How AI Can Help Restaurant Operators Spot Workers’ Comp Risk Sooner

Workers’ comp claim files often contain more safety information than operators can use in real time.

A manager may record where an injury happened. An adjuster may note the task the employee was performing. A medical record may identify the body part involved. Another note may mention equipment, footwear, a spill, a missing guard, or a shortcut that created risk.

Those details can point to prevention. The problem is that they sit in text, making it easy to miss one claim at a time.

In a recent Self-Insurance Podcast episode, CRMBC CEO Kaya Stanley spoke with Benjamin Burch, a data scientist and associate vice president at Midwest Employers Casualty, about how AI can help workers’ comp teams process that kind of information.

Burch’s framing is practical. Midwest Employers Casualty does not treat AI as a replacement for people. Internally, Burch said, the company thinks of it as assisted intelligence rather than artificial intelligence.

AI can surface a pattern. People still decide what the pattern means.

Why claim notes matter for restaurant operators

Most operators already track the obvious claim details: date, location, injury type, cost, status, and whether the employee returned to work.

That data is useful, but it rarely explains why the injury happened.

The explanation is more likely in the note field: the surface where the employee slipped, the equipment involved, the station where the cut happened, or the safety step that was missed.

A single claim can look routine. Five similar notes from the same work area can point to a training gap, a layout problem, an equipment issue, or a documentation habit. The sooner that pattern is visible, the sooner an operator can check what is happening on the floor.

Burch called this unstructured data, meaning information that does not fit neatly into a spreadsheet. Structured data can be counted quickly. Unstructured data has to be read. That is why useful safety clues can sit in claim notes without ever reaching the next safety conversation.

OSHA’s restaurant safety materials list common hazards such as burns, cuts, slips, trips, falls, strains, and sprains. Operators know these risks. The value of claim-note review lies in seeing whether one of them keeps recurring in a specific part of the operation.

How AI can help review claim notes

Burch gave a specific example from claims analysis.

His team reviewed a customer’s claim notes to identify incidents in which a safety procedure may not have been followed. The goal was not to prove a violation. It was to flag claims worth a closer look.

The review surfaced employees cleaning a machine without the proper guard, employees not wearing the right safety equipment, and employees standing on rolling chairs to reach a shelf. Those are operational details a manager can act on.

The AI-assisted pass took about 15 minutes. Reading the notes for roughly 500 claims by hand could have taken half a day, and a reader could still miss something.

That faster first pass makes a broader review possible. A claims team can review more history. A safety partner can compare locations. An operator can walk into a claims review with sharper questions. A manager can see the same unsafe behavior described in different words across different files.

AI does not have to make the decision to be useful. It can organize the evidence so that the people closest to the claim can review it more quickly.

A small claim note review operators can try

Start small.

Pick one contained question. For example: do our recent claim notes show safety issues that repeat by task, equipment, location, body part, or behavior?

Gather a small, approved set of de-identified claim summaries or incident notes. Remove names, medical details, legal comments, and anything your organization has not cleared for AI review.

Use a focused prompt:

Treat the output as the start of a review, not the conclusion. Compare the themes against manager notes, training records, maintenance activity, staffing levels, shift timing, and location history. If the tool points to a possible pattern, use it as a reason to look closer, not as a finding.

Where AI should stop

Burch drew a clear line between review work and decision-making.

AI fits work that involves scale and repetition. It can review large files, strip duplicate information, summarize records, and extract details from text-heavy documents.

Other decisions do not belong in a model. Future medical treatment, compensation discussions, negotiation, claim resolution, and conversations with injured workers stay with people because they turn on context and care.

This line matters more in workers’ comp than in many other settings. A claim file is a record set that also represents an injured worker, an employer, a medical process, and a claims process. AI can help organize the file. It should not move responsibility away from the people who make the call.

Be careful with claim data

Burch said Midwest Employers Casualty established governance and security measures before allowing employees to use AI tools. When sensitive information is involved, the company uses private implementations rather than public versions of AI tools. The sequence is the point: rules first, access second.

Claim information can include private employee details, medical history, legal exposure, and sensitive business records. Operators should not paste claim files, medical notes, employee records, or legal comments into public AI tools unless that use has been approved.

A safer starting point is a small, de-identified set of incident summaries or general claim descriptions. Even then, an operator should know which tools are approved, who is allowed to use them, and who reviews the output before anyone acts on it.

NIOSH has published guidance on managing workplace risks tied to AI. The NAIC has issued guidance for insurers on AI governance, risk management, and internal controls. Both support the same practical rule for operators: set the rules before using AI with sensitive information.

What operators should take from the conversation

Start from a claim or safety question the operation already has. A location with repeat slips. A station with repeat cuts. A piece of equipment that keeps showing up in claim notes. Incident reports too thin to help the claims team later. A claims review that drags because the useful details are scattered across PDFs and notes.

AI can help sort that information, group the themes, and point to where a manager should look next. When a piece of equipment keeps appearing, that is a reason to review training, guards, maintenance, and supervision. When slips cluster in one area, that is a reason to look at matting, drainage, cleaning schedules, footwear, and traffic flow.

It may also show that the answer is not an AI project at all. Sometimes the fix is better documentation, better training, a cleaner reporting process, or a more focused claims review.

Burch’s closing reminder is worth keeping in view.

“It’s still, at the end of the day, just a tool.”

AI may point an operator toward the guard, the matting, or the training record faster than a manual read. The operator still has to check the issue, talk with the team, and decide what needs to change.