Deloitte: AI Just Crossed From Pilot to Production in Restaurants

I open the Deloitte landing page in a booth at the back of a midtown bistro, somewhere between the lunch wind-down and the dinner prep huddle. The owner is squinting at a labor report on his phone; a runner is restocking ramekins. The survey loads. State of AI in Restaurants. Three hundred and seventy-five executives, eleven countries, one tidy headline: AI investment is about to spike.

The headline number — 82% of restaurant execs plan to increase AI spend in the year ahead, 73% “somewhat” and 9% “significantly” — will be quoted in every vendor deck between now and the next NRA show. It deserves to be. But the more interesting line, the one that actually tells you what’s happening on the floor, is buried a few pages in: only 9% of those same execs say their teams use generative AI on a daily basis. That gap — between the budget line and the routine — is the real read for anyone trying to figure out whether 2026 is the year AI stops being a slideware concept and starts pulling weight at the pass.

The budget line of 2026

Deloitte’s framing is unusually direct for a Big Four survey. The report describes AI as having “crossed from pilot to production,” and the spend pattern backs it up. Of the 82% planning to invest more, the bulk are dialing up existing programs rather than starting fresh. Sixty-three percent are already using AI to automate the guest experience — drive-thru voice, recommendation engines, loyalty personalization, chatbot reservations. Fifty-five percent are running it through inventory and forecasting. Those aren’t moonshots. Those are line items that will show up in the FY26 budget memo and survive the first round of cuts.

The category mix matters because it tells you who is winning the procurement fight. Guest-facing AI is sticky because it shows up in the marketing report. Forecasting AI is sticky because it shows up in food cost. The work that doesn’t make the top of the chart — back-of-house labor scheduling, prep-list automation, smart waste tracking — is the same work most operators tell me is the most painful manually. That misalignment is going to define which vendors get a third meeting this fall and which get a polite “circle back in Q1.”

Mark interpretation: the 82% number is not really a forecast. It is a permission slip. Boards and PE owners have spent two years asking what the AI plan is. Execs now have a defensible answer — we’re investing, here’s the survey — and that gives them air cover to keep the existing pilots breathing through the next budget cycle. Whether those pilots earn their keep is a separate question.

What “daily use” actually means at the pass

The 9% figure is the one I keep circling. Deloitte is measuring whether generative AI shows up in the daily workflow of an actual restaurant employee — not whether the CIO has a ChatGPT Enterprise seat. Nine percent is a small number, but it is not zero, and the composition is telling. Most of the daily users sit in marketing, menu R&D, and guest-comms teams at chains with more than 250 units. The line cook in unit 814 is, statistically speaking, not yet talking to a model.

That tracks with what Toast’s 2025 Voice of the Restaurant Industry survey found earlier this year, which I’ll be unpacking in a forthcoming desk review: operators are bullish on AI in the abstract and cautious about it in the specific. Toast’s own blog rundown of how restaurants are using AI today reads less like a revolution and more like a punch list — menu copywriting, review-response drafting, basic forecasting nudges. Useful work. Not yet transformative work.

The gap between Deloitte’s 82% spend signal and the 9% daily-use signal is where the next twelve months get interesting. Either the spend pulls usage up — execs force adoption through workflow integration and training — or the usage number stays flat and the spend gets quietly reclassified as R&D in the FY27 deck. I have seen both endings before.

What I’m watching from here

Three things I will be tracking in the back half of 2025 and into 2026.

First, the labor-side rollouts. Guest-experience AI is the easy win because it lives on the marketing P&L. Labor-side AI — scheduling, prep, training — lives on the operations P&L, where every dollar is contested. If the 82% spend signal is real, we should start seeing labor-side announcements from the top 25 chains by Q4.

Second, the format question. Sweetgreen’s Infinite Kitchen is the most aggressive bet that automation and AI converge at the unit level rather than the cloud level; I am working on an upcoming May piece on whether that thesis holds up at scale. Deloitte’s data does not really capture format-level bets — the survey is a CIO instrument — but they are where the operational economics will or will not pencil out.

Third, the vendor consolidation. Eighty-two percent of execs spending more does not mean eighty-two percent of vendors get paid. It means the winners get bigger contracts and the losers get a polite email. Watch the next two earnings cycles.

The pilot era is ending. The production era — messy, contested, budgeted — is beginning. That is the actual headline.

— Maya covers restaurant tech for TableTransfers. Tips: [email protected].