Agentic DSP Workflows: When AdTech Starts Running Itself

The phrase ‘agentic AI’ has been circulating long enough to lose meaning — until it shows up in your DSP and starts making bid decisions without waiting for you.

Yahoo’s integration with Kochava, reported by Digiday, is one of the clearer signals yet that ‘agentic’ is moving from conference slide to live infrastructure. StationOne, Yahoo’s DSP, is now being pitched with interoperability hooks that allow AI agents to act on measurement signals from Kochava’s attribution layer — closing the loop between identity, measurement, and media execution in ways that previously required a human analyst sitting between the systems. That’s not an incremental upgrade. That’s a structural shift in where human judgment enters the workflow.

What ‘Agentic’ Actually Means at the DSP Layer

Strip away the marketing and the architecture is straightforward: instead of a media buyer pulling reports, interpreting signals, and adjusting targeting parameters, an AI agent monitors those signals continuously and executes changes autonomously — within guardrails the team defines upfront. Yahoo’s pitch pairs this with Kochava’s identity graph, meaning the agent isn’t just optimising on last-click attribution. It’s working with a more complete picture of the user journey, across devices and channels, in something closer to real time.

The practical implication is significant. In Southeast Asia, where cross-device fragmentation is severe — a single user might interact with a brand across a Shopee ad, a LINE message, a mobile browser session, and a CTV unit in the same week — the ability to resolve that identity and act on it programmatically, without manual stitching, is genuinely valuable. The problem is that the value is only accessible if your own first-party data infrastructure is clean enough for the agent to work with. Garbage in, autonomous garbage out — at scale.

The Identity Stack Is Now a Competitive Prerequisite

The Yahoo-Kochava pairing isn’t unique in what it’s trying to solve; it’s notable because of how explicitly it positions identity resolution as the foundation for automation. You cannot have agentic media buying without a functional identity layer. The two are inseparable.

This reframes how marketing and data teams should be thinking about their clean room investments. A clean room isn’t just a privacy-compliant way to collaborate with a platform’s data — it’s the on-ramp to automated decision-making infrastructure. Brands that have already built interoperable first-party data environments, with consistent user identifiers across CRM, web, and app, are positioned to plug into these agentic systems relatively quickly. Brands that haven’t done that work yet are now facing a compounding disadvantage: they’re behind on identity resolution and behind on the automation layer that depends on it.

For teams in markets like Thailand, Indonesia, or the Philippines — where platform-owned identity (Grab, Shopee, TikTok) often dominates over brand-owned data — this creates a specific strategic question: are you building toward identity independence, or are you comfortable being a permanent tenant in someone else’s graph?

---

Disney’s Engagement Metric Problem Is Everyone’s Engagement Metric Problem

A parallel signal came from an unlikely source. Disney’s first earnings call under new CEO Josh D’Amaro, covered by AdExchanger, revealed $25.2 billion in quarterly revenue — up 7% year-over-year — but ESPN ad revenue down 2% despite subscription and affiliate revenue growing 6%. D’Amaro’s stated focus: engagement, not just reach.

The distinction matters beyond Disney’s internal strategy. What’s happening at ESPN is a microcosm of a broader tension in programmatic advertising: inventory that looks healthy by traditional reach metrics can underperform on engagement signals, and advertisers with access to engagement-weighted buying — the kind of signal an agentic DSP can act on in real time — will increasingly out-allocate toward quality rather than volume. The brands still buying on CPM alone are flying partially blind.

For Southeast Asian advertisers running video across platforms like YouTube, TikTok, and regional CTV inventory, this is a useful forcing function. Engagement data — completion rates, interaction signals, post-exposure search behaviour — needs to be feeding back into your DSP logic, not sitting in a separate analytics dashboard that someone reviews on Fridays.

What Teams Should Do Before the Agents Arrive

The risk with ‘agentic DSP’ as a concept is that it creates paralysis — teams waiting to see how the technology matures before committing to infrastructure decisions. That’s the wrong posture. The infrastructure decisions are what determine whether you can participate when the technology is ready.

Three things worth doing now: First, audit your identity resolution capability honestly. Can you match users across your owned channels with a confidence level above 70%? If not, that’s the priority — not the agent. Second, define the guardrails you’d want an AI agent to operate within before you need to set them. Spend ceiling per audience segment, frequency caps, channel allocation limits — these are strategic decisions that shouldn’t be made reactively under time pressure. Third, pressure-test your measurement setup against the assumption that last-click attribution is no longer the primary signal. If your current setup can’t surface multi-touch or engagement-weighted data, an agentic system won’t fix that — it’ll just automate the wrong optimisation faster.

The playbook for agentic media buying hasn’t been written yet. But the infrastructure decisions that will determine who gets to use it are being made right now.

---

Key Takeaways

• Agentic DSP workflows require a clean, interoperable identity layer as a prerequisite — automation amplifies your data quality, for better or worse.
• In fragmented Southeast Asian markets, the gap between brand-owned and platform-owned identity graphs is a strategic risk that compounds as automation matures.
• Engagement signals need to be integrated into DSP logic now — not reviewed in hindsight — to take advantage of the shift toward quality-weighted programmatic buying.

The harder question underneath all of this: if AI agents are making increasingly autonomous decisions about where your media budget goes, what does the media buyer’s role actually become? Strategy and guardrail-setting are the obvious answers — but that requires a level of strategic clarity about audiences, objectives, and acceptable trade-offs that most teams haven’t had to formalise before. The technology will force that conversation.

---

At grzzly, we work with brands across Southeast Asia at exactly this intersection — helping teams build the identity and measurement infrastructure that makes automation viable, not just theoretical. If you’re trying to figure out where your stack stands before the agentic wave hits, we’d rather have that conversation early than help you catch up later. Let’s talk