswax customers (Beeswax is a demand-side bidding platform Comcast owns) series-level CTV reporting, telling buyers which specific show their connected-TV ad ran on. That data comes from the MRM ad server, not the bid request. A DSP that doesn't own an ad server cannot build that view. The margin advantage and the antitrust advantage are the same advantage, expressed in different places.

On fraud: Ben Edelman's Fia finding deserves more attention than it got. An invisible second browser tab fires its own affiliate link 100% of the time, even after detecting a competing one, stealing commission credit from whoever actually drove the sale. Merchants paid for traffic they never received. AppLovin's carrier "install helpers," apps with system privileges doing one-click installs outside the Play Store, manufactured installs advertisers were billed for. In both cases, last-touch attribution is what made the fraud legible. Multi-touch would have given the cheaters more places to hide.

On earnings: Meta posted $60.8 billion, up 27%, with ad prices up 12% and impressions up 14%, then guided soft and got punished on capex. Microsoft hit $90 billion, up 18%, with Azure near a $100 billion run rate. Paparo's structural observation is worth keeping: Google owns its chips (TPUs) and its model (Gemini); Microsoft resells someone else's silicon. Vertical integration is becoming the cost story across this entire business.

Our call: When the judge issues final remedies in the Google ad-tech case, the order will explicitly constrain Google Ad Manager or products like BuyerDirect, not just require divesting AdX. Remedy-phase judges punish live workarounds because signing a remedy the defendant has already defeated makes the court look outmaneuvered. The opposite outcome is possible, but it's the less likely one. Revisit by 2026-12-31: right if the final order names the ad server or direct-booking products among constrained assets; wrong if it's limited to AdX divestiture alone. Confidence: medium.

Your draft

Google shipped a product called BuyerDirect inside Google Ad Manager that lets ad buyers book publisher inventory directly, and Ari Paparo thinks it quietly guts the DOJ's whole antitrust theory. The rest of the episode is a tour of fraud mechanics and earnings, but the decision sitting under it is real: if the government spent years fighting to spin out the wrong box, the remedy that lands on your desk in 2027 may not change your economics at all.

What's actually being decided here isn't yours to decide. It's whether operators plan around a structural break in Google's ad stack, or plan around Google routing spend to a new front door and keeping the money. Type 1 for the DOJ, and hard to reverse once a judge signs off. Type 2 for you: cheap to hedge, cheap to wait. Forcing function is the final remedies ruling.

The Market Analyst. Paparo's read is that the DOJ asked for the ad exchange (AdX, the auction where inventory clears) to be spun out and left the ad server (Google Ad Manager, the tool publishers use to schedule and serve every impression) alone. BuyerDirect lets buyers transact through the ad server and skip the exchange entirely. If that holds, the remedy carves off a business Google can just route around. For a generalist: the government aimed at the cash register and Google is building a side door. The people who lose are anyone who bet their roadmap on a forced breakup creating open space. Index Exchange, PubMatic, Magnite, OpenX all wanted a weakened AdX. They may get a hollowed-out one instead.

The Skeptic. Slow down on the "DOJ missed the boat" victory lap. BuyerDirect is a product announcement, not a shipped, adopted, revenue-generating channel. Paparo himself floats that a judge could block it as part of final remedies, which means the whole argument depends on a court doing nothing. That is the load-bearing assumption, and it is weak. Judges in remedy phases are explicitly watching for exactly this move: defendant launches a workaround while sentencing is pending. In plain terms, Google building a bypass is as likely to enrage the judge as to outsmart the DOJ. Betting Google gets away with it is a bet on judicial passivity, and that is not the way to bet.

The Operator. Forget the courtroom. The thing that actually changes my Tuesday is FreeWheel giving Beeswax customers series-level CTV reporting, meaning I finally learn which show my connected-TV ad ran on. And it comes from the MRM ad server, not stuffed into the bid request. That is a real capability gap for everyone who does not own both the buy side and the sell side. If you are a DSP without an ad server, you cannot assemble that view, full stop. On fraud: Ben Edelman's Fia finding, an invisible second browser tab firing its own affiliate link 100% of the time even after detecting a competing one, is the kind of thing your logs already contain and nobody reads. The tooling to catch it exists. The will does not.

The Customer / End User. The publisher and the advertiser are the ones getting worked in these stories, and they mostly do not know it. Edelman's Fia scheme steals affiliate credit from whoever actually drove the sale, so the merchant pays a commission to a browser extension for traffic it never sent. The AppLovin forced-install work, carrier "install helpers" with system privileges doing one-click and countdown installs outside the Play Store, means advertisers paid for installs a user never chose. For an advertiser: you were billed for conversions that a piece of software manufactured. The uncomfortable part is that last-touch attribution, the crude model everyone loves to trash, is what let Edelman explain the fraud cleanly. Multi-touch would have given the cheaters more places to hide.

The CFO. The earnings prints set the weather. Meta did $60.8 billion, up 27%, with ad prices up 12% and impressions up 14%, then guided soft and got punished on capex and legal costs. Microsoft did $90 billion, up 18%, Azure near a $100 billion run rate. Paparo's structural point is the one to bank: Google owns its chips (TPUs) and its model (Gemini), while Microsoft resells someone else's silicon. Vertical integration is becoming the cost story in this business, on-chip and on-stack. That same logic runs straight back to BuyerDirect and FreeWheel: owning both ends of the pipe is what lets you route around a remedy or assemble reporting nobody else can. The margin advantage and the antitrust advantage are the same advantage.

The tensions worth sitting with. First, the Market Analyst and the Skeptic split hard on BuyerDirect: is it a checkmate or a provocation? One says Google out-designed the DOJ; the other says launching a bypass mid-remedy is how you hand the judge a reason to break up more, not less. Second, the Customer and the Operator disagree on attribution: last-touch is primitive and leaves money on the table, yet it is precisely what made the fraud legible. Better measurement and easier fraud detection are pulling in opposite directions.

What this hinges on: whether the final-remedies judge treats BuyerDirect as a legitimate product or as evidence Google will route around any narrow fix. Everything downstream, whether a breakup creates real open-market space or just moves Google's toll booth, flows from that one call. The council leans toward the Skeptic. A defendant shipping a workaround while the judge is still writing the remedy is the loudest possible signal that a narrow remedy fails, and courts in the remedy phase are built to notice. Before you plan around a weakened AdX, price in the scenario where the judge widens the remedy to include the ad server precisely because BuyerDirect proved the exchange spin-out was toothless.

Prediction: When the judge issues final remedies in the Google ad-tech case, the order will explicitly constrain Google Ad Manager (the ad server) or products like BuyerDirect, not just require divesting AdX, in direct response to Google's ability to route buyers around the exchange.

Confidence: Medium. Remedy-phase judges punish exactly this kind of pending-case workaround.

Why: Google launched BuyerDirect, a way for buyers to book inventory through the ad server and bypass the exchange the DOJ wants spun out, while final remedies are still being decided. The mechanism that connects that to my call is basic remedy law: a court crafting a fix watches for the defendant neutering it in real time, and a launched bypass is documentary proof that an AdX-only divestiture accomplishes nothing. The opposite outcome, the judge shrugging at a live workaround and signing a remedy the defendant has already defeated, is the less likely one because it would make the court look outmaneuvered, which judges avoid.

Revisit by 2026-12-31: We're right if the final remedy order names Google Ad Manager or direct-booking products among the constrained assets or conduct. We're wrong if the order is limited to AdX divestiture with no ad-server or BuyerDirect provisions.