AI Agents Bought Rouge Care's TV Ads, Returned 5x ROAS

Wellness brand Rouge Care ran an experiment that ad ops teams should pay attention to: it handed the execution of its television advertising campaign to autonomous AI agents, allowing software to make buying and optimization decisions with limited human intervention. According to the brand, the approach returned roughly five times its media investment — an early, if anecdotal, proof point for the emerging concept of agentic media buying.

What "AI agents buying ads" actually means

In traditional programmatic buying, humans configure a campaign inside a DSP: setting budgets, targeting parameters, frequency caps, bid strategies, and creative rotation, then monitoring performance and adjusting. Agentic buying reframes this by delegating those decisions to AI systems that can interpret goals, query available inventory, place bids, and reallocate spend in near real time — closing the loop between measurement and optimization without a human touching the dashboard for every change.

For Rouge Care, the agents operated across TV and connected-TV (CTV) inventory, adjusting where and when ads ran based on performance signals. The reported 5x return is a business outcome, but the mechanism behind it is what matters for the ad-tech ecosystem: it represents software making the granular, continuous decisions that programmatic pipelines already support but that humans have historically bottlenecked.

Why CTV is fertile ground for agentic buying

Connected TV is uniquely suited to this model. Unlike linear television, CTV inventory is transacted programmatically through SSPs, exchanges, and DSPs using OpenRTB, with impression-level addressability, log-level reporting, and server-side ad insertion (SSAI) delivering ads into streaming ad pods. That data density — bid requests, win rates, completion rates, and outcome signals — is exactly what an autonomous optimization system needs to function.

An AI agent operating against CTV supply can, in principle, evaluate thousands of bid opportunities, weigh contextual and audience signals, and shift budget toward the app environments, dayparts, and pod positions that convert. The same automation is far harder to apply to legacy linear buys, where transactions are still largely negotiated and insertion orders are slow to change.

The technical implications for the buy side

If agentic buying scales, it changes assumptions across the programmatic stack. DSPs would need to expose richer, machine-readable interfaces — APIs and outcome feedback loops — rather than dashboards designed for human traders. Bid strategies become policies an agent tunes rather than static settings. Measurement and verification (viewability, IVT/ad-fraud detection, completion tracking) become even more critical, because an autonomous system optimizing toward a metric will exploit whatever signal it is given. Poor or gameable measurement could lead an agent to chase fraudulent or low-quality inventory efficiently.

This raises the stakes for supply-chain transparency standards — ads.txt/app-ads.txt, sellers.json, and the OpenRTB supply chain object — as the guardrails that keep autonomous spend flowing to legitimate inventory. An agent buying at machine speed across CTV needs trustworthy provenance data to avoid concentrating budget in spoofed or misrepresented supply.

Signal, not certainty

A single brand's 5x result should be read as a directional case study rather than a benchmark. Small campaigns can post outsized returns for reasons unrelated to the buying method — favorable creative, a receptive audience, or timing. The more durable takeaway is that the tooling to let software run programmatic video and CTV campaigns end-to-end is maturing, and brands are willing to test it.

For ad ops teams and publishers, the practical questions are about control and accountability: how do you set the objective function an agent optimizes toward, what constraints (brand safety, frequency, budget pacing) do you hard-code, and how do you audit decisions after the fact? The shift mirrors what programmatic itself once represented — automation replacing manual insertion orders — but pushes the automation up the stack, from execution to decision-making.

Where this goes next

Expect DSPs, SSPs, and CTV platforms to lean into agent-friendly interfaces and outcome-based optimization as agentic buying gains attention. The winners will be the platforms that combine clean, machine-readable supply data with reliable measurement — the two inputs an autonomous buyer depends on. Rouge Care's experiment is a small data point, but it points to a larger structural change in how programmatic video and CTV budgets may be deployed.

Stay on top of video ad serving and programmatic. Follow Adelerate.