Why your board keeps asking about AI

Your CEO walked out of the last board meeting and asked you, the CFO or COO or owner-operator, "What's our AI strategy?" You probably mumbled something about "looking into ChatGPT." The board nodded politely. The meeting moved on. And then you opened ChatGPT and typed "how should a restaurant use AI?" and got back a 600-word answer that was mostly garbage.

You're not alone. Every multi-unit restaurant operator I've talked to in the last six months has had the same conversation. The board read a McKinsey report. The accountant heard something at a conference. The Toast rep mentioned AI "capabilities." Everybody is asking. Nobody has a real answer.

Here's the thing. You're not behind. The category is empty. Most restaurant operators are not yet using AI in their business in any meaningful way. The ones who are, are mostly pasting spreadsheets into ChatGPT and calling it a strategy. That's not using AI. That's having a chat with a robot.

But there are three things you can do this weekend — actually do, with your own hands, on your own computer — that count as using AI in your business. Each one takes about 20 minutes. None of them require an engineer, an analyst, a consultant, or a budget approval. By Monday morning you can walk into your GM huddle and have a real answer to "what's our AI strategy?" that goes beyond "we're looking into it."

What "using AI" actually means in 2026

Before we get to the projects, let's level-set on what AI actually does. Because most of the noise you're hearing is from people selling you something.

AI in 2026 starts with two things and grows into many.

The two starting moves. First, AI reads numbers and tells you what's worth looking at. The same way your old DM used to scan the daily reports and tell you which stores to call — except it does it in 30 seconds, for every store, every day, without missing a shift. Second, it answers questions in plain English. You ask "how was food cost across all stores last week?" and it pulls the answer in sentences, not in a chart you have to interpret. That's where everybody starts. That's what this Saturday is about.

Where it goes. Here's where it gets interesting. Once your AI has access to more than just the numbers — once you've uploaded your recipes, your operating playbooks, your training docs, your brand standards, your core values — the same tool stops being a faster DM and starts being something closer to an army of agentic employees who never sleep. The AI can answer not just "which store had a bad week" but "which store had a bad week, and based on our playbook, here's the three-step fix the GM should be running, and here's the script for the call you should make to them this afternoon." The AI knows your business the way your best DM knows it — except it knows every store, every shift, every menu item, every operating standard, all at once. That's not 2030. Operators are doing this in 2026 right now. It's the chef's knife on Saturday, and a brigade in the kitchen by next quarter.

Everything else you're hearing about — AI-powered menu engineering, AI dynamic pricing, AI voice-ordering — is some specific application of these same fundamentals. The hype machine has just dressed each one up with a new label.

What you need on your computer before Saturday morning

A few specifics. Don't overthink this.

Claude desktop app or ChatGPT desktop app. Both are free to try. Both work fine for our purposes. I'd pick Claude if you want plain-English business answers; ChatGPT if you want it to be a little more proactive. Either one is the chef's knife. You don't need both. Pick one and learn it.

Your weekly reports, exported as something you can share. A PDF is fine. A CSV is better. A screenshot is acceptable in a pinch. Toast, Restaurant365, Crunchtime — whatever your stack is — every system can export a sales-by-store report or a labor-cost report for last week. Most operators already have these sitting in their email from Monday morning. Pull last week's.

A quiet hour on Saturday morning. Coffee, optional but recommended. Phone face-down. This isn't going to take long. It's going to feel weird, the way using a new knife feels weird the first time.

That's the whole setup. No accounts to create with other vendors. No integrations to wire up. No IT involvement. We're going to use AI the way an operator uses it — with what's in front of you.

Project 1 — Have Claude read last week's sales report and tell you what to look at

Twenty minutes. The highest-leverage AI move you can make.

Open Claude. Drag last week's sales-by-store report into the chat window. Type:

I'm a restaurant operator with [X] stores. This is last week's sales report across all my locations. Look at it and tell me three things: which store had the worst week and what's the most likely reason, which store had the best week and what's worth studying about it, and which store's numbers look suspicious in a way that needs a phone call. Be specific. Use the store names.

Hit return. Wait 15 seconds.

What you'll get back is something like a really good DM's read — except this DM has seen every store's numbers, isn't tired, and has no political opinions about which GMs to defend. The report will name specific stores. It will give you specific reasons. It will tell you which stores need a phone call this week.

Will it be perfect? No. It will get some things wrong. It doesn't know that the Lubbock store had a water main break last Tuesday. It doesn't know that the new GM in Plano just got divorced and is having a bad month. You'll need to translate. But you would have needed to translate anyway — even reading the report yourself, even talking to your DM. The AI gives you a clean first pass.

The first time I watched a 50-store operator do this, he stared at the screen for about ten seconds and then said, "Why didn't anyone tell me about the Sunday dinner drop at the Sherman store three weeks ago?" He'd had the report in his email every Monday. He'd never sat down with it the way Claude just did. The report was always there. The pattern was always there. Nobody, including him, had pulled the thread.

Project 2 — Draft your Monday GM email

Fifteen minutes. The most useful thing AI does for time-poor operators.

Same Claude window. Same uploaded report. Now type:

Write me a draft email to my district managers for Monday morning. It should be 4-6 sentences. It should call out the two stores that need the most attention this week and what to look at. It should mention any positive trend worth celebrating. It should end with a specific question I want each DM to answer by Wednesday. Use the actual store names and numbers from the report.

Wait 20 seconds.

The draft you get back is not your final email. It might use a tone you don't like. It might miss context Claude doesn't have. But it's better than the blank page you'd otherwise be staring at on Sunday night. Edit it. Send it. The whole exercise took 15 minutes; writing that email from scratch usually takes 45.

The trick — and the reason this matters more than it sounds — is that the email forces a discipline. Every Monday now, you know exactly what your DMs are being asked about. Every Wednesday, you know what they answered. That feedback loop, by itself, runs better stores. AI didn't make the loop. AI just removed the 30-minute friction that kept the loop from happening every week.

Project 3 — Ask the question you've been meaning to ask for six months

This is the one I want you to actually do. Even more than the first two.

Every operator I talk to has a question they've been meaning to ask for six months and never had time to. "Are my busiest stores actually my most profitable?" "Is my labor variance worse on Saturdays or Sundays?" "How many of my GMs have been at the same store for more than 18 months, and does it correlate with sales?"

You know the question. You've been telling yourself for months that you'll get to it when things slow down. They don't slow down.

Pull whichever reports are relevant. Drop them into Claude. Ask the question. Plain English. The way you'd ask your CFO if you had one. "Look at these reports and tell me whether my busiest stores are actually my most profitable, and if not, which ones are over-rotating on sales without converting to profit. Name names."

What you'll get isn't a board-ready slide. It's a first answer. Sometimes it'll be wrong. Sometimes it'll surprise you. Either way, you'll have done something this Saturday that's been on your list since November.

That's what counts as using AI in your business. Not the consultant pitch. Not the dashboard. The specific six-month question, asked and answered, on a Saturday morning, by you.

What you just did, and what it doesn't get you yet

You did three things in about an hour. You read a report better. You wrote your DM email faster. You answered a question that's been bugging you for half a year. That's real. It's not nothing.

But it's not the whole game either. Notice what's missing. Every Saturday, you still have to find the report. Drop it in. Type the question. The AI isn't doing this on its own. It isn't watching your business while you sleep. It isn't alerting you Monday morning that the Lubbock store's Sunday lunch service is trending bad three weeks in a row. You are still the one driving.

This is the difference between "using AI like a consumer" — which is what we just did — and "using AI in your business." The consumer version is your Saturday morning. The business version is the AI watching your data continuously, flagging the patterns before you ask, surfacing the questions you didn't think to ask, and answering them in plain English on a tab you keep open all day. You stop doing the lifting. The system does it.

To get there, the data has to live somewhere the AI can read it directly — not in a PDF you're dragging into a chat window. That's a bigger conversation than this weekend's project. We'll write about it separately. For now, the chef's knife is enough. Most operators don't even own the chef's knife yet. Start there.

Frequently Asked Questions

What's the difference between Claude and ChatGPT for this?

Both work. Claude tends to give cleaner business-language answers without much prompting; ChatGPT tends to be more proactive about asking follow-up questions. If you've used neither, start with Claude. If you've already got ChatGPT loaded, stay with it. The difference is real but small for this kind of use.

Is it safe to upload my sales data to Claude or ChatGPT?

For most multi-unit operators, yes — with one setting to check first.

If you're on the free or consumer version of ChatGPT or Claude, both platforms let you opt out of having your chats used to train their models. In ChatGPT, go to Settings → Data Controls and turn off "Improve the model for everyone" (the exact wording shifts, but it's in that area). In Claude, go to Settings → Privacy and check the data-sharing toggle. Flip both off before you upload anything you care about. Takes about 30 seconds.

If you're on a paid business version (ChatGPT Team or Enterprise, Claude for Work), that setting is on by default — your data isn't used for training and isn't visible to other users. If your CFO eventually wants to run AI more seriously across the team, the business tiers are worth the $25-30 per seat per month for the cleaner data posture alone.

Even with the right settings, the practical rule is: it's fine for individual reports — sales by store, labor variance, food cost variance, that kind of thing. Don't upload your full customer database, your investor financials, or anything you wouldn't share with a vendor. If you're publicly traded or have specific compliance requirements, run it past your accountant or legal counsel before getting in the habit.

How is this different from what Restaurant365 or Toast already give me?

Restaurant365 and Toast give you reports. The reports are clunky, the dashboards are buried three menus deep, and every chart looks like it was designed in 2014. You know this. You've sat in the Toast back-office trying to find last week's labor-by-daypart and given up after the fourth click. You've waited 40 seconds for a Restaurant365 page to load just to see one number.

What you just did with Claude on Saturday is the opposite. You asked a question in plain English and got an answer in plain English, in 15 seconds, with the specific store names called out. That's not what either of those platforms does today. Some are bolting AI features on top of their dashboards in 2026 — a chat sidebar here, a "smart summary" button there — but the core experience is still a 2014 reporting tool with AI lipstick.

Here's the bigger thing nobody is telling you. Toast sees your POS data. Restaurant365 sees your accounting and inventory. Crunchtime sees your back-of-house labor and recipes. Your scheduling tool sees the schedule. Your guest feedback platform sees the reviews. Each one of these is a slice. None of them sees the whole picture. The minute you ask a question that spans two systems — "did our labor variance correlate with the menu changes we rolled out last quarter?" — every one of these tools throws its hands up. Their AI features can only think about the data they own. They can't think across your business because they don't have your business in front of them.

What you did on Saturday with Claude works for one report at a time. The next level — the one your competitors are quietly figuring out — is when AI has every report, every system, every operating doc, all at once. That conversation comes later in this series. For now, just know: the dashboard your existing tools give you is the floor, not the ceiling.

Can I do this for inventory and labor reports too?

Yes. Same exact pattern. Drag in the report, ask Claude what's worth looking at. Labor variance reports are particularly good for this — Claude is excellent at spotting which stores are out of pattern.

My CFO wants to do an "AI initiative." Is this enough?

No, but it's the right first move. Tell your CFO you spent Saturday morning using AI to find a six-month-old blind spot in the business. That's a credible thing to say in a board meeting. Tell them you want to talk next quarter about the bigger version — wiring AI into the data layer so it's running continuously, not just on Saturdays. That conversation happens later. Get the Saturday version working first.

What if I want AI to actually be running in my business continuously, not just on Saturdays?

That's the bigger conversation. The short version: your data needs to live somewhere AI can read it directly, with the right structure to be useful. There are a handful of platforms in restaurant operations that handle this — pre-connected to your operational stack, with the data formatted so AI tools can answer questions about it live, across every store, without you uploading anything. Look for one that integrates with the systems you already run. For now, your Saturday morning project is enough.

Where can I learn more about the basics of using AI in business?

Two places worth trusting. Anthropic's documentation at docs.claude.com is good for understanding what Claude can and can't do. The original Bourdain chapter on cooking like a pro from Kitchen Confidential — for a completely different reason, but it's the same teaching mode and it's worth reading if you're trying to figure out what good practical instruction sounds like.

The hardest part of using AI in your restaurant business isn't the technology. It's deciding to spend an hour on Saturday morning with a tool you haven't tried yet. The chef's knife is the same way. Buy the knife. Practice on a few rutabagas. By the third one, you'll know what you're doing.