If you ask an AI agent to build a pricing system, it will happily do so. The problem? It might build exactly what you asked for—and destroy massive amounts of value in the process.
I’m seeing this happen across publishers, retail media networks, and ad tech platforms alike.
To be clear, there are plenty of places where AI should be used aggressively in pricing:
- Answering internal questions about pricing policies
- Automating deal desk approvals against explicit threshold rules
- Accelerating analytics so teams can query complex data directly
Notice what those examples have in common: the business defined the rules, and AI is executing them.
The danger occurs when AI crosses that line. Specifically, there are three traps you need to watch out for:
1. Algorithmic Price-Setting & Collusion Risk
We all know traditional price collusion—executives meeting in a room to manipulate market rates. But AI introduces a much less obvious version of this risk.
When pricing agents continuously monitor competitor prices and react to one another, you can end up with automated systems producing the exact same collusive outcome without human intervention. Do not let an AI agent autonomously optimize prices against competitors. You need explicit guardrails around what competitive data is used, what actions the system can take, and where human oversight is required. You do not want to discover regulatory boundaries by becoming the test case.
2. Relying on AI for Real Innovation
When Uber and Lyft built their advertising businesses, they didn’t just tweak impression pricing. They capitalized on owning the entire ride experience to sell the ride itself—creating the “Cost Per Ride” (CPR) model. That idea didn’t come from asking “how do we price impressions better?” It came from asking “what do we actually have the ability to sell?”
AI is great at looking across existing models, finding analogies, and adapting them. Show it CPMs, sponsorships, and takeovers, and it will suggest sensible variations. But real breakthroughs require stepping outside existing categories. If you hand product and pricing innovation to AI too early, you get a slightly refined version of what already exists instead of something fundamentally new.
3. Building Frameworks Without an Objective Function
Before asking AI to design a pricing model, you must step back and answer the core management question: What does success actually look like?
In digital advertising, higher prices and higher margins are not always the same thing. Raising CPMs can destroy total revenue if demand drops fast enough. Maximizing margin percentage can leave profitable volume on the table. Or your goal might be completely different: making packages simpler for sales to pitch, reducing discounting, or migrating clients to a new product line.
Each scenario requires a completely different pricing architecture. AI becomes dangerous here not because it gives bad answers, but because it gives a sophisticated answer to the wrong question. If you tell an AI agent to “optimize pricing,” it will—but defining what to optimize for remains a human decision.
The Bottom Line
Use AI aggressively when rules are clear, data is well-defined, and management has already decided what good looks like. But keep strategy firmly in human hands. AI can execute a pricing strategy remarkably well—it can also execute the wrong strategy with terrifying efficiency.
Best,
James Deaker
The Yield Doctor