I listened to two of the largest demand-side platforms (DSPs) in programmatic advertising describe the same broken supply chain in the same month, at the two different events, using nearly the same language. This is what it sounded like.
The Duplication Disconnect
When an advertiser bids on a programmatic ad impression, that single impression often arrives at the DSP through multiple supply paths simultaneously. The same ad slot for the same user gets duplicated across SSPs, resellers, and intermediary nodes. If a DSP bids on both the original and its clone, it can't actually win both. One is real inventory, the other is virtual. The DSP wastes compute, the advertiser wastes bid evaluations, and the SSPs generating duplicates capture fees on traffic that never converts to a purchase.
Dr. Neil Richter, Amazon's Vice President of Advertising Science, called duplicated bid requests "replicants." Mike O'Sullivan, The Trade Desk's Vice President of Product, abbreviated the problem as "DoSOL" (duplication, obfuscation, and sometimes lies). Both sat across from Jounce Media's Chris Kane in this past March. Both agreed the programmatic pipes are clogged with virtual inventory no buyer can actually purchase. Both want to fix it.
What struck me is how differently they see the solution.
Amazon Shares the Algorithm
Richter's premise is simple. Every ad impression can reach a buyer through dozens of paths, and most of those paths are wasteful. Amazon DSP scores each one on marginal value to the advertiser's campaign, the same way baseball's wins above replacement (WAR) metric scores a player on what they add beyond a freely available substitute. The scoring shifts per campaign. One path wins on return on ad spend (ROAS). Another wins on unique reach. The math decides, not the relationship.
The tool that operationalizes this is Dynamic Traffic Engine (DTE), an open-source SDK on GitHub that any sell-side platform (SSP) can install on its own servers. Amazon publishes the scoring model and the SSP applies it locally.
The part worth noting is that Dr. Richter explicitly told Kane that if the highest-utility path doesn't route through Amazon DSP, that's the path the platform takes. The best SSPs are already in single-digit waste percentages. Amazon wants to bring the rest of the ecosystem closer to that benchmark by giving the math away.
Trade Desk Builds the Telescope
O'Sullivan described a different instrument. OpenPath, now four years old with more than 200 publisher integrations representing 5 to 10 percent of partner revenue, functions as what he called a "canary in the coal mine." Open Ads, the next-generation wrapper, extends that canary closer to the publisher's primary ad server. The purpose is verification so that when Trade Desk buys through its own direct path, it establishes a control signal against which every other supply chain can be measured. Where auction integrity is higher, TTD bids more.
Kane pressed on whether publishers need Open Ads to get full access to TTD demand. O'Sullivan's answer was careful but directional: where trust is greater, bids are higher.
The Scorecard Arrives on Cue
Forrester released its first Omnichannel Advertising Platforms Wave earlier this week. Amazon Ads landed as the clear Leader. Trade Desk fell outside the top three. On agentic AI, the capability most closely mapped to cooperative routing infrastructure, Amazon scored a 5. Trade Desk scored a 1.
🆘 SOS Insight: Amazon gives the algorithm away to make itself indispensable. Trade Desk builds integrations it controls to make itself indispensable. SSPs lose margin either way, but through completely different mechanisms.
Richter was asked onstage about Jeff Green's public claim that Amazon won't have a DSP in five years. His response was that he'd spent the last five years upgrading it, and Forrester had just named it a market leader on execution and strategy.
