Could AI Decide What Each Customer Pays? The Business Case and the Trust Problem
A system that recommends a price is not automatically setting a different price for every person. The McDonald’s coverage is a useful prompt for businesses to distinguish how pricing tools work, who approves their output and whether customers will regard the result as fair.
What the McDonald’s coverage does—and does not—show
A 29 September Reuters report, syndicated by CNBC, raised questions about McDonald’s use of AI-assisted price recommendations. McDonald’s public explanation says its tools provide restaurant-specific recommendations, that franchisees decide whether to accept them, and that the company does not use dynamic or individualised pricing. Those are the company’s stated position and should be attributed as such; neither a recommendation nor the use of AI proves that each customer is quoted a different price.
The reporting also describes pressure and disagreement around how pricing tools are used. That makes the story relevant to business governance: a system can influence a decision without formally owning it. The organisation still needs to know who sets the price, which data informs a recommendation and what happens when a manager rejects it.
Businesses should avoid collapsing several practices into the phrase “AI pricing”. The distinctions matter to customers and to the design of controls. The underlying model may forecast, compare or recommend; the commercial policy determines what can be offered and who can approve it.
Five different things a pricing system might do
Dynamic pricing changes a price over time, often in response to demand, capacity or availability. A hotel room costing more on a busy weekend is one familiar example. Customer segmentation groups customers into broad categories for analysis or offers; it does not necessarily mean each person receives a unique price.
A personalised offer selects a promotion for a customer or group, such as a voucher for a product they may buy. Individualised pricing goes further: the price itself differs for a particular person, potentially based on information about that person. AI-assisted pricing is a description of a decision-support method, not a pricing policy: software analyses data and suggests a price, which may still be reviewed or rejected by a human.
These practices can overlap, but they are not synonyms. A retailer could use AI to forecast which products will sell and recommend a store-level price without using personal customer data. Conversely, an apparently simple discount campaign can involve profiling if it is targeted using individual behaviour. The data and the effect—not the AI label—determine what questions the business should ask.
The commercial case is about better decisions
A well-designed system might help a business understand historical sales, stock levels, seasonal patterns, promotion results or local demand. That could improve forecasting, reduce avoidable waste, make a promotion more relevant or flag a price that needs review. The benefit is not automatic; it depends on whether the data is accurate and whether managers can act on the information.
Poor data can produce recommendations that are inconsistent or commercially damaging. An automated response to an unusual sales spike might raise prices when the cause is a data error. A model trained on historical outcomes may reproduce patterns that managers would not choose today. A short-term increase in margin can also cost long-term loyalty if customers see the result as arbitrary or exploitative.
Trust is therefore part of the business case, not a separate communications exercise. A price that is technically permitted may still be difficult to defend. If staff cannot explain why a recommendation appeared, or if customers cannot understand a promotion, the organisation may face complaints, lost sales and reputational harm.
Set policy before connecting data
Before a pricing pilot, write down the objective and the pricing rule. Decide whether the system may recommend, change or publish a price. Set clear floors, ceilings and excluded situations, such as sensitive products, distress events or a sudden data-quality failure. Identify which data is necessary and which is out of bounds.
Assign approval to a named role. A manager should be able to see the basis of a recommendation, override it and record why. Test it against a baseline and compare outcomes by product, location and customer group. Track not just revenue but complaints, conversion, margin, stock waste, price volatility and the frequency of overrides.
If personal data influences an offer or price, assess the purpose, transparency and data-protection implications before launch. Keep a human route for questions and corrections. Monitoring should include a way to pause the system when the data or result falls outside expected limits.
Customer acceptance is part of the evaluation
A useful pilot asks not only “Did the model improve the metric?” but “Would we be comfortable explaining this outcome to the person affected?” Describe the pricing approach in plain language, make promotion terms easy to find and avoid claiming that a price is personalised when it is not—or hiding personalisation behind a generic label.
McDonald’s says franchisees retain control over its recommendations and rejects the claim that it uses dynamic or individualised prices. For another business, the lesson is not that AI pricing is inherently wrong. It is that a recommendation system must be governed as part of the organisation’s pricing policy, with people accountable for the result.
AI may help a company price more responsively. Whether that is good business depends on the accuracy of the recommendation, the controls around it and the customer’s reason to trust how the price was reached.