The Trade Desk just made AI media buying more operational. On August 27, the demand-side platform introduced Kokai Zuma, a redesigned release that puts agentic AI, forecasting, and measurement into the daily workflow of programmatic buyers. The company says early modeling results produced an average 32% improvement in cost-per-acquisition performance.

That headline deserves context: the figure comes from a vendor analysis of 62 campaigns comparing model versions, not a universal guarantee. But the strategic signal is still important. AI is moving from a reporting assistant to the layer that forecasts inventory, changes campaign settings, and helps buyers act across a fragmented media market.

What The Trade Desk actually launched

Zuma is not a new ad network. It is the latest release of Kokai, The Trade Desk’s platform for planning, buying, and measuring advertising across the open internet. The update combines a refreshed interface with a more deeply embedded version of Koa, the company’s AI system.

According to The Trade Desk’s announcement, Koa analyzes more than 20 million ad impressions every second. In Zuma, specialized agents support campaign creation, audience building, troubleshooting, performance analysis, and dynamic frequency decisions. A conversational Koa assistant is intended to help buyers find information and take action without navigating every platform screen.

CEO Jeff Green described the goal plainly: “AI is helping our customers move faster than ever, while making measurement more intuitive and actionable.” That is a more consequential promise than simply generating recommendations. It points toward a media platform where the buyer’s job becomes setting objectives, approving changes, and evaluating outcomes while software handles more execution.

The 32% CPA claim is a signal, not a benchmark

The number will attract attention in every agency pitch deck, but disciplined marketers should read the footnote before reallocating budget. Independent coverage from PPC Land notes that the 32% average improvement is based on 62 campaigns and a comparison between an upgraded model and its predecessor. The result is statistically significant according to the release, but The Trade Desk has not published campaign-selection criteria, confidence intervals, or an independent replication.

That makes the statistic useful as a directional performance signal, not a promised outcome. It tells agencies that model quality and forecasting infrastructure can create measurable gains; it does not tell every advertiser what their CPA will be. Results will vary with conversion volume, creative quality, audience constraints, inventory mix, attribution design, and the accuracy of the conversion signal.

The practical response is to treat vendor AI claims like any other media hypothesis: establish a baseline, define the primary business outcome, run a controlled holdout or geo test where possible, and monitor lead quality—not just platform-reported CPA.

Why agentic media buying changes agency work

Zuma’s larger implication is organizational. When agents can build campaigns, expand audiences, troubleshoot delivery, and modify frequency, the value of manual platform operations falls. The value of strategy, signal design, creative testing, and governance rises.

This is the same shift already visible in broader marketing operations, from AI-assisted analytics to conversational media buying. Agencies should not position AI as a faster version of the old button-clicking workflow. They should package the higher-value layer: objective setting, offer and audience strategy, experiment design, cross-channel interpretation, and accountable approvals.

There is also a governance issue. Several Zuma capabilities—including the conversational assistant, frequency features, conversion-lift enhancements, and some bulk-editing improvements—are in closed or open beta. Agencies should not promise a client workflow around a feature that has no general-availability date. Every automated action needs a visible approval boundary, a change log, and a rollback plan.

What marketers should do next

  • Audit the signal first. Confirm that purchase, qualified-lead, margin, and offline conversion events are deduplicated and recent enough for machine learning.
  • Pilot one outcome. Choose a campaign with enough volume to measure, define a baseline CPA or revenue-per-impression target, and compare the AI-assisted setup with a holdout or pre-registered test.
  • Measure incrementality. Platform optimization can improve reported efficiency while shifting attribution. Add lift testing, geo experiments, or matched-market analysis where the economics support it.
  • Rewrite the agency role. Move human hours away from repetitive trafficking and toward business strategy, creative iteration, QA, and explaining trade-offs to the client.
  • Set guardrails before scale. Require approval for budget changes, audience expansion, brand-safety settings, and any action that can materially alter spend or customer experience.

Kokai Zuma is not proof that every ad account should hand control to an agent. It is proof that the competitive frontier in paid media is becoming the quality of the system around the model: clean signals, useful experimentation, trustworthy measurement, and humans who know when not to automate.

Ready to turn AI media buying into a measurable growth advantage? Real Internet Sales helps businesses build AI-ready marketing systems, stronger measurement, and practical automation. Call 803-708-5514 or visit realinternetsales.com.