News

Four DSPs now hold about 85% of programmatic spend—run AI bidding right

How to run AI bidding in a DSP: set a business outcome, load first-party data, give the model room to learn, then steer goals and audiences instead of bids.

AI bidding in a DSP means the platform prices every auction automatically against a goal you set—conversions, return on spend, viewable impressions—rather than a rate you enter by hand. Google Display & Video 360, The Trade Desk, Yahoo DSP, and Amazon Ads Agent all build AI or machine learning into campaign optimization, and together they held around 85% of global programmatic ad spend in Q1 2026, according to PPC Land's analysis of Guideline data. The machine prices every impression; the operator sets the goal and feeds it signal.

How to run it

  1. Pick the platform whose signal matches your goal. Amazon Ads Agent fits when reach across Amazon's own properties is relevant. The Trade Desk's AI analyzes up to 15 million ad opportunities each second across the open internet. DV360 applies machine learning to bidding and connects with Analytics 360 to reach specific audiences. Yahoo DSP's Blueprint AI engine guides performance throughout the platform, and its identity solutions work across addressable and non-addressable inventory.
  2. Set a business outcome, not a bid rate. Declare the objective you want—conversions, cost-per-action, awareness. One failure mode is choosing the wrong goal, or none at all, and then wondering why the AI optimizes toward cheap inventory rather than results.
  3. Give the model room to learn. A budget too thin early on starves the algorithm of auction data. Set a floor and ceiling—maximum CPM or target CPA—as guardrails, then leave the pricing alone while signal accumulates.
  4. Load your first-party data before activating. The Trade Desk's targeting uses first- and third-party data and its durable identity framework. In DV360, connecting Analytics 360 lets you reach the specific audiences you care about with greater precision. Yahoo DSP captures audiences with identity-backed precision. The AI is only as good as the signal you bring in.
  5. Activate automated bidding in the platform's optimization settings. In DV360 this is automated bidding, one of its campaign management features. In The Trade Desk, you enter your campaign inputs and its AI-driven decisioning helps optimize. In Yahoo DSP, the Blueprint AI engine guides performance throughout the platform. In Amazon Ads Agent, DVA+ offers simplified, outcome-based campaign creation, with advanced settings available when you need full control.
  6. Adjust goals and audiences, not bids. Once the learning period closes, the right lever is the objective or the targeting—not the bid itself. The AI reprices in response. Operators who keep reaching for manual bid adjustments are fighting the model rather than directing it.

Why this moment is different

Two of the four platforms now running 85% of global programmatic spend are moving the same direction: from AI that prices impressions to agents that run the campaign loop. Yahoo DSP's MCP servers accept plain-language requests to plan, activate, troubleshoot, and report, and Amazon Ads Agent's conversational experience helps plan, create, optimize, and measure campaigns through natural language—so on Yahoo DSP a workflow or an AI assistant, not just a human, can operate the platform. The Trade Desk's AI is built to work alongside operators.

The likely consequence is that bid pricing fades as a source of human competitive advantage. What remains is goal-setting, audience quality, and creative—the inputs the machine needs to do its work. Operators still running manual bids risk being slower not just than peers who have handed pricing to the model but than the agents those peers can now wire into the platform itself. Amazon went from under 10% to just under 20% of the Guideline dataset in roughly fifteen months. The consolidation has been rapid.

Operators who want to stay in the game connect first-party data, declare a real business goal, and activate automated bidding now—then watch where the model finds signal they missed. That is the work. The bidding itself is already done by the machine.

Get the Daily Frontier

Stories like this one land in your inbox every weekday morning. Subscribe free at factua.com/blog.

Know someone who should be reading this? Send them to factua.com/blog to subscribe.