Wingstop's Smart Kitchen vs. Olo's Catering Pilot With Chipotle: Two Competing Visions of Where AI Lives in the Stack

It’s a Friday afternoon in mid-April, and I’m doing what I usually do when two earnings prints are stacked back-to-back the following week: I’m building a 2x2. On the whiteboard in my home office, the x-axis is “builds proprietary kitchen AI” vs. “buys it from a vendor.” The y-axis is “owns the guest-ordering layer” vs. “rents it.” I’m trying to plot the top-25 U.S. QSRs and fast-casual chains by 2024 system sales — and what I’m finding, as I move from McDonald’s to Chick-fil-A to Domino’s to Wingstop to Chipotle to Sweetgreen, is that the diagonal is doing a lot of work.

That diagonal is the thesis of this piece. Two earnings prints are coming next week and the week after — Chipotle on April 23, Wingstop on April 30 — and both are expected to formalize what their operators have been hinting at for months. Chipotle, per reporting on its emerging partnership with Olo, is anticipated to disclose a Catering Plus pilot, with Olo positioning itself as the catering-fulfillment layer for a brand that sells more catering than people realize. Wingstop, per CFO commentary and trade-press previews, is expected to confirm that its in-house Smart Kitchen platform is already live in more than 200 restaurants, with a target of about 400 by the end of April — figures that have been previewed in Nation’s Restaurant News and Restaurant Business but not yet ratified by a Q1 print.

Two operators. Two prints. One frame. And the frame, as I’m going to argue for the next 2,000 words, is not really about technology at all.

Two prints next week, one frame

Let me set the table on what I think we’re about to see.

On Wednesday, April 23, Chipotle reports Q1. The number that matters for this column is not transactions or comps — it’s whatever the company says about its catering channel and the technology backbone behind it. Reporting in the trade press has Chipotle and Olo running a Catering Plus pilot, with Olo taking on the workflow that has historically been Chipotle’s weakest digital surface: large-order catering. If Chipotle confirms the pilot on the print, it’s the first time the chain has publicly outsourced a meaningful chunk of its digital ordering stack to a vendor. That’s a tell.

On Wednesday, April 30, Wingstop reports Q1. The number that matters here is not same-store sales — those are going to be loud and good and everyone is going to talk about them. The number that matters is the Smart Kitchen store count. CFO Alex Kaleida has previewed in investor settings that the platform is operating in 200-plus units already, and the company has telegraphed a roughly 400-unit footprint by the end of April. If those numbers hold up on the print, Wingstop will be the first publicly traded QSR to confirm a proprietary, in-house kitchen-operations AI live in more than 10% of its U.S. system in a single quarter.

Two confirmations, separated by seven days. Both stories live in the same conceptual frame — operator-led AI — but they answer the make-vs.-buy question in opposite directions.

My base case: both prints land, both confirmations are issued in roughly the language I’ve described, and by Friday May 2 the sell-side will have two new operator templates to argue about. The contrarian piece sits in the argument itself. Most of the sell-side will treat Wingstop-builds and Chipotle-buys as evidence that the two chains have different views of “where AI is going.” I don’t think that’s what’s happening. I think they have different labor cost structures and different channel mixes, and the build-vs.-buy split falls out of the math.

Why Wingstop builds and Chipotle buys

Here is the contrarian thesis stated as plainly as I can manage: enterprise restaurants do not make build-vs.-buy decisions on the basis of technology. They make them on the basis of internal labor cost relative to vendor opex. Once you have enough scale, every line item on the engineering org chart becomes a candidate to insource. Once you don’t have enough scale on a particular channel, every line item becomes a candidate to outsource.

Wingstop has roughly 2,500 system units globally as of year-end 2024 and is pursuing a $3 million average unit volume target. AUV times unit count is the number that funds an engineering org. At $3M AUV across U.S. units, the system is throwing off enough royalty and franchise fee revenue at the corporate level — Wingstop’s corporate revenue model is fee-based, not store-P&L — to support a software-platform engineering team that can compete with what a Toast or an Olo would sell them. The cost of one senior engineer in Dallas, fully loaded, is somewhere around $280K to $340K. A 40-person platform org is roughly $13M to $14M a year. Against a system that’s about to clear $5 billion in U.S. sales, that’s a fee model the franchisor can pay out of corporate without flinching.

Compare that to what Wingstop would pay a vendor for an equivalent capability. SaaS pricing for kitchen-display systems plus operations AI for a 2,500-unit system is — and I have seen these contracts in due diligence — somewhere between $200 and $600 per store per month, before professional services. Call it $400 average. That’s $12M a year before implementation, customization, integration work, and the consultant army that always shows up. The first-year vendor bill is well north of the in-house engineering org’s annual cost. The second year is close to even. By year three, the vendor cost has compounded and the internal cost has not, because the marginal cost of a deployed engineering team is approximately the loaded salary plus benefits, and the marginal cost of a vendor is whatever the contract says it is plus whatever the vendor’s pricing power can extract on renewal.

That math is why Wingstop builds.

Now run the same math at Chipotle, but only for the catering channel. Chipotle has roughly 3,500 U.S. units and does — by my back-of-envelope from prior disclosures — somewhere between 6% and 9% of revenue through catering. Even on the high end, that’s a $750M channel against a $10B+ business. The engineering effort to build a competitive catering-orders surface, with the address validation, large-order pricing logic, group ordering, billing, delivery integration, and corporate-account workflow that catering customers expect, is not materially smaller than the effort to build the consumer-facing app. Chipotle’s digital team is already heads-down on the core ordering experience, the rewards program, and store-level digital throughput. Building Catering Plus in-house would mean hiring a separate squad and supporting it in perpetuity to serve 7% of the business.

That’s the same calculus as Wingstop’s, run the other way. The channel does not justify the engineering org.

Hence: Wingstop builds the kitchen, Chipotle buys the catering. Both decisions are correct. Both are about labor cost relative to vendor opex, with channel mix as the moderating variable.

The Olo PT move, decoded

This frame also explains a piece of sell-side activity I’m watching closely. Eric Martinuzzi at Lake Street is anticipated to raise his Olo price target to $10 from $9 in the post-print window — reporting in the trade press has already telegraphed a positive view on the Chipotle pilot. The $1 PT move is small, and the model adjustments behind it will be modest, but the directional signal is worth decoding.

Olo’s pitch has been, for several years, that as enterprise restaurants get more sophisticated about digital, they will consolidate their fragmented vendor stack onto Olo. The bear case has been the opposite: that as enterprise restaurants get more sophisticated, they will pull capabilities in-house. The Chipotle pilot is the first major datapoint that suggests there is a third, more interesting outcome: enterprise restaurants will insource the capabilities where they have scale economics, and they will outsource the capabilities where the channel can’t justify the build. For Olo, that means catering, large-order workflows, off-premise, third-party-marketplace integration, and probably loyalty integration plumbing are durable use-cases. Core consumer ordering for the largest chains is not.

If you are modeling Olo, the implication is that the right way to think about TAM is not ”% of enterprise digital orders” — it’s ”% of enterprise digital orders in channels that don’t justify in-house engineering at the customer’s scale.” That’s a smaller, more defensible number. It is also a number that does not get eaten by Wingstop building Smart Kitchen, because Smart Kitchen is operations AI, not order capture. The Olo bull case is narrower than the bulls have been telling it. It is also more durable than the bears have been telling it.

My base case on Olo: the stock works on a multiple expansion as the catering thesis gets validated, and the long-run model is “the off-premise plumbing layer for chains too big to use a SaaS-y POS and too channel-constrained to insource the plumbing.” Lake Street’s $10 PT is, I suspect, an opening bid in a re-rating that has more room to run if Chipotle’s pilot scales.

What this implies for Toast and Square

The make-vs.-buy frame has consequences past Wingstop, Chipotle, and Olo. The two companies in restaurant tech with the most to lose if my framing is right are Toast and Square (Block) — and I want to be careful about which one, because they are pointed at different customer cohorts.

Toast lives between the SMB tier and the lower end of the enterprise tier. Roughly: Toast is dominant in the 1-to-50-location operator, increasingly competitive in the 50-to-500-location operator, and aspirational in the 500-plus operator. The 500-plus operator is exactly the cohort where my make-vs.-buy line crosses. At 500 corporate-owned units and reasonable AUVs, the math on insourcing operations AI starts to make sense. At 1,000 units and Wingstop-style AUVs, it almost certainly does. Toast’s enterprise pipeline is going to face a question its mid-market customers don’t ask: why are we paying you per store per month when we can hire 30 engineers and own the roadmap?

Toast’s defense against this is its breadth — payroll, payments, capital, marketing, ordering, kitchen — and the argument that even at scale, you don’t want to run six engineering orgs to replace it. That’s a real argument, and it works for the operators who genuinely don’t want to be in the software business. It does not work for operators who already are in the software business and just haven’t put it on the balance sheet yet. Wingstop is in the software business. So is Chick-fil-A. So is Domino’s. So is, increasingly, Chipotle.

Square’s restaurant business sits in a different place — more SMB, more single-unit and small-multi. The make-vs.-buy line doesn’t reach down into that cohort. Square’s exposure to my thesis is indirect: as Toast loses enterprise share to insourcing, Toast pushes harder downmarket, which compresses Square’s restaurant ARPU growth. That’s a second-order effect, not a primary one.

The first-order effect of my framing is on Toast’s enterprise ambition. If the Wingstop print on April 30 confirms what’s been previewed — 200-plus stores, 400-by-end-of-month, in-house build, no third-party kitchen platform — every Toast enterprise pitch in May is going to be a referendum on whether the prospect believes their AUV and unit count put them on the Wingstop side of the line or the Chipotle side.

The make-vs.-buy line item every CFO should add

Here is the practical takeaway. If you are a restaurant CFO reading The Bottom Line — and I know some of you are, and I appreciate the tips, please keep them coming — there is a line item I would add to your annual technology review starting this year.

Call it “imputed insource cost.” For every vendor relationship over $1M in annual spend, compute the cost of replacing that vendor with an internal engineering team of the size required to deliver the same capability. Use a 30-engineer team for kitchen operations AI, a 50-engineer team for an order capture and aggregation layer, a 20-engineer team for loyalty plumbing, and a 15-engineer team for analytics. Fully load each engineer at $300K, including infrastructure overhead. Add a one-time build cost in year one equal to 1.5x the annual run rate.

Compare imputed insource cost against vendor cost over a five-year horizon. If imputed insource is more than 1.4x vendor cost over five years, you are correctly outsourcing. If it is between 0.9x and 1.4x, the decision is a coin flip and probably hinges on strategic considerations like talent retention and roadmap control. If it is under 0.9x, you are leaving money on the table and your board should be asking why.

The reason this line item matters now, in 2025, and didn’t matter in 2018 or 2020, is that the cost of building software has fallen sharply. AI-assisted engineering, cheaper foundation models for the AI-flavored capabilities, and a more mature open-source ecosystem for restaurant-adjacent workloads (payments, KDS, inventory) all mean that a 30-engineer team in 2025 produces what a 50-engineer team produced in 2020. The vendor side of the math has not moved as much — vendor pricing power is sticky and rises with revenue per location.

This is the math that’s driving Wingstop. It will increasingly drive Chick-fil-A, Domino’s, Yum Brands, and Restaurant Brands. It will not drive Chipotle on the catering channel, and it will not drive any chain on any channel where the channel can’t justify the build. The result is the diagonal in my 2x2 — the line that runs from “buy everything” in the bottom-left to “build everything that matters” in the top-right.

As a later piece on restaurant-tech M&A frames it, the back half of this decade is going to be defined by chains pulling capabilities back in-house wherever scale economics justify it, and vendors retreating into the channels where they can defend a durable structural advantage. The Wingstop and Chipotle prints next week are the first clean test of that thesis on the same week, in the same column of the trade press, with the same analyst spreadsheets being updated by the same buy-side junior associates. And in a later case study on operator-led automation, the same logic applies to Sweetgreen’s Infinite Kitchen — built, not bought, because the throughput economics at unit level justified the engineering org.

What I will be watching most carefully on the April 30 print is not the Smart Kitchen store count, which I am reasonably confident lands roughly where Wingstop has previewed. I will be watching the CapEx line and the corporate G&A line. If those lines do not move materially, it tells you that Wingstop is funding its in-house build out of cash flow without flinching — which is the strongest possible signal that the math I have outlined above is generalizable to any QSR with $3M-plus AUVs and 2,000-plus units.

If those lines do move materially, the in-house build is more expensive than the model assumes, and the line item every CFO should add — imputed insource cost — needs a wider error band. Either way, we will know more on May 1 than we know today.

Two prints, one frame, one diagonal. The whiteboard will be there Monday morning.

— Marcus writes The Bottom Line. Tips: [email protected].