Tag: AI Leadership

McDonald’s AI Pricing Dispute: Who Really Gets the Final Say?

McDonald’s says its franchisees make the final call on menu prices. Reuters reported something messier. On September 29, 2026, Reuters published an investigation into McDonald’s existing pricing recommendation system and reported that five franchisees said they felt pressure to follow the company’s recommendations.

McDonald’s responded on October 1 that franchisees independently set their prices and can reject recommendations. The company also said the system does not set individualized customer prices and does not use real-time dynamic pricing.

That disagreement matters beyond hamburgers. If a company can truthfully say, “A human makes the final decision,” several practical questions remain. Who decided what the system should optimize? What information does the person making the decision actually see? Can that person disagree without creating another organizational problem? Who owns the result if the recommendation turns out to be wrong?

AI product executive Jason M. Riggs works on AI products in restaurant automation, where apparently simple outputs depend on a long chain of product decisions. In our conversation about AI leadership and decision speed, he described the complexity behind AI-powered drive-thru ordering, where customer requests can depend on menu data, modifiers, promotions, brand rules, point-of-sale systems, and human judgment during product development.

That does not give Jason inside knowledge of McDonald’s pricing system. It does give him a useful lens for examining what “human judgment” actually requires once AI starts recommending decisions that affect customers and the business.

For an executive, founder, or operator rolling AI into real decisions, the useful question is not whether a human technically remains somewhere in the loop. The harder question is whether that person has enough context, authority, and accountability to make the judgment meaningful.

The AI Recommendation Starts Long Before Someone Clicks Yes

The easiest place to look for human judgment is at the end of the process. The system makes a recommendation, someone looks at it, and that person accepts or rejects it. A lot has already happened before that moment.

Jason sees this in restaurant AI from the product side. When he described AI-powered drive-thru ordering, he pointed out that something as ordinary as ordering a hamburger gets complicated fast. Different brands use different language. Menus have modifiers, combos, coupons, limited-time offers, and rules about how those choices get entered into the point-of-sale system.

Drive-thru ordering and pricing are different problems. Jason did not build or evaluate McDonald’s pricing system. His example is useful for a narrower reason. It shows how much business judgment can disappear behind a clean recommendation on a screen.

Someone had to decide what the system was supposed to accomplish.

Reuters reported that McDonald’s pricing engine analyzes restaurant data and produces pricing recommendations for individual locations. Its reporting also described company involvement in parameters used to generate those recommendations, based in part on interviews with former employees of Tiger Analytics, which worked on the system.

Even if a franchisee ultimately decides whether to accept a recommendation, the recommendation did not arrive from nowhere. Someone selected inputs. Someone decided which outcomes matter. Someone decided what should count as an improvement and how the recommendation should be presented.

Those choices happen upstream.

The same issue shows up in smaller businesses. Suppose you give a sales manager an AI recommendation about which leads deserve immediate follow-up. The manager still has the final say, but the system may already have prioritized deal size, recent engagement, industry, or another signal before the manager sees anything.

If the system ignores something the manager knows from experience, the quality of the final human decision depends on whether the manager understands enough about the recommendation to challenge it.

Pricing can make those hidden choices particularly important because several reasonable objectives may pull in different directions. A recommendation could favor transaction growth, profitability, traffic, customer perception, competitive positioning, or some combination of them.

There is no need to guess which objective McDonald’s system prioritizes in a particular case. The Reuters reporting supports a more useful point. A recommendation is the output of choices made before the local operator ever receives it.

McDonald’s says franchisees independently set restaurant prices and may reject its recommendations. That distinction matters. The software recommends. The franchisee has the formal authority to act.

For an executive introducing AI into a consequential decision, formal authority is only one layer to inspect. You also need to know what problem the system was built to solve, what information shaped its answer, and whether the person receiving that answer knows enough to question it.

Otherwise, the final human review can become approval of a decision whose assumptions were settled somewhere upstream.

The next issue is what has to be true inside an organization for the ability to say no to function as real decision authority.

Having the Final Say Is Different From Being Free to Use It

Jason gave a useful example from his own restaurant AI work.

He described an AI system questioning an automatic upsell because the model considered it a mistake. From the model’s perspective, that recommendation could appear reasonable. From the business’s perspective, the upsell was intentional.

Someone had to know the customer, the business rule, and why the rule existed. Someone had to recognize that a technically plausible recommendation conflicted with what the company was actually trying to accomplish.

That is a more demanding version of human judgment than simply checking whether the AI produced an answer.

The analogy to pricing has limits. Jason’s drive-thru experience tells us nothing about whether a particular McDonald’s pricing recommendation is good or bad. It does show why an AI-assisted business decision can carry assumptions that are not obvious when it finally reaches the person expected to approve or reject it.

At McDonald’s, the company’s position is clear. Franchisees independently set menu prices and may reject the recommendations they receive.

Reuters reported a more complicated experience from some operators. Five franchisees told Reuters they felt pressure to follow the recommendations. That does not establish that franchisees lack the authority to say no. McDonald’s disputes any suggestion that the system removes their pricing control.

That tension is worth examining without pretending we can resolve it from outside the company.

Authority on paper and usable authority inside an organization are not always identical.

Imagine an operator receiving an AI-assisted recommendation that conflicts with something they know about their local customers. They may know a nearby competitor has changed an offer, that a particular item is unusually price-sensitive, or that a short-term event distorted recent demand.

The important issue is not merely whether there is a way to reject the recommendation. It is what happens when the operator does.

Can they explain the decision using information the system did not have? Does the organization want that local judgment, or does disagreement automatically look like resistance? Is the recommendation one input among several, or does it gradually become the default answer that requires special justification to challenge?

Jason described a related accountability problem in corporate decision-making: “nobody wants to make a decision because if you make a decision, you’re accountable, and you have sort of this group accountability, which means nobody’s accountable.”

AI can make that problem stranger.

The system recommended it. Corporate supplied the tool. The local operator approved it. Responsibility can become distributed across the process even though somebody still has to live with the outcome.

Meaningful human judgment requires more than keeping a person somewhere in the workflow. The person needs enough context to question the recommendation, enough authority to act on that judgment, and a clear understanding of who owns the result after the decision is made.

The Software Doesn’t Live With the Consequences

A person being present in the workflow does not automatically mean that person owns the decision.

AI can produce an answer, recommendation, or next step, but the system does not experience what happens after somebody follows it. Jason put this bluntly while talking about large language models: “they don’t care about your time, they don’t care about your money.”

He was not talking about McDonald’s pricing technology. The broader management problem still applies to AI-assisted decisions. Software can recommend an action without having any stake in whether it helps the customer, damages a relationship, squeezes a margin, creates extra work, or turns out to be wrong. Someone inside the business still owns those consequences.

That person needs more than permission to click yes or no. They need enough information to understand what the recommendation means and enough authority to challenge it when something important is missing.

Suppose a pricing recommendation looks mathematically reasonable but conflicts with something a local operator knows about the market. Meaningful judgment requires a way for that information to matter.

If the operator can technically reject the recommendation but has no useful way to explain why, or expects every disagreement to trigger another layer of approval, the organization has not necessarily created better human oversight. It may have created a slower rubber stamp.

Decision speed matters here.

Jason has argued that faster execution can expose slower corporate decision structures. Teams can produce analysis, prototypes, and recommendations faster than before, then sit waiting for finance, leadership, or another alignment meeting to make the next call.

Adding AI does not fix that problem if every automated recommendation creates another queue of humans waiting to approve it.

The answer is not to have a person manually bless every routine transaction.

In Jason’s restaurant work, the goal is eventually to let a well-designed system handle the situations it was built to handle. Human judgment matters heavily while the product is being designed, while exceptions are being defined, and when something consequential falls outside the expected pattern.

Keeping a human involved everywhere can defeat the reason for automating the work in the first place. The harder management job is deciding where judgment belongs.

For a consequential recommendation, someone usually needs to understand what the system is optimizing, recognize information it may not have, know when an exception deserves attention, and have clear authority to act. Someone also needs to look afterward and determine whether the decision produced the result the organization wanted.

The McDonald’s reporting makes that ownership question visible without resolving it for us. Reuters reported that several franchisees felt pressure to follow pricing recommendations. McDonald’s says franchisees retain pricing authority and may reject them. Those accounts should not be collapsed into a conclusion the evidence does not establish.

They do expose a useful test for any company using AI-assisted recommendations.

If a recommendation goes badly, who is expected to explain the decision?

If the answer is the operator, manager, or executive who approved it, that person needs more than ceremonial involvement. They need the information and authority required to exercise judgment before the consequences arrive.

Jason warned about the opposite habit directly: “Don’t let your human judgment atrophy, because it will.”

Keeping that judgment active does not require slowing every decision down. It requires deciding where human responsibility sits and giving that person enough room to use it.

Audit One AI Decision Before You Add Another Approval Layer

The practical way to use Jason’s ideas is to inspect one consequential AI-assisted decision already happening inside your business.

Do not start by asking whether the company needs more AI governance. That can easily produce another committee, policy document, or approval step. Start with one real recommendation that somebody is already expected to accept, reject, or act on.

First, write down what the recommendation is supposed to improve.

Is the system trying to increase revenue, reduce cost, improve conversion, shorten response time, protect margin, reduce errors, or improve customer experience? If nobody can state the goal clearly, it becomes difficult to judge whether the recommendation is useful.

Jason described the danger of approaching technology backward when he talked about a company pushing employees to use more AI simply because leadership wanted more AI. His reaction was blunt: “Here’s a solution. Go find some problems.”

The warning still applies after a system is already in place. You need to know what problem the recommendation is meant to solve.

Next, name the decision owner.

Not the department. Not “leadership.” Not “the team.” Put one role or one person beside the decision.

That does not mean one person performs every step or ignores everybody else’s expertise. It means someone has clear authority to make the call when the recommendation matters.

Then define what evidence would justify overriding the system.

This gives local knowledge, customer information, unusual circumstances, or better data a way into the process without turning every disagreement into an argument over whether somebody “trusts AI.”

A sales manager might know a large account has gone quiet because the buyer changed jobs. An operations leader might know a supplier delay makes the recommended schedule impossible. A restaurant operator might know something about local demand that is not represented in the data feeding a recommendation.

Human intuition does not automatically beat the system. Disagreement still needs a path.

If the only acceptable answer is whatever the system recommended, the person reviewing it does not have much judgment to exercise. If every disagreement requires three meetings and executive approval, you have created a different problem.

Finally, decide who looks at the outcome.

A recommendation should not disappear after somebody approves it. Someone needs to check what happened.

Did the decision improve the thing the system was supposed to improve? Did it create a customer problem somewhere else? Was the override justified? Did the recommendation expose something the human decision maker had missed?

That is where the organization can learn instead of merely accumulating approvals.

For one AI-assisted decision, write down four things: the goal, the owner, the evidence that can justify an override, and the person responsible for reviewing the result.

This is not a framework Jason named, and it is not a claim about how McDonald’s should structure its pricing process. It is a practical application of the problem his comments expose.

A recommendation can arrive in seconds. You do not necessarily need another layer of review around it. You need to know who owns the decision, what would justify saying no, and whether anyone checks what happened next.

Faster Decisions Still Need an Owner

The phrase “final say” sounds reassuring until you try to define it.

A person may technically approve or reject a recommendation, but that only matters if they know what the system is trying to do, can see when important information is missing, and have enough authority to act on their judgment without turning every exception into a bureaucratic fight.

Reuters reported that several McDonald’s franchisees felt pressure to follow pricing recommendations. McDonald’s says franchisees remain free to reject them. The reporting does not let us settle what every operator experiences inside that system, but it exposes a management question that will keep showing up as companies put AI into consequential decisions.

Who actually owns the call?

Jason has made the same broader point from a different direction. Faster execution only helps if decision-making keeps pace. In his words, “speed without clarity just creates chaos.”

Pick one AI-assisted decision in your business and make sure you can name the person who owns it, the evidence that would justify overriding the recommendation, and who checks the result.

If you cannot name that owner, the decision process is still unfinished.

An AI recommendation can arrive in seconds, but someone still has to own the call. Jason M. Riggs addresses that tension in “Decision Speed with Human Judgment,” his contribution to Decisions, Courage, and Action, the book we co-authored. Pick up the book on Amazon to read Jason’s chapter and the other co-authors’ perspectives on making decisions and following through.

About Jason M. Riggs

Jason M. Riggs is an AI product executive with more than 25 years of experience across software, hardware, cloud, and applied AI. In the September 4, 2026 episode of Marketer of the Day, he was identified as Chief Commercial and Product Officer at Audivi AI, where his work includes AI-powered drive-thru automation. His experience focuses on product leadership, decision speed, organizational accountability, and the role of human judgment as AI makes execution faster. He is also the author of The MACH-10 PM and The MACH-10 Leader. Connect with Jason M. Riggs on LinkedIn.

Media and interview inquiries: To request an interview about AI leadership and human judgment, visit his press page.

Resources

• Reuters: Inside McDonald’s push to have AI price your Big Mac
• McDonald’s: Separating Fact from Fiction: AI Does Not Set Prices at McDonald’s
• Decisions, Courage, and Action
• Jason M. Riggs on Marketer of the Day
• MACH-10 Leadership Series