What AI superintelligence actually means for QSR

In 1995, researchers Eric Darr, Linda Argote, and Dennis Epple published a study of 36 pizza stores in the same franchise system with 10 different franchisees. What they found is that greater operator experience was directly associated with lower production costs. That shouldn’t be surprising. But the researchers also pointed to knowledge transfer between stores belonging only to the same owner, while having found no evidence of transfer between stores belonging to different franchisees. So operators running practically the same exact restaurant, were not benefitting from each other's knowledge.

Growing up inside ChurroMania, a 150 store multinational franchise, I believe in franchising because I've watched my family build a business around teaching other people what we learned. The reason a franchisee buys a brand, is because they’re buying the benefit of decisions someone else has already worked through. There is massive value in the training, purchasing knowledge, operating standards, and access to experienced people.

However, we think there's considerably more value available than we have figured out how to deliver.

The detail we lose when we summarize the business

Across a franchise, two stores can both have slow lunch service and need different fixes. At one, delivery orders may use up the prepared ingredients before walk-in customers arrive. At another, the ingredients may be ready, but a new team may struggle to assemble orders quickly. The first store may need to change its prep timing or quantities; the second may need more training or a different way of assigning staff.

At enough scale, unusual situations recur. Somewhere in the system, somebody may already have found a better way to handle the situation another operator is facing this morning.

Some of that knowledge travels through field visits, calls, and revised procedures, but considerable detail disappears when the lesson becomes a slide saying that one store format performs better than another. An operator needs to know which part of the operation improved, what the manager changed, and whether a less experienced team with a different order mix could reproduce it. Ideally, this also happens before it becomes a problem.

Existing docs don’t help solve the problem: franchise disclosure documents serve a legal purpose and operations manuals establish how the business should run ceteris paribus. Neither is designed to deliver a fresh, situation-specific account of what hundreds of stores tried yesterday and what happened afterward.

We don't want the answer to a preventable problem to depend on the right two operators being on the same call.

One Friday, several possible explanations

Imagine a store whose Friday lunch orders are having slower throughput while its labor percentage is climbing. This is an illustrative example, but the decision is familiar: should the operator change staffing, preparation, deployment, or something else?

A labor report can make reducing hours look reasonable. A sales report may suggest the store has enough demand to support its schedule. Neither, alone, explains why guests are waiting.

Now suppose another location had a similar problem. That manager discovered that the new 3PD channel they added had orders arriving right before the walk-in peak, exhausting the prepped ingredients just as the dining room filled, which meant the team was spending the rush catching up on prep. The manager moved preparation earlier and put an experienced person at the station where orders were accumulating. The store is back to normal.

The useful lesson in this example is much more detailed than “add labor” or “prepare earlier.” It includes the delivery pattern, preparation capacity, menu mix, staffing skills, and the point where the operation fell behind.

Even then, we haven't established that the change caused the improvement. Perhaps the following Friday was quieter. Perhaps the promotion ended. A nearby event may have changed the type of orders coming in. A system that recommends the staffing change without checking those possibilities can distribute a bad conclusion faster than people ever could.

The proposed lesson needs to retain those conditions. It should identify genuinely comparable shifts, show what changed, and tell the operator what remains uncertain. Sometimes the honest recommendation will be a small test rather than a confident instruction.

Keep the action and the result together

Software architect Martin Fowler describes event sourcing as capturing changes to an application's state as a sequence of events, and those events can be retained so the system can reconstruct what happened. Restaurant operating knowledge needs something beyond that basic record: why a change was proposed, who authorized it, whether it happened, and what followed.

Back to the Friday example. A recommendation to move prep earlier is not evidence that anybody changed the schedule. An updated schedule doesn't prove the shift was staffed as planned. A faster ticket time doesn't prove the store improved if accuracy suffered or waste increased.

The system needs to follow the decision far enough to establish an outcome worth learning from. Where an action is already authorized, it should carry it out and check the result. Where the operator needs to decide, it should provide the relevant evidence without handing them another research project.

Franchisee privacy also has to survive the process. One operator's payroll or exact sales should not become another operator's open book. We need permission to learn across businesses with limits on what can be shared, and a way to make a useful lesson available without exposing unnecessary private information.

Those permissions and sharing limits must be built into the system.

What superintelligence should do for an operator

For us, superintelligence for QSR means a company can use far more of its operating experience than any individual could hold in their head. It can examine specific conditions, find relevant precedents, help take an action, and keep learning from the result.

This is the opportunity we care about at Clave. The operator doesn’t need to know which specific franchisee encountered the same problem last month, which call discussed it, or where somebody saved the follow-up, but the knowledge should be available when the decision needs to be made.

A successful system would also spare that operator from repeatedly manning the answer. Once a routine is understood, authorized, and reliable, it can happen without pulling the manager back into the same work every week.

If that works, the next Friday lunch could start with preparation ready and the team deployed properly, leaving the manager more attention to coach the new employee and notice the guest waiting at the counter. That is how we want the franchise's accumulated experience to reach the restaurant.