For one fashion platform we researched, Monday’s dashboard looked like a victory lap. 8% increase in profits in 21 days, thousands of price re-adjustments every hour, and 17,000 price decisions per day. AI has learned to read demand, inventory, weather, social signals, and influencer activity faster than any revenue team. Conversions went up. Cart size has increased. Revenues decreased.
After three months, sales were down 40%.
This story is becoming common. And the problems it reveals are not technical.
Most executives today are asking the right questions about AI. “What can we automate?” Where will we create productivity? What capabilities can we extend? These are important questions. But they miss the more important point. It’s about what AI is already making decisions for you, and are those decisions reflective of the company you want to be?
The real risk with AI pricing is not that the machine will set the wrong price. That is, companies inadvertently turn pricing policy into an optimization problem, only to realize too late that algorithms have made strategic, ethical, and political choices that management never discussed.
Shifts that most executives don’t fully register for
For most of the past decade, AI in business has acted as a co-pilot. We’ve recommended, flagged, and ranked options. It was decided by humans. Although imperfect, this architecture maintained clear lines of responsibility.
Its architecture has quietly changed. We do not recommend next-generation Agentic AI systems currently deployed for pricing, customer management, and revenue optimization. they act. They pursue goals autonomously, remember past interactions, learn from outcomes, and continually adjust at each step without the need for human validation.
In pricing, this change is already visible. Salesforce made the first move. Agentforce introduces action-based pricing. Customers can pay not only for access to the software, but also for the work that the AI agent completes. The unit of value is shifting from humans to actions performed by machines. This is not an incremental optimization. It is a fundamental shift in how decisions are made, by whom, and at what speed.
McKinsey’s latest AI research shows that while the use of AI is now widespread, most organizations are still in experimentation or pilot mode, and only a minority are scaling agent AI. This makes governance even more urgent. Companies are looking to give machines more autonomy before the routines for deciding when human validation is needed are mature.
3 Ways of AI Pricing You May Have Missed
First, speed. Pricing agents now operate faster than human review cycles. Prices can fluctuate multiple times a day based on signals (demand, inventory, competitor actions, time of day) that cannot be processed in real time by human teams. Decisions are individually defensible. But their cumulative impact on customer trust, brand positioning, and long-term profits are rarely audited with the same frequency.
Second, personalization. In reality, agent systems can create different markets for different customers. If two people access the same platform at the same time, they may see different prices, different bundles, and different emergency signals tailored to what the system infers about their willingness to pay. Revenue growth is real. The same goes for exposure. The gap between “personalized offers” and “discriminatory pricing” can collapse overnight as customers compare notes.
Third, memory. Unlike previous AI tools, the agent system builds a cumulative model of each customer relationship. Past decisions shape future decisions. A price setter who learned six months ago that a particular customer segment would accept a 15% premium will continue to apply, refine, and extend that logic unless someone explicitly repurposes it. Individual decisions do not trigger a review. Drift is invisible until it actually occurs.
The real risk is not the wrong price. it’s an unconsidered decision
Ticketmaster has issued a warning, indicating it’s nothing new. Between 2019 and 2022, the number of dynamic price tickets for North American concerts sold by Ticketmaster increased by more than 700%, according to a U.S. Senate investigation. The system did not fail because it could not be optimized. It failed because the optimization conflicted with consumers’ sense of fairness.
The subcommittee launched an investigation into Ticketmaster’s role in soaring ticket prices. The algorithm was not malfunctioning. It worked exactly as designed, but no one defined how far was too far.
The commercial costs of perceived unfairness are not theoretical. Bain has long argued that online retailers need to offer customers fair prices, not necessarily the lowest prices, and pricing research shows that perceptions of fairness and loyalty are deeply intertwined. But the companies that build the most sophisticated pricing engines are often the least capable of explaining, defending, and modifying what those engines are doing. Machine precision trumps institutional governance.
The failure mode is rarely a bad algorithm. It is an organization that delegates decisions without defining limits.
Testing what to delegate and what not to delegate
Not all pricing decisions have the same organizational weight. A useful distinction distinguishes between frequent, local, and reversible transactional decisions with limited impact and organizational decisions that concern brand promises, perceptions of fairness, long-term relationships, or social license to operate.
Transaction decisions are ideal candidates for full automation. Adjusting prices based on real-time inventory, regional demand signals, or channel dynamics fits right into this category. This is where speed and consistency truly trump human judgment.
Institutional decisions are different. These are things like Ticketmaster where the optimization is done correctly but the relationships are wrong. A test that can help in these cases is called the 4Rs. Before delegating pricing decisions to an AI agent, executives should ask four questions: reversible Quickly without setting a harmful precedent? Is it enough? routine How can it be standardized? is that so bound by rules By clear company rules? and is there an error Repairable Even though there is no public crisis? If the answer to these questions is no, you should not delegate the decision entirely. If any one of these conditions fails, full delegation to machines becomes a governance risk rather than a productivity gain.
The deeper problem is structural. Every agent’s pricing system encodes a vision of what is important, what can change, what is negotiable and what is not. Companies that do not make these choices explicitly are making them implicitly by default through the objective function that their data scientists optimized last quarter.
Questions every CEO should be able to answer
The executives we spoke to who have navigated this transition best have one characteristic: They treat AI pricing governance as a leadership issue rather than an IT issue. Clarify what pricing agents are allowed to make on their own and what requires human decisions without consulting a dashboard.
They effectively created a pricing constitution, a rule that generates all future prices before that price is generated. Such a configuration must define what variables agents can and cannot use, how much price variation is allowed between customer segments, when humans must approve changes, and whether long-term metrics such as trust, churn, repeat purchases, and complaints can override short-term profits.
The question is not whether AI will make pricing decisions. That’s already the case. The question is whether the rules the company is following are rules that it will uphold before its customers, its board of directors, and, if necessary, a Congressional hearing.
If you can’t answer it today, you’ll need to define the rules before the notification arrives next Monday morning.
François Candron He is a partner at the private equity firm Seven2 and an executive fellow at the HBS AI Institute (formerly known as the D^3 Institute).. read other luck Column by François Candron.
Paul-Louis Andres is the director of Seven2.
Augustine Manchon is a professor at HEC and Paris Dauphine and a pricing strategy and governance advisor to the CEO of Manchon & Company.
