Wingstop's AI Smart Kitchen Cuts Food Waste 34 Percent and Ticket Times in Half — and Just Did It at 2,586 Restaurants Simultaneously.

THE CRAFT

Before the Smart Kitchen, this was the workflow.

A customer placed an order — at the counter, on the app, on DoorDash. A paper ticket printed in the back of the house, on a printer that may or may not have had toner. The ticket got clipped above the wing station, or it fell behind the countertop, or it sat in a stack of other tickets the crew was working through. Crew members eyeballed how much chicken to drop, how many fries to cook, how many tenders to par-cook against what the next ten minutes might bring. There was no system telling anyone what was coming. There was no system flagging that the order was running late. There was no system spotting that the line had thirty tickets in queue and the labor schedule had two people on the floor. On a Friday night, on Super Bowl Sunday, on the first Saturday after a winter storm when everyone in three zip codes wanted wings delivered, the time from the printer to the front-of-house bag could push past thirty minutes. At a thousand stores, it pushed past forty-five.

That was the operational reality for Wingstop for twenty years. The food was always fresh-cooked. The menu was deliberately constrained — fourteen wing flavors, bone-in and boneless, a short list of sides. The brand promise was simple. And the kitchen was being held together by line managers who had built an instinct for what was about to walk through the door. It worked at the unit volume the company was running. It would not work at the unit volume the company wanted to run.

Michael Skipworth’s Smart Kitchen — deployed across all 2,586 domestic Wingstop restaurants in ten months during 2025 — is what replaces that workflow. It is, by my read of the trade press, the SEC filings, and the analyst commentary, the most operationally complete enterprise AI rollout in the QSR sector right now. It is also one of the few mid-cap AI deployments I have seen this year where the system architecture, the operational metrics, the deployment cadence, and the executive’s discipline about what the AI is and isn’t for are all visible enough to write about honestly.

This issue is about that deployment. What the system actually does. How it was built. Why Skipworth chose to ship to every store in ten months instead of running a three-year pilot. What the operational numbers look like at the restaurant level, in the dayparts that matter, and at the cohort the company runs itself. The story is the AI. Near the end, in a single honest paragraph, I’ll address what is happening to Wingstop’s same-store sales right now in the broader QSR consumer environment — and why I think the AI is part of what is keeping that pressure from being much worse.

THE OPERATOR

The situation

The pre-Smart-Kitchen Wingstop ran on the same architecture every quick-service restaurant in America runs on, with one wrinkle: the food is cooked fresh, to order, and the operational tolerance for chaos is lower than it looks. Wings have to be par-cooked at a defined point in the demand curve. Bone-in and boneless run on different cook timers. Fourteen wing flavors mean fourteen sauce pans being maintained at the line. Sides — fries, corn, cheese sauce, ranch — have their own portion-discipline problem: a crew member eyeballing the fry-basket fill is the difference between a profitable order and a write-off.

The original system for managing all of that was a paper ticket and a clip above the wing station. When the line manager was experienced and the lunch rush was predictable, the system worked. When the line manager was new, or when a Sunday football game and a third-party-delivery surge hit the kitchen in the same fifteen minutes, the system did not work. Tickets fell behind countertops. Sauce flavors got mixed up. Crew members par-cooked too much chicken in anticipation of a rush that did not come, or too little in anticipation of a slow afternoon that turned into a stampede. The trade press reporting around the rollout documents stores without the Smart Kitchen running peak-time speed of service at thirty to forty-five minutes. That was the operational state of the brand on a bad Friday night, at scale, before the AI.

Skipworth’s read on this — what he has said in every earnings call and what the Restaurant Business reporting confirms — was that the kitchen was the constraint on AUV growth. The food was fine. The brand was fine. The marketing was fine. The unit economics at $1.5 million in AUV were fine. The unit economics at $2 million were fine. The unit economics at $3 million — the long-term target — required a kitchen that could move twice as many orders through the same physical footprint without doubling the labor headcount and without crashing the customer experience on the highest-volume days. The Smart Kitchen is what Skipworth bet on to get there.

The move

The Smart Kitchen was codeveloped with an undisclosed startup partner. Skipworth has said publicly the cost of the technology is “not material to industry-leading cash-on-cash returns” — language designed to reassure franchisees that the install would not eat their unit economics. Three pieces sit underneath what the customer eventually experiences as “my wings came out fast and they were right.”

The forecasting engine. Every fifteen minutes, an AI demand-forecasting model predicts what each individual restaurant is going to need over the next ninety-six fifteen-minute windows. Inputs: weather, day of week, local sports schedules, school calendars, current historical patterns at the unit level, current menu mix, third-party delivery queue depth, loyalty-program activity, time of year, and roughly two hundred and ninety other variables Wingstop has not publicly enumerated. The output is a prep plan and a labor-batching plan — the system telling the line, in advance, how much chicken to par-cook, how many fries to drop, how many tenders to hold, and how many crew members the schedule needs to be holding open. The forecasting cadence is the part of the system most QSR competitors are not yet running. Industry norm is one demand forecast per shift, sometimes one per day. Wingstop runs ninety-six per store per day.

The kitchen display system. Four touchscreens replace the paper ticket. The flow is gamified — a crew member swipes a task complete the way a video game registers a level-clear, and the order automatically populates onto the next station’s screen. The wing station sees raw cooks; the sauce station sees them next; the fry station and pack-out station see the line for sides and assembly. Crew turnover, which used to be the choke point for ticket time when a new hire was working the line, drops as a problem because the system walks the new hire through each step. The KDS also shows the crew, in real time, how many ounces of fries to put in the basket. Predictive portioning replaced the eyeball.

The order-ready handoff. A board in the front of the house tells customers, third-party delivery drivers, and the front-of-house crew exactly what is ready and what is still in progress. The bag-grab — the moment when DoorDash drivers used to stand at the counter for ten minutes waiting for someone to call out their order — is now visible to everyone. BTIG analyst Peter Saleh has flagged that the next move on top of this layer would be a customer-facing “Wing Tracker” similar to Domino’s pizza tracker, which would convert an operational asset into a marketing asset.

Installation, in Wingstop’s own description, is “primarily wiring and screens.” About four weeks per store.

The standard QSR technology playbook for an operational change at this scope is the three-stage rollout. Year one is the pilot — usually fifteen to thirty stores, often in a single geography, watched closely by a steering committee. Year two is the regional expansion — a few hundred stores, refined operating procedures, vendor renegotiation. Year three is the systemwide deployment. By the time the last franchisee gets the system, the company has been at it for four to five years. The technology that started the rollout is often two generations old by the time it lands at the last store.

Skipworth chose not to run that playbook. The Smart Kitchen rolled out to all 2,586 domestic restaurants in ten months. The company-operated stores received the system first — the operational shake-down cohort — and the franchise system followed close behind. The compression of the timeline was deliberate. Skipworth’s read, on the record across multiple earnings calls and trade-press interviews, is that the cost of pilot drift is higher than the cost of compressed enterprise-wide deployment. Pilot drift is the operational equivalent of vendor lock-in: by the time the company has finished a five-year pilot, the operations team has gotten bored, the original vendor has been acquired, the rollout muscle has atrophied, the competitive timing window has closed, and the technology that started the pilot has aged out. The compressed timeline runs more deployment risk in a single calendar year — but it captures the operational and competitive value before either decays.

That decision is the part of the story other mid-cap leaders most need to take seriously. Almost every CEO I know with a board-approved AI initiative is currently running the three-stage playbook. The pilot is comfortable. The pilot is defensible. The pilot is the conservative path. The pilot is also, in many cases, what kills the deployment before the system ever lands in the hands of the line staff who would have produced the actual value. Skipworth’s bet was that the people on the floor with a working system for nine months is worth more than a steering committee with a perfect spec sheet for five years. The compression was the strategy.

The result

The operational results are documented in Wingstop’s earnings disclosures, the Restaurant Dive and Restaurant Business reporting on the rollout, the QSR Magazine analysis of the operational case, and the Canopy writeup that tracked the system through the install cycle. The headline numbers, all corroborated by multiple independent sources:

Ticket time. From over twenty minutes to under ten — a roughly fifty percent reduction at restaurants where the Smart Kitchen is live and running under operational discipline. New installs see a forty percent reduction within four weeks of going live. As of early 2026, Skipworth disclosed that roughly fifty percent of the system was hitting the ten-minute target as a daily-and-weekly average, with a focus now on hitting it inside the peak dayparts specifically. Speed-of-service compliance — the share of restaurants meeting the speed target — improved sixteen percentage points in Q1 2026 versus Q4 2025. Ten percent more orders are meeting the ticket-time target since the start of 2026.

Food waste. Down thirty-four percent. The predictive portioning and the demand forecasting together took out the par-cook excess that used to walk into the back of the house as overproduction. In a category where food cost is north of fifty-seven percent of restaurant-level expenses, a thirty-four percent reduction in waste lands directly on margin.

Labor efficiency. Up fifteen percent. The labor savings come from two places: the system tells the schedule when to hold a crew member open and when to send them home, and the gamified KDS shortens the training ramp for new hires from weeks to days. Labor-as-percent-of-restaurant-sales has been roughly flat in Q1 2026 versus the prior year (23.9 percent vs. 23.8 percent), in a year where labor cost inflation across QSR was running mid-single-digits. Flat labor cost in an inflationary labor market is, itself, a margin defense.

Throughput. Up nine percent during peak hours. On Super Bowl Sunday 2026 — the highest-volume day on the QSR calendar — Smart Kitchen stores held average quote times to sixteen minutes under conditions that would have pushed pre-AI stores to forty-five minutes or more.

Order accuracy. Up five percentage points in Q1 2026. Smart Kitchen stores also report an eight-point increase in guest satisfaction scores versus non-Smart-Kitchen stores. Saleh’s note: this is a marketing asset waiting to be turned on.

AUV. The cleanest financial proof point sits in the company-operated cohort. Wingstop runs 71 restaurants itself. Those 71 stores, which received the Smart Kitchen first and run it under the most uniform operational discipline, are operating at an AUV of approximately $2.5 million — a 25 percent premium to the system AUV of $2.0 million. Skipworth confirmed the $2.5 million figure on the Q4 2025 earnings call. The path to the $3 million store, which is the long-term unit-economic target Wingstop’s investor materials center on, is being walked first by the cohort that has had the system longest. That is the AI proof point.

Not all of Wingstop’s business outcomes right now are rosy, and any honest write-up of the AI deployment has to address that. Domestic same-store sales were down 5.8 percent in Q4 2025 and 8.7 percent in Q1 2026. CFO Alex Kaleida and Skipworth himself have attributed the decline to macro pressures on the QSR consumer — the same pressures showing up at Chipotle, Sweetgreen, Papa Johns, and Wendy’s in the same window. Two waves of winter storms in early Q1 took 700 restaurants and then 400 more out of operation for stretches of the quarter. GLP-1 weight-loss drugs are now in 23 percent of U.S. households, and the QSR category as a whole is repricing the relationship between calories and dollars. The traffic problem is real and it is industry-wide. Here is the part of the story I think is genuinely underwritten in the trade press: William Blair’s Sharon Zackfia had projected a seven percent Q4 comp decline; Wall Street consensus was minus 6.7 percent; Wingstop reported minus 5.8 percent. The actual print was meaningfully better than the projection. The 71 company-operated Smart Kitchen stores are running comps growth, not decline. The system stores hitting the ten-minute ticket-time target are showing the strongest customer-satisfaction movement. Without the Smart Kitchen — without the speed-of-service defense, the food-waste reduction, the throughput protection, and the AUV uplift in the company-operated cohort — the comp damage would have been larger, the margin damage would have been larger, and the analyst downgrades currently coming through the tape would be coming through deeper. The AI did not stop the consumer pullback. It is, by the evidence available, part of what is keeping the consumer pullback from being significantly worse.

The Craft of AI read

The dominant AI deployment story in mid-market right now is the one Skipworth’s earnings call performed for Wall Street on May 7: operational KPIs and financial KPIs presented in the same paragraph, with the implication that the AI is the cause of the good number and the macro is the cause of the bad one. The audience for that story is usually a board, an analyst community, or a press corp that is not going to do the disaggregation themselves. The Brief reader has to.

A Buffett-style read of Wingstop’s deployment goes like this. The operational case study is unusually rigorous: speed of service, food waste, labor efficiency, throughput, and accuracy are all measured directly, with third-party corroboration. The financial case study is muddled because two things changed in 2025 at the same time — the AI got deployed, and the consumer broke. The 71 company-owned stores are the closest thing Wingstop has to a clean test cohort, and their results suggest the AI is worth somewhere on the order of 80 basis points of restaurant-level operating margin at maturity. That is not nothing. It is also not “game changer,” in income statement terms, on a per-store unit-economic base where the consumer can move the comp number by 8 percentage points in a quarter.

The deeper lesson, the one I think operators need to write down before their next AI budget review: the AI almost always shows up in the operating metrics first, and shows up in the financial metrics second — and only if the macroeconomics cooperate. Wingstop’s macroeconomics did not cooperate. The AI worked anyway. The income statement does not show it cleanly. That is not a Wingstop problem. That is going to be the structural pattern for nearly every AI deployment of the next eighteen months, because the macro is moving faster than the technology can compound. The leader who can tell the difference between “the AI did not work” and “the AI worked but the consumer changed” will fund the next phase of their program correctly. The one who cannot will defund a working program and fund a vanity one — based on the wrong metric.

Things to consider

• Find the operational pain the AI is actually built to solve. Name it specifically. Wingstop’s pain was a kitchen architecture that worked at $1.5M AUV and started losing margin everywhere it scaled past $2M. The Smart Kitchen was not a strategy bet. It was a constraint-removal bet on the part of the operating system that was already known to be the choke point. Your equivalent: the workflow that is currently absorbing margin you cannot trace, carried by the practiced eye of one or two people who would not be able to explain it to a new hire if asked. Find that workflow. Name it. The AI deployment that solves a specific operational constraint produces measurable financial outcomes. The AI deployment that solves “productivity” produces a slide.

• Pick the forecasting cadence the system runs at — and run it faster than the industry norm. Wingstop runs ninety-six demand forecasts per store per day. The QSR industry norm is one per shift. The forecasting cadence is the load-bearing move, because the kitchen display, the labor schedule, the portion guidance, and the prep plan all derive from it. If the forecast updates every fifteen minutes, the operating system adapts every fifteen minutes. If it updates once per shift, the operating system runs blind for six hours at a time. Identify the cadence your equivalent system needs to run at. Run it faster than your competitors think is necessary. The cadence is the discipline.

• Skip the pilot if you can defend the deployment risk. Skipworth deployed to 2,586 restaurants in ten months. The industry norm for the same change is three to five years. The pilot is a comfort. The pilot is also where most AI initiatives lose their political momentum and their competitive timing window. If the system has been validated in a small cohort and the operational case is clear, the cost of compressed enterprise-wide deployment is almost always lower than the cost of pilot drift. Run the deployment risk on a calendar your competitors are not willing to run.

• Replace the paper ticket. Specifically. By name. Every business has a paper-ticket equivalent: the manual process that lives next to the official system and holds the operation together when the official system fails. At Wingstop it was a ticket on a clip above the wing station. At your business it might be a shadow spreadsheet, an email chain that runs the approval, a Slack channel that runs the escalation, or an analyst who maintains the master file in her head. The AI deployment that earns its keep is the one that replaces the paper-ticket equivalent with a system every person on the line can see, swipe, and trust. The AI deployment that adds another layer on top of the paper ticket produces a dashboard nobody reads.

• Prove the AUV uplift in the cohort closest to your direct control before you ask the rest of the system to believe you. Skipworth’s 71 company-operated stores are at $2.5M AUV against a $2.0M system average. That 25 percent premium is the proof point that funds the next phase of the rollout and the next round of franchisee development commitments. Find your equivalent: the business unit, the region, the customer segment, the team running under the cleanest operational discipline. Get the financial proof in that cohort. The cohort spread is the only number that pays for the next phase of the program.

THE WORKBENCH

Three questions to carry out of this brief

How much of your operating margin is currently absorbed by work nobody has properly mapped? The hours your best people lose to manual rework. The errors that walk out the front door because the official system never had a field for the edge case. The duplicate effort across two teams who never realized they were doing the same thing. The approval chain that runs through three inboxes because the procurement tool was too rigid to adopt. That margin is not theoretical. It is bleeding out of the business today, this quarter, in dollars your CFO can name but cannot trace. Most leaders cannot put a number on the cost. The AI deployment that earns its keep is the one that can.

How often does your operating system actually see new data? Wingstop’s Smart Kitchen forecasts demand ninety-six times a day at every store. The QSR industry norm is once per shift. Whatever your equivalent is — pricing updates, capacity reforecasts, customer-segment recalibration, inventory rebalancing, risk re-scoring — what is the gap between how often your system refreshes and how often the conditions on the ground actually change? The cadence gap is where the operational losses live, and it is the gap an AI deployment is uniquely well suited to close.

Which of your processes have never been traced end-to-end on a single sheet of paper? Wingstop did this work the year before they wrote a line of code. Every order, every cook timer, every portion check, every handoff, every bottleneck under peak load, every dependency on a manager’s read of the room. Most of the operational margin Skipworth recovered — the thirty-four percent food-waste cut, the labor-efficiency lift, the half off the ticket time — was not invented by the AI. It was sitting in the operation the whole time, paid for daily, waiting for someone to map the work properly. The question is how much of that money is sitting in your operation right now, and whether you have a system that can see it before a competitor builds one.

The Ground-Up Workshop

The conversations I keep coming back to right now are with leaders who have stopped asking “what AI tool should we buy?” and started asking the harder question underneath it: “what is the work we actually do, who is carrying the cost of the inefficiency around it, and how much margin is currently disappearing into processes nobody has measured?” Workflows first, technology second. That is the only sequence I have seen produce an AI deployment that survives the eighteen-month mark.

The Workshop is the alignment step: your leadership team in a room, mapping the broken processes where the work actually lives today, and drafting the target operating model that has AI built into the work — not bolted on top of it.

THE QUESTION

How much of your operating margin is being absorbed today by work nobody has properly mapped? Not the work that will break two years from now when you double in size. The work that is bleeding margin right now — this quarter, in hours your best people lose to manual rework, in errors that walk out the front door, in process steps that survive because nobody has traced them carefully.

If you can put a dollar number on that inefficiency, the AI program on your roadmap has a real target. The work itself is the bullseye. The people doing the work are who the deployment has to serve. The metric the work produces is the scoreboard. The AI is the system that finds the bloat, names it, and gives the people closest to the work the tools to remove it. That is a deployment that produces a number a CFO will recognize and a board will fund for a second round.

If you cannot put the number on it, the AI initiative on your roadmap is probably keyed to a use case picked from somebody else’s slide deck rather than to a workflow inside your own business. The deployment will produce a slide of its own. Some operational metric will move slightly. The board meeting six quarters from now will be a debate about whether the program is worth funding for another year — and the debate will be unwinnable, because nobody mapped the inefficiency at the beginning.

Michael Skipworth mapped the inefficiency. The order that nobody could find behind the countertop. The portion size nobody could measure. The peak-time wait that had been pushing past thirty minutes for two decades. He shipped a system to 2,586 restaurants in ten months that solved each of those, and the 71 stores he runs directly are now operating at roughly twenty-five percent above the system on the metric his board cares about — even in a year when the consumer is pressing hard against his category. The top line will recover when the consumer recovers. The work Skipworth did on the kitchen is what will make the recovery worth more, per store, when it arrives.

If you want to talk through where the operating margin is sitting inside your business right now — and what it would take to find it before someone else does — hit reply, or send a note to grant@thecraftofai.com. I run a small number of these conversations each quarter. We start with your business, not with my workshop.

— Grant grant@thecraftofai.com