Grigoriy MisilyukMany direct-deal workflows still rely on slow, manual coordination across multiple teams and systems. Even when agreements are standardized and follow repeatable processes, coordination still takes a lot of time: inventory checks, pricing, approval chains, and endless emails.
More deals mean more coordination and more people involved. Time lost here is money lost.
The solution: AI agents that automate coordination for structured deals by matching buyer requests to available inventory and preparing campaigns for approval. Early industry experiments suggest that structured direct deal workflows are among the most realistic early use cases for agent-based automation. Experience confirms that implementation success depends entirely on infrastructure, data quality, and how each business defines "standardized.”
Here's the complete picture: why it works for direct deals, what challenges it presents, and whether testing makes sense for your business.
Why start with direct deals?
Direct and controlled deals became the first use case for AI agents for two practical reasons, not just market size.
First, structured environments enable automation. Direct deals operate with predefined parameters: specific audiences, fixed budgets, agreed dates, and clear KPIs. Predictable variables mean programmable decisions — agents can support availability checks, pricing logic, and proposal generation within predefined rules and approval frameworks. Compare this to OpenRTB: billions of impressions per day, unpredictable supply quality, dynamic bidding — far less structured, far harder to automate with AI agents.
Structured environments also make it easier to design communication protocols — the second critical enabler. Standards like AdCP create a common language for AI agents to match buyer requests with seller inventory. When a buying agent sends a request (audience, budget, dates), the seller agent checks if the inventory meets those criteria. If there's a match, agents coordinate activation: prepare setup, trigger workflows, and queue for approval. In standardized scenarios, agents can match supply conditions against incoming requests with minimal manual coordination.
This matching logic works when both sides operate within predefined parameters. Buyers define what they need, sellers define what they offer, and agents coordinate the match, preparing deals for human approval. Open-market buying remains more complex because of the scale, speed, and variability involved. Direct deals offer the right balance: high enough volume to justify automation, structured enough for reliable automation.
Challenges of implementing AI agents for programmatic direct deals
Let’s start by noting that the use of AI agents in programmatic is just another tool. Agent-to-agent interaction does not replace OpenRTB, nor does it replace the human negotiation of atypical direct deals, or direct deals with variable terms or large traffic volumes. For now, we’re only talking about standard structured direct deals.
Human involvement matters
AI will not replace sales teams or account managers. People remain essential for defining performance criteria, configuring which datasets and access permissions agents use to create structured requests, and approving deals.
Data quality dependency
Inaccurate or missing data creates specific operational problems. Agents amplify data quality problems rather than hide them. Real-time data infrastructure is required — not just having data, but ensuring it's complete, current, and accessible.
Agents need automated access to data via APIs and integrated systems. Manual processes, overnight updates, or systems that don't share data create gaps. If infrastructure relies on periodic refreshes or separate data sources, agent proposals will be inaccurate or impossible to complete.
Infrastructure as a competitive factor
Shared platform vendors define which data sources agents can access and what custom logic you can add to the agent's workflow. This works well for getting started and experimenting — the infrastructure is ready to use. Owning your platform and agent adds custom data integrations and deeper workflow control, which matters more as your operations scale or require differentiation. The two approaches are often complementary: many businesses start on shared platforms, then move to their own infrastructure for greater control.
Successful adoption requires addressing these constraints: human oversight, data quality, and infrastructure decisions. These challenges are worth tackling — let's examine why.
Benefits of AI-powered direct deal automation
The challenges are real, but the operational benefits make experimentation worthwhile.
Faster deal execution. In early testing, deals that previously took around two weeks closed in roughly 48 hours once AI agents were in place.
Transformed manual work. After infrastructure is set up, agents handle routine tasks, allowing sales teams to focus on high-value activities: complex negotiations, relationship building, and proactive business development.
Workflow consistency. Following already-standardized workflows, agents apply them consistently every time — reducing errors, making training easier, and ensuring compliance with business rules.
Scaling standardized inventory sales. Agents handle multiple standard deal requests simultaneously. Your team shifts focus to reviewing and approving deals.
These benefits compound over time as teams scale operations. But they depend on standardization, data quality, and infrastructure readiness — which is why testing on your own operations matters more than theoretical assumptions.
Testing agent-based trading in practice
Before adopting an agent-based system, you should evaluate whether your workflows, data, and infrastructure are ready. That's what experimentation reveals.
We built our AdCP Sales Agent MVP as an early-stage environment for exactly this: controlled testing of structured, approval-driven direct and controlled deals. It's an experimentation layer, not a finished product — a way to evaluate operational feasibility before committing to full implementation.
How agent-to-agent trading fits into programmatic infrastructure
Testing the sales agent in this environment helps publishers, SSPs, and wider sell-side participants understand:
How ready their infrastructure is for agent-based trading.
Whether their inventory and data are structured enough for agent trading.
Whether their workflows are suitable for automation.
Which direct deals can realistically be standardized and automated.
Where automation breaks down or requires human intervention.
The value is the operational assessment: clear answers to the questions above, grounded in your own operational reality. Publishers, SSP owners, and monetization teams can test this with their own inventory and approval logic, without needing a dedicated AI team.
Summary
The central question isn't whether these AI systems work — it's whether they fit your business model.
Agents automate manual processes in direct and controlled deals: inventory availability checks, comprehensive request matching, and proposal generation. Systems process structured requests and reduce coordination time between teams. The benefits: potential reduction in routine operational work, easier scaling, and increased efficiency.
The limitations are clear. Agents work only in standardized scenarios with defined rules. They don't fully replace negotiations and depend entirely on data quality. If inventory data is inaccurate or pricing rules are outdated, agent proposals will be wrong.
This automation doesn't eliminate manual work — it transforms it. Teams shift from executing standard requests to handling non-standard scenarios and system maintenance. This makes sense for businesses with high volumes of standardized deals. If you're evaluating agentic AI for your operations, reach out — we're happy to discuss implementation approaches.
