Your CEO walked into your office last quarter and said, "figure out our AI strategy." Maybe your board said it. Maybe both.
You went home, opened ChatGPT, dropped a labor report into Claude, watched it do something useful, and thought okay, now what?
Then you went looking for advice. McKinsey decks. Vendor pitches. LinkedIn thought leaders. And what you got back was a list that looked something like this: connect your POS, connect your inventory, get an AI assistant on top, ask it questions in plain English, save your analyst some time.
None of that is wrong. We sell exactly that. Half a dozen good vendors sell it too. If you haven't done it yet, do it — it's table stakes now, and your category leaders are already there.
But every pitch ends in the same place. And the place every pitch ends is the same place your competitor's AI strategy ends. Which means if you do what everyone else is doing, the best you've built is parity. Faster, sure. But not different.
There's a second half. Nobody is selling it yet. That's where the real ground is.
What every advisor sells you
Walk into any restaurant industry conference this year and you'll hear the same playbook five times before lunch:
- Audit your data. POS, labor, inventory, guest feedback, delivery.
- Get it into one place — warehouse, platform, dashboard, whatever.
- Layer AI on top so non-technical folks can ask questions.
- Watch analyst workload drop and insight velocity climb.
- Report savings to the board.
True, true, true, true, true. And every operator in your category is hearing the same five steps right now. So when you implement it, you've matched the field. You haven't changed the game.
Here's what bothers me about how this gets pitched. Step 5 is where the conversation stops. The CFO guide ends. The webinar wraps. The vendor closes the deck.
But that's the easy half. The structured half. The half where the data is already in databases somebody else built, where the connectors already exist, where every vendor in our category is racing to add an AI layer on top this year.
The other half is harder, less obvious, and worth more. And no one is telling you about it yet.
Think of agents as employees
You keep hearing the word "agents." Every vendor deck. Every AI strategy webinar. Sooo abstract. Sooo hand-wavy. Half the time you nod along, half the time you wonder if anyone in the room actually knows what they're talking about.
Here's the frame that makes it real. Stop thinking of an agent as a piece of software. Start thinking of it as an employee. One that doesn't sleep, doesn't take vacation, and never asks for a raise. But still — an employee.
The moment you do that, AI stops being abstract.
Because you know how to onboard an employee. You've done it a thousand times.
Think about Day 1 with a new hire. You don't sit them down and say "build me a spreadsheet for this huge project, due Friday." That'd be insane. You'd never do it. Day 1 you onboard them.
You have them read your core values. You introduce them to the department leaders. You walk them around, show them where the files live, who to ask about what. You explain the company vision. The quarterly objectives. The KPIs your team is on the hook for. The history. The wins, the losses, the things that worked and the things that didn't.
Two weeks in, they start doing actual work. Three months in, if they're good, they're fully productive.
Three months. That's how long it takes a smart human to absorb the culture, the vision, the SOPs, the playbooks, the unwritten rules — enough of it to operate without supervision.
Now contrast that with an AI agent.
Feed an agent every culture doc. Every onboarding SOP. Every playbook. Every quarterly objective. Every KPI dashboard. Give it access to the same shared drive, the same daily manager logs, the same training portal you'd give a new GM. Give it the same context a human would need three months to absorb.
It can start doing meaningful work today. Not three months from now. Today.
And here's the part your board hasn't done the math on yet: a human takes a month to onboard. An agent takes a morning. You can onboard three agents a day. Three agents a day, instead of one human a month.
That's not a faster analyst. That's a different cost structure.
What the un-onboarded agent does vs. the onboarded one
Picture two agents at your company. Same model under the hood — Claude, GPT, doesn't matter. The only difference is what they've been onboarded with.
Agent #1, hired on Friday, started Monday, given a POS login and nothing else. Looks at the numbers. Sees Store 14's lunch sales dropped 12% last week. Tells you Store 14 had a bad week. You knew that. Your area manager knew that. The store knew that. You've paid a vendor for a faster way to learn what everyone already learned by Tuesday morning.
Agent #2, hired the same day, but actually onboarded. Read your DM's last six weekly emails. Read the playbook you wrote two years ago for diagnosing a sales dip. Read the daily manager logs from Store 14 — including the one where the GM noted she was about to go on PTO and her second was covering. Read the post-mortem from the last time this happened in 2024.
Same Store 14 dip. Different answer. Agent #2 tells you Store 14's second-string GM is covering, she always struggles with the lunch transition without her usual team, last time this happened recovery took eight days but only after a specific shift change at 11am. It drafts the email to the area manager. It schedules the staffing change in your system. It flags two other stores running the same pattern and likely to dip next week.
Agent #1 told you a number. Agent #2 ran the play.
The difference between them isn't the model. The model is the same. The difference is the onboarding. And the onboarding material isn't in your POS. It's in your shared drive, your daily manager logs, your email, your training portal, the document where you wrote down what makes Store 14 different.
It's the stuff your operators use every day. The stuff nobody has thought to hand to a new hire — because we haven't yet thought of agents as new hires.
Why this is a CFO question, not an IT question
Your CEO is going to ask you the question every CEO is asking their CFO this year: how does AI change our cost structure?
Half-version answer: we were able to reduce analyst headcount and save our managers 40 hours per month. P&L cost improvements of 50-100 basis points. Real. Big. But not transformative.
Full-version answer: we run this company with materially fewer people doing today's work, and we reinvest the headcount into growth. HQ shrinks by half. District managers oversee twice the stores without breaking. GMs stop doing admin and get their hours back for guest experience and waste reduction — the two things only a human in a restaurant can actually do. And we plug all those savings back into growing stores instead of paying it out as the cost of staying flat.
That's a different operating model.
The full-version answer is the conversation your board is actually trying to have with you. They aren't asking whether you bought a dashboard. They're asking whether AI is going to let you run more stores with the same team, or the same stores with a smaller team. The answer depends on whether you've given your AI access to the unstructured operating context, not just the structured data.
Which is why this is your problem, not IT's. IT can plug in data sources. Operators can use a dashboard. But the question underneath all of this is the same question you ask when you onboard a great new hire — what does every employee at this company need to know to be effective?
When the employee is human, you answer that question once a year, slowly, through training programs. When the employee is an agent — and you're onboarding three a day — that question becomes the most important strategic question your company has. What gets written down. What gets centralized. What context lives somewhere an agent can read it, instead of locked in someone's head.
That's not an IT decision. That's a CFO decision. You're the one who owns how this company gets built.
If your AI strategy doesn't have a plan for the unstructured half, your AI strategy is half a plan.
The honest part
Two things I want to be straight with you about, because this is where vendor content gets dishonest.
It isn't easy. Getting your playbooks, your emails, your manager logs, your tribal knowledge into a form an AI can actually use is real work. Most operators in your size band don't have anyone whose job it is to do this. It's not a six-week project you delegate to IT. It's an operating capability you have to build, and then keep building.
You don't have to do it all at once. Companies that get this right start small. One playbook. One weekly email. One area where the gap between great and average operators lives in someone's head and nowhere else. Get that one thing into a form an agent can use. See what happens. Add the next one.
The CFOs who win the next five years won't be the ones who ran the fastest dashboard rollout. They'll be the ones who figured out that AI strategy is, at the bottom, a what do we centralize question — and started treating the unstructured operating context as a strategic asset.
What Expo does
Expo is the data layer that connects everything a restaurant operator runs on — POS, labor, inventory, guest feedback, delivery platforms — into one AI-native system you can ask questions of in plain English. We work with multi-unit operators from 20 to 200+ stores. Pricing runs $99–$250 per location per month.
Why this piece, from us: most of what gets sold as "AI for restaurants" is the structured half. That's what we sell too. It's table stakes. The work that actually changes the cost structure is the unstructured half — the playbooks, the emails, the institutional knowledge — fed into agents alongside the operating data. We help our partners figure that out. We wrote this piece because the argument deserves to exist in writing, in plain language, where any CFO can read it.
If you're a CFO working through these questions for your company, Book a demo and we'll show you what we're seeing across our partners.