Issue 30
THE TAKE
AI is moving into one of the most sensitive decisions a company makes: price.
In September, Reuters reported that McDonald’s pricing engine uses machine learning to analyze transaction data from across its U.S. restaurants and recommend menu prices for individual locations. Local market conditions help shape those recommendations. Reuters reporting
McDonald’s says franchisees independently set their prices and decide whether to use its recommendations. The company also says the tool does not set prices for individual customers or change prices in real time or by time of day. McDonald’s October 1 statement
The governance issue begins when a recommendation becomes difficult to challenge.
AI can also influence decisions about underwriting, promotions, discounts, renewals, supplier negotiations, and customer offers. These are commercial decisions that affect revenue, margin, trust, and regulatory exposure.
My view is that boards do not need to approve individual prices. That would be ridiculous. But they do need to understand the boundaries around algorithmic pricing: what data is used, what outcome is being optimized, who can reject the recommendation, how exceptions are handled, and who owns the final business judgment.
THREE THINGS BOARDS SHOULD KNOW
Pricing algorithms create business risk, not just technical risk. A model can improve margin and still damage trust, create regulatory exposure, or produce a customer outcome leadership would not defend publicly.
Optimization needs boundaries. If the system is told to maximize revenue or margin without guardrails, it may produce technically rational decisions that are strategically or reputationally wrong.
Accountability does not move to the model. AI may recommend the price, but management still owns the decision and the consequences.
THE QUESTION I’M HEARING
“Should boards really be involved in how AI influences prices?”
Boards should not manage individual pricing decisions. They should govern the decision framework.
For any company using these systems, that starts with clarity around the data and objective. Is the model using transaction history, geography, demand, competitor pricing, loyalty data, customer behavior, or demographic signals? Is it optimizing for revenue, margin, volume, retention, or market share? Different goals create different outcomes, and some outcomes may not align with the company’s values, brand, or legal obligations.
My recommendation is that any material AI pricing engine should have defined guardrails, human ownership, outlier review, legal review, and scenario testing. A human approval step should involve real authority to challenge the output.
Do not only test whether the model improves margin. Test what it recommends during shortages, local emergencies, demand spikes, competitor disruptions, or unusual customer behavior. Decide in advance which recommendations require escalation and who can stop their use. If the algorithm produces a result leadership would not defend in public, the control model is not finished.
THE BOTTOM LINE
AI can make pricing smarter, but it cannot own the judgment behind the price.
When AI influences what a customer pays, management still owns the outcome.