Here's how I cut my restaurant labor by $103,000 using AI

From January through September, hourly labor at my two restaurants went from 21.1% of sales to 16.2%. Sales were down 5.5% over the same months. (We were up for the first four months of the year and then had a pretty rough summer.) At last year's rate, this year's hourly wages would have come to about $444,000. They came to $341,000.

I didn't fire anyone to do it.

Here's how, in five steps:

The conventional wisdom in restaurants is for labor to run at 30% of sales. If you're alcohol-heavy that can run a little lower. If you're fine dining it may run a little higher.

My managers wrote the schedule with basic common sense. If it ran high, we adjusted the next one. We staffed up when we expected more business. I looked at the numbers after the fact and said "this is fine" or "we need to tighten up." The schedule never had a real budget behind it. I figured budgets were for big groups with more resources. It’s like… spray and pray for restaurant operations.

The question a lot of us ask is, "What should labor be in a restaurant?" A more useful question is, "What labor do I need to operate my restaurant?" Better still: "What human labor do I need to consistently produce the best sales in my restaurant?"

This is what some people call thinking from first principles. It puts you in a position to throw out industry standards and vague numbers and force yourself, as the owner, the operator, the person in charge, to ask, "What do I need to produce the outcome I'm looking for?" I decided 30 was out the window.

Now let me just say: this is an article about how AI helped me save money in my business. If you skip the first-principles part and go looking for a magic bullet, you may be disappointed at the end. In a practical sense, AI removes friction, dramatically increases speed (the speed of data, the speed of information) and shines a light on data and insights we were previously in the dark about. You still have to apply it to real choices and real conditions in your business, and that's what this is all about.

The number I built around was sales per labor hour (SPLH), your sales divided by the hourly hours you paid for. Across both stores we went from $60 in the first nine months of 2025 to $75 this year. There are two ways to improve it: cut labor or raise sales. Doing both is ideal.

Reverse engineer the goal. Take the sales you expect, divide by your goal, and that's how many hours you can schedule. A $10,000 day at a $75 goal gets 133 hours. My advice is to make it feel impossible. It needs to be aggressive enough to make you re-think your processes. If you say you want to shave 10% you’ll continue working within your existing framework. If you push your mental construction of your process until it breaks, you’ll know there’s the possibility of a breakthrough coming.

My impossible goal was 20% labor, total, including salaries and taxes. (I also set a goal of 20% COGS in order to run a 40% prime cost, but that’s a piece for another time.)

Breaking the schedule into a percentage of sales for each job code tells every manager where they can place their resources: bartenders, barbacks, line cooks, BOH prep, FOH prep and managers. (We don't have servers.) Then the schedule gets built to hit those numbers.

Parkinson's law says work expands to fill the time available for it. Labor does the same thing: schedule a shift more hours than it needs and it will use them. So we decided we were no longer going to let labor fill the space, and we forced ourselves to live within the goal.

When you say you're hitting that number come hell or high water, it forces you to rethink how you open, how you close and how you prep. Simplify. We did change some processes to get there. Your sales set the schedule, and it's up to you to build a restaurant that can run that way.

So how did that translate? We set three constraints. We had to hit the sales-per-labor-hour goal. We were not going to compromise on quality. And we were not going to fire anyone.

The biggest area for improvement was prep. We had already centralized prep for both restaurants, because running two prep teams for two stores with the same menu made no sense. But I'd left the prep schedule up to the team. They came in every day, looked at what they needed to make and made it. Wildly inefficient to start and stop a dozen items each day with cleanup and setup in between. Henry Ford would have slapped my face.

Instead we started making larger batches every other day, with no compromise in quality or freshness. You can prep 10 gallons of sauce in about the same time it takes to prep 5. Quality maintained, hours dropped. We also started having the cooks prep more produce on the line. I'm embarrassed to admit we weren't doing enough of that before.

Then a cook left on good terms, and we didn't replace the role. We added hours for the other cooks so the busiest hours stayed covered, and we adjusted the closing duties.

Those are the practical results of living inside the constraint. These aren’t the only changes that were made, but they’re representative.

Here's a wild fact. For us, almost the same number of people can handle a $1,000 hour and a $3,000 hour. At one of my stores this year, a $1,000 to $1,500 hour averaged 6.6 people on the clock. A $2,000 to $3,000 hour averaged 7.6, and hourly labor in those hours ran 3.2% of sales. A restaurant has a floor, the people you need just to function, and the sales that same crew can handle go a long way above it.

The floor is also why a slow day looks bad. A Monday can run 37% because the way we're set up you still need two cooks, and FOH prep and the porter take in the week's orders and do repairs that day. That's a week of work landing on one slow day. So I judge labor for the week in total, and you can "buy" a better weekly number with your best sales days.

It's stressful to see a slower week and still be asked to hit the improved number, and my management team let me know that. To give them some peace of mind, I added a running average and a year-to-date number. They'd beat the goal regularly. Then sales would dip and they'd decide we were screwed. The running number showed them they had a little room to borrow from. In the slow parts of the year there isn't much you can do. You still need cooks.

For almost a year I got up at 4 AM to focus only on this and on building my other tools. Three or four hours a day while my house was still asleep, going through labor and sales reports, bank records, P&Ls and invoices until I understood where every dollar went.

I did it the analog way. We ran a punch check and went through every punch, every day. Then I checked sales per labor hour against the schedule and against actual, every day, to make sure we were on track. I built spreadsheet tools for it that I'd never seen anywhere before. Alone in the early morning, it's just you and the numbers and the stark sense that you have to do something about it.

Hours dropped about 25%, roughly 9,200 for the nine months, and overtime fell by half. Labor came down while sales were up early in the year, and the savings kept adding up through the summer, which was brutal and would have hurt real bad if we were still working with our previous setup. I think we’re running labor that actually works for these businesses, but I constantly tell myself not to get complacent again.

Once the system worked by hand, I moved it to Claude one piece at a time. The first was the punch tracker. Every morning it emails me who clocked in early or late and what it cost. Then I built the labor budget tool by job code, with budgeted, scheduled and actual labor side by side. Now it's a dashboard I can look at any time, with my 7-day and 30-day averages.

I'd like Claude to fix the punches too. I haven't been brave enough to try that yet, so I still click the approval button.

Set an impossible sales-per-labor-hour goal, actively rethink and reformulate your processes, give every job code its share, build the schedule to hit it and check actual against it every day. Guarantee is a strong word, but if you follow this process I can't imagine not being able to find meaningful and maybe impossible-seeming improvements in your labor numbers at your restaurant.

If you want help setting this up in your restaurant or business, answer six quick questions and grab a time on my calendar.

CC

Figures are hourly wages from Toast, January through September 2025 vs 2026, both stores combined. Salaried managers, payroll taxes and tips aren't included.