The 10 AI systems running inside my Chick-fil-A right now

I get asked about AI in restaurants almost every week. Usually the question is some version of “should I invest in AI?” and I never know how to answer that one, because it is like asking whether you should invest in electricity.

So here is a more useful answer: what AI actually looks like inside one working restaurant. Mine. Chick-fil-A Miami Shores, open 5 AM to midnight, six days a week, about a hundred people on the team.

Ten systems run my restaurant with me right now. I did not hire an engineer to build any of them. Every one of them follows the same rule, which I will get to, and that rule matters more than the list.

The ten

1. Daily transaction capture. Every morning it pulls the prior day’s sales and grades the day against our rolling baseline. I stopped guessing whether yesterday was actually good or just felt good.

2. Daily labor briefing. Yesterday’s scheduled hours versus actual punches versus the grade, plus today’s outlook and overtime watch. My managers used to build this by hand when they remembered to. Now it is waiting for them.

3. Cash integrity watch. Reads the overnight sales report for refunds, discounts, cash over/short, and deposit variance, and tracks patterns by cashier. Trust, but verify, every single morning.

4. Catering confirmations. The day before every catering delivery, it drafts a personal confirmation email to the customer. We missed too many details the manual way. That miss rate is what pushed me to build it.

5. Catering thank-yous. The day after delivery, a thank-you note drafts itself for my review, and a recovery note if something went wrong. Every customer hears from us twice.

6. Review responses. Drafts replies to delivery-platform reviews in my voice for the routine ones and routes anything serious straight to a human. Response time went from days to hours.

7. Compliance nudges. Tracks required training for every team member and drafts a friendly reminder when someone is falling behind, with the tone scaled to urgency. Nobody gets surprised by an expired certification anymore.

8. Team care. Every Monday it surfaces birthdays, work anniversaries, guest shoutouts, and new team members who need a check-in, and drafts a personal note from me for each one. Hold that thought. This one is where the story is.

9. Catering radar. Watches our catering accounts for lapsed customers and seasonal patterns and hands me a short ranked list of who to call. Detection only. I make the calls.

10. Forecast tripwire. Once a week it checks whether our scheduling forecast has drifted from reality and flags the days worth fixing. Quiet most weeks. Worth it the week it is not.

The rule they all follow

Every system drafts. A human decides.

Not one of these ten sends an email, posts a reply, or messages a team member on its own without a human review step or a veto window. That is not because the AI is not good enough. It is because the moment my team or my customers figure out they are talking to a machine wearing my name, every note I have ever sent becomes worthless.

The machine does the noticing. The person does the caring. That division of labor is the entire design.

What I actually got back

Everybody wants the ROI number. Here is the honest version: the hours came back first. The morning reporting my managers used to grind through is waiting for them when they walk in. The catering follow-ups that used to slip on busy weeks stopped slipping. The compliance chase that ate a manager’s afternoon every month runs itself.

But the return that matters showed up somewhere else. A while back I asked one of these systems a question I had never thought to ask: who on my team deserves recognition that has not received it?

It gave me a name I did not expect. A kitchen team member who had been coming in early, over and over, to cover a station for a teammate who was out for health reasons. She never mentioned it. She never asked for credit. She also had zero activity in our team chat in thirty days, which means she was completely invisible in every channel I normally look at.

The week before, a different report had flagged the same person as an overtime risk. Same hours. Same data. One question made her a cost to manage down. The other question made her the person holding the kitchen together.

The AI did not change. The question did.

I wrote her a thank-you note in Haitian Creole, her first language, and gave it to her myself. If you want the questions I ask now, I keep the whole list at jonharmeling.com/prompts. Steal them.

Where to start

Not with a tool. With a pain point that repeats every week, and one hour on a Saturday to describe it out loud to an AI assistant. That is genuinely how most of my ten started.

And when the hours come back, and they will, decide on purpose what they are for. Efficiency is what the systems produce. What you spend it on is the part with your name on it.