Burrito Season is Ava Cado's stress test

It is a Tuesday afternoon in a strip-mall Chipotle in the kind of suburb where the Target anchors one end of the parking lot and a Hand & Stone anchors the other. I am here because Burrito Season — the company’s internal name for the March-through-May hiring window — opened this week, and I wanted to see what the tablet looked like in a real restaurant before I wrote about what the tablet was doing to the funnel. The general manager, who has been with Chipotle for nine years and has hired through every iteration of the company’s applicant-tracking system, walks me to a kiosk on the back wall near the office. She taps a QR code printed on a tent card by the salsa station. The kiosk wakes up. Hi! I’m Ava Cado. Want to apply? The kid in front of me, who has just put down a chicken bowl, types his phone number into the screen and starts a text-message conversation with the bot while he eats. By the time he is done with the bowl, he has a Thursday-morning interview slot on his phone and the GM has been pinged in the same thread. He has not spoken to a human. The GM has not done any administrative work. The kid is, presumably, a future crew member.

I am writing the same scene the trade press has written a half-dozen times since Chipotle announced Ava Cado in October 2024, and I am writing it again because the deployment has reached the moment it was built for. Burrito Season is when Chipotle hires roughly 20,000 people in twelve weeks to staff up for the peak guest period that begins in early spring and runs into Memorial Day. The hiring bot has been live across the fleet for almost eighteen months. The Q4 2025 print landed February 4. The first week of Burrito Season is this week. If Ava Cado was ever going to be stress-tested, it is being stress-tested now.

The contrarian thesis I want to argue is this: Ava Cado is the cleanest in-market AI deployment in fast-casual, and it is solving the wrong half of the labor problem. The throughput numbers are strong — application completion lifted to roughly 85% from roughly 50%, applicant flow doubled, time-to-fill compressed by as much as 75% per CEO Scott Boatwright’s comments to CNBC. Those numbers move a labor-margin line in a way almost nothing else in the public restaurant tape moves. But the labor-margin problem at a 3,700-unit footprint with a 7,000-unit target is not, structurally, a fill problem. It is a retention problem. Ava Cado is a beautifully built top-of-funnel machine. The bottom of the funnel still leaks. That is what Burrito Season is going to show us.

What I watched, in the GM’s office

The first thing worth saying is that Ava Cado is not the kiosk. Ava Cado is the conversational hiring agent, built on the Paradox platform, that runs as a layer on top of Workday Recruiting. The kiosk is one front door. The QR code on the tent card by the salsa station is another. The careers page on the Chipotle website is a third. A text from a friend who already works at the store is a fourth. All four front doors deliver the candidate into the same SMS-based conversation, which is the actual product.

The GM walked me through the back-office view. From her seat, Ava Cado is a dashboard showing every applicant from her store in the last seven days, their stage, their availability windows, and a column flagging candidates within a half-mile walking radius. She clicked on the kid from the dining room. His application was 92% complete by the time he finished the bowl. The remaining 8% was the I-9 paperwork that has to be physically witnessed; everything else — schedule preferences, transportation, work-authorization questionnaire — had been collected through the SMS thread while he was eating. Two clicks, the GM said, and I have a Thursday at ten-thirty with him. Two clicks is the entire pitch.

The pitch is not “AI hires for you.” The pitch is that the layer between the applicant and the manager has been compressed to two clicks, and the manager’s time gets reallocated from data entry to the conversation that actually matters — the in-person interview, the trail shift, the offer. That reallocation is the labor lever the PwC case study writes up in detail as the operator-side ROI of the deployment.

The stack: Workday Recruiting plus eighteen integrations

The piece of this that the trade press almost never writes about is the part of the build that took the longest and cost the most. Ava Cado is not a standalone product. It is a Paradox conversational layer that sits on top of Chipotle’s Workday Recruiting system, and the value of the deployment is created in the integration layer between the two. PwC was Chipotle’s systems-integration partner for the build, and the PwC case study puts the number at eighteen system integrations between Paradox and Workday — the seams that make the funnel feel like one product to the candidate and one dashboard to the manager.

What does it take eighteen integrations to do? At minimum: a candidate-data sync between the SMS thread and the Workday applicant record; a scheduling sync between Paradox’s availability picker and the manager’s calendar; a job-requisition sync that pulls open positions from Workday into the conversational flow; a background-check trigger that fires once an offer is accepted; an I-9 verification handoff to E-Verify; a tax-credit eligibility check during onboarding; a learning-management-system enrollment that pushes training modules into Workday Learning; and a payroll-system trigger that creates the employee record so the first paycheck clears on cycle. The remaining ten are the unglamorous plumbing: location code mapping across 3,700 stores, multilingual handling across English, Spanish, French, and German, manager-permission roles, state-specific audit logs, district rollups, EEO reporting, source tracking, mobile deep-linking, GM push notifications, and the analytics warehouse feed.

Eighteen integrations is the number that explains why Chipotle is the case study and most of the rest of the QSR field is not. The Paradox platform has been productized since well before Chipotle’s deployment, and there are several credible competitors. The hard part is the connective tissue between the bot and the system of record. Without the integrations, Ava Cado would be a chat widget that creates work for the manager. With the integrations, it is an administrative-layer replacement that creates time for the manager. Two entirely different products, two orders of magnitude apart to build.

What the integration number also tells you is the replacement cost of Ava Cado for any operator who wants the same outcome. The PwC-built integration layer is, conservatively, a multi-million-dollar engineering program with an eighteen-month timeline. At twenty-thousand Burrito-Season hires alone, the savings clear it easily. At the long-term run rate of nine to ten thousand hires per year against a 7,000-unit target — a number Chipotle disclosed in the original Ava Cado announcement — the math is stronger. For a 300-unit chain, the same integration program would not clear the bar. This is one of those AI deployments where scale is the moat.

The PwC numbers, taken at face value

I am going to walk through the published outcomes the way I always walk through vendor case-study numbers — taking them at face value first, and then saying what I think the print actually means.

Application completion rate: 50% to 85%. The headline number. Before the deployment, half of the applicants who started a Chipotle application abandoned it. After the deployment, roughly fifteen percent abandon. The mechanism is the SMS-based interaction: the candidate is not asked to complete a multi-page form on a mobile browser. They are asked one question at a time, in the messaging app they already use, with the option to pause and come back. Eighty-five-percent completion is not a number any traditional applicant-tracking system delivers, and it is the cleanest piece of evidence that the conversational-AI interface, specifically, is doing the work.

Applicant flow: doubled. Same funnel, twice the volume. Some of this is the completion lift — completed applicants are counted in flow numbers in a way that abandoned ones often are not. Some of it is reduced friction creating real new applicants who would not have started a traditional application. Boatwright has been careful on calls not to claim that Ava Cado is marketing — that it is bringing new candidates into the brand who would not otherwise have considered Chipotle as an employer. The 2x flow number is, on the company’s own framing, a funnel-conversion number, not a brand-awareness number. Ava Cado converts demand. It does not create it.

Time-to-fill: down by as much as 75%. Boatwright told CNBC, per Fox Business, that the time it takes to fill an in-restaurant position has dropped by as much as 75% — application-to-start-date from twelve days to four. The 75% number is the kind that does not survive contact with cross-validation often. This one has: Boatwright restated it on the Q4 2025 earnings call this past February, Paradox restated it in the partner case study, and the Nation’s Restaurant News coverage of the Burrito Season ramp leans on it as the single most cited operational figure in the company’s playbook. The mechanism is the integration layer: every handoff that used to take a half-day of email now fires automatically when the previous stage completes.

The operator’s read on the three numbers, taken together: this is a funnel-velocity contract, not a labor-headcount contract. Ava Cado does not reduce the number of crew members the store needs. It compresses the cost and the time of getting those crew members in the door. The labor savings appear in the manager’s P&L line — the assistant-manager hours and general-manager hours that used to go to applicant-tracking work and now go to floor coverage — not in the crew-hours line itself. For an operator reading this and thinking about whether to write the same kind of integration cheque, that distinction is the one that matters. The contract is a manager-time contract, dressed up as a hiring-funnel contract. Both are real. They are not the same.

The “almost every 24 hours” cadence and what it requires

Boatwright has been making a specific claim on recent calls about the pace at which Chipotle is opening new restaurants. The phrase is “almost every 24 hours” — i.e., the chain has been opening, on average, very nearly one new company-owned location per calendar day. The company opened 334 company-owned restaurants in 2025 against a 365-day calendar, which is the math the phrase points to. The 7,000-unit North American target is the destination. The unit-opening cadence is the path.

The piece of this most operators miss when they read the unit-opening number is what it requires of the hiring funnel. A new Chipotle opens with somewhere between 20 and 25 crew members hired and trained in the four-to-six weeks before the doors open. At an opening-a-day cadence, that means the chain is hiring, on average, roughly twenty new crew members every single day, just for new units — before you count replacement hires across the existing 3,700-store footprint. Burrito Season pushes that number up by a multiple. The funnel has to support both the seasonal ramp and the steady-state new-unit pipeline simultaneously, for twelve weeks, without breaking.

A 12-day time-to-fill at that volume would, structurally, not work. The math of a 12-day fill window against a 24-hour opening cadence implies that every new store’s hiring pipeline has to start almost two weeks before the previous store’s closes — and the manager doing the hiring is usually the same manager covering the floor of an existing store. A 4-day fill window lets the same manager run the funnel for the next opening as a part-time second job, with the administrative load handled by Ava Cado. The cadence works because the funnel was rebuilt. This is the part of the RetailWire coverage of the unit-opening pace most of the trade press has not yet drawn the line through clearly. The AI hiring tool is not just a labor-cost lever. It is a unit-growth-pace lever. The 7,000-unit target is not reachable on a 12-day fill window.

The contrarian read: traffic is the thing Ava Cado does not fix

Here is the part of the case study that the trade-press write-ups rarely include. The same Q4 2025 print that contained Boatwright’s reiteration of the Ava Cado numbers also contained a 2.9% transaction decline against the prior-year quarter, partially offset by a 1.2% check increase, for a 1.7% comparable-sales decline. The 5.4% full-year revenue lift came from new units, not from same-store transactions. Same-store mix was negative. Traffic was pressured.

That dynamic is not, in any direct sense, Ava Cado’s problem to solve. Ava Cado is a labor-margin and unit-growth-pace tool. The transaction-decline story is a guest-side story — value perception, menu mix, the broader QSR competitive environment, and the macro pressure on the lower-income consumer that the entire industry has been navigating since the back half of 2025. Chipotle is hardly alone in that pressure; the print was modestly better than the QSR sector aggregate. But the dynamic does mean that the labor savings Ava Cado generates do not flow to the bottom line the way they would in a flat or growing same-store environment. They land into a comp-sales hole and get partially absorbed.

For an operator reading this and thinking about the structural lesson, the point is narrow and important: an AI hiring tool, no matter how cleanly built, does not move topline. Ava Cado is solving a cost problem and a growth-pace problem. It is not solving a traffic problem. If you are evaluating an equivalent build for your own chain — Paradox or one of its competitors, integrated into your own HRIS — make sure you are buying it on the cost-side and growth-pace contract, not on a vague hope that better hiring will somehow lift transactions. It will not. The two systems are decoupled. Better hiring helps the store deliver on demand. It does not create demand.

The McDonald’s case I wrote about in the drive-thru study is the cleanest parallel. The voice-ordering deployment was framed as a labor and capacity tool; it did not move same-store traffic, and the leadership team was careful never to claim it would. The vendors that surrounded the deployment were sometimes less careful. The same trap exists in hiring-AI. Vendors who imply that better hiring will produce same-store-sales lift are selling sentiment. Vendors who promise unit-growth-pace acceleration and manager-time recapture are selling product. Read the contract carefully.

The retention question, which is the one I am actually watching

The deeper question Burrito Season will surface is retention. A doubled application flow, an 85% completion rate, and a 4-day fill window are all top-of-funnel metrics. None of them tells you what happens at day 30, day 60, or day 90 of the new hire’s tenure. Restaurant industry crew turnover has historically run somewhere in the 130-to-200-percent range annually. A faster, cleaner top-of-funnel does not, by itself, change that number.

What it can do is feed more applicants into a retention funnel that is, downstream, leaky. If 20,000 Burrito-Season hires turn over at 150% in their first year, you are running the same hiring conveyor for the next twelve months whether you want to or not. Ava Cado makes the conveyor cheaper and faster to run. It does not, on the published feature set, reduce the rate at which the conveyor has to run. The retention work — onboarding-quality scores, manager-relationship survey signals, first-30-day-experience data, predictive turnover modeling — is a different software product, and Chipotle has not, to my reading of the public record, deployed an equivalent fleet-scale tool against retention specifically. Training is not retention. They are correlated and they are not the same.

My bet — and I will call it a bet because the data to support it is not yet public — is that the next major AI hiring-platform announcement from a large QSR operator in 2026 or 2027 will be a retention-focused deployment, not a hiring-funnel one. Either Chipotle will extend Ava Cado’s scope into the first-90-days experience (the Paradox platform supports this, the integration plumbing is in place), or a competitor will leapfrog with a retention-first product. The Q3 2026 calls will be the print.

What I am watching during Burrito Season

A short list of metrics that, if Chipotle continues disclosing them, will tell us how the stress test went.

Time-to-fill, restated on the Q2 2026 call. The 75% number should hold if the funnel works. If it slides — to 60%, or the call simply does not mention it — that is the tell that the funnel hit a capacity limit at the Burrito Season ramp. The integration layer has a ceiling somewhere.

Application completion rate, restated on the same call. This should be the most stable number across the year because it is mechanical, driven by interface design rather than load. If it moves down meaningfully, the chatbot is being asked to handle volume beyond its tuning envelope. That would be the canary for throughput-limited Ava Cado.

Crew turnover, in any form Chipotle is willing to disclose. The metric the company is most reluctant to publish and the one I would most want to read. Even directional commentary — “improved year-on-year,” “held flat,” “deteriorated” — would be enough to read against the funnel numbers and triangulate whether the top-of-funnel work is feeding a retention machine that holds the gains or a leak that gives them back.

Margin-mix in the same-store print. Comp-sales-versus-mix is the leading indicator for whether the labor savings from Ava Cado are flowing to operating margin or being absorbed by a softer guest mix. If transactions stabilize and margins improve, the bet is paying out.

The bet

Read the deployment, the integrations, the published numbers, the unit-opening cadence, and the Q4 2025 traffic context together, and the operator’s-eye summary is the following.

Ava Cado is the cleanest case in the public restaurant tape for an AI deployment that crossed the chasm from pilot to fleet, with disclosed numbers, with a real integration program behind it, and with a CEO who continues to reference it on every earnings call. The eighteen-integration build is the moat. Paradox is replaceable; the Workday-integrated, PwC-engineered, four-language, fleet-deployed layer is not. That layer is what makes the 24-hour unit-opening cadence possible, and the unit-opening cadence is what makes the 7,000-store target reachable. If you are an operator of any size and you are looking at the public record for one AI deployment that has demonstrated unambiguous ROI at chain scale, it is this one.

But it is a labor-margin and growth-pace tool, not a topline tool. It does not fix the same-store transaction decline that showed up in Q4 2025. It does not, at least not yet, address the retention half of the labor problem that determines whether the funnel needs to run twice in a year or only once. It compresses the cost of running the hiring conveyor. It does not yet slow the conveyor.

My bet — and the read this column will revisit at the Q3 2026 print — is that Ava Cado’s second act, sometime in the next eighteen months, will be a retention extension. The integration plumbing is in place. The data is in Workday. The Paradox platform supports it. The Burrito-Season cohort that lands this week will be the first class large enough to make the retention math visible by the September staffing report. Either Chipotle will pull on the retention lever next, or a competitor will. The forward-looking Four Margins of AI Spend frame I have been building toward will need a fifth category — retention AI — by the time we get to the end of the year, and Ava Cado’s evolution is the deployment that will probably define what the category looks like.

For now, Burrito Season is the stress test. The throughput numbers should hold. The retention numbers are the ones I would read in September.

— Priya covers the Operator beat for TableTransfers. Tips: [email protected].