Restaurant groups automate the loud thing first. The money is quieter.

On a Friday at 6:40, the host stand at a Chicago restaurant group is doing three jobs at once. Seating a walk-in, holding a line at the door, and calling the pass about an eight-top that is somehow both confirmed and not on the book. None of that is a technology problem yet. It is a process problem that technology can help with, but only once someone can actually see the whole line.

That order matters, because the industry is about to spend a lot of money in the wrong place.

Before you rent another tool, settle what a restaurant should own: the guest record question

The leak is labor and empty seats, not the menu

Start with where the cash actually goes. For full-service operators, salaries and wages including benefits ran a median of 36.5% of sales in 2024, well above the historical average, and staff turnover sits at 75 to 80% a year. Labor is the biggest controllable line on the plate, and most of it is scheduled against a guess.

The other leak is the seat that never arrives. No-shows are a small percentage of covers on a normal night, but they cost the global industry an estimated 16 billion dollars a year, and a booked table that ghosts you is lost twice: the cover, and the walk-in you turned away to hold it.

Against that, look at where restaurants are actually pointing AI. About a quarter of operators now use it, 26% by the National Restaurant Association's count, and the top uses are marketing, review replies, and menu copy. All fine. None of it touches the two lines that decide whether the month works.

When we map a restaurant group's operating week, the same four rooms always show up: the book, the pass, the schedule, and the invoice pile. Fix those, in that order, and the automation pays for itself. Here is how we sequence it.

The five workflows, in the order we would fix them

1. The reservation and no-show flow. Confirmations live in the booking tool, the waitlist lives in someone's head, and deposits are a policy nobody enforces on a busy Friday. Before any AI, make the flow one loop: booking, reminder, waitlist backfill, and a deposit rule for large parties. Automated reminders and waitlist logic can cut no-shows sharply, and reminder systems have been shown to reduce them by up to 90%. That is covers you already sold, recovered.
2. Prep and kitchen visibility. The prep list lives in a notebook, and the line finds out what it is short on at 7pm. A demand forecast off your own reservation and sales history turns into a prep sheet the day before, and a shared view of what to fire and when. This is where AI-assisted forecasting earns its place, because the input is data you already own.
3. Order and ticket routing. Tickets cross stations, modifiers get lost, and an 86 on the line does not reach the host or the app for twenty minutes. Fix the routing rules first. Automate the propagation second. A clean routing map beats a smart kitchen display bolted onto a messy one.
4. Inventory and vendor coordination. Counts are done by hand, invoices sit in a pile, and variance surfaces weeks late when the P and L lands. Matching invoices to receipts, flagging price variance, and forecasting the order are all repeatable enough to assist or automate, and they protect the food cost line the same way scheduling protects labor.
5. Staff scheduling. Coverage is guessed, then patched with texts. Tie the schedule to the same demand forecast that drives prep, and you are matching hours to covers instead of to last week. On a labor line running north of a third of sales, a few points of accuracy is real margin.

The two that look like AI wins and are traps

Two projects always come up first in the pitch deck, and both are premature.

The first is a guest-facing AI concierge, a chatbot that answers questions and takes bookings, layered on top of a reservation process that is not yet one loop. It answers faster on top of the same broken flow, and now the mess has a friendly voice. Clean the book first. Then a booking assistant is genuinely useful.

The second is full kitchen automation before the process is legible. If the prep, routing, and 86 flow are not written down and consistent, automating them just runs the mess faster and hides it behind a screen. You end up paying to scale the exact thing you needed to fix.

Neither is a bad idea. Both are the right idea in the wrong order.

AI does not fix a broken process. It runs it faster. That is a good thing only after the process is worth running.

Sequence beats software

The pattern under all of this is boring and it works. Map how the week actually moves, not how the org chart says it does. Find the friction: the handoffs, the queues, the exceptions, the timing. Decide, workflow by workflow, what to automate, what to assist, what to route, and what to just measure. Then build the execution path and put AI where the work repeats.

Done in that order, AI stops being a line item you hope pays off and becomes a layer on a process you can already see. Done in the other order, you buy software to run a mess, and the mess is what you were trying to sell your way out of.

If we had one week with a Chicago group, we would not build anything. We would map the book, the pass, the schedule, and the invoices, and show you exactly where the covers and the hours are leaking. The automation comes after that, and it comes cheaper.