Inside Wendy's FreshAI Year Two — The First Brand to Stick With Drive-Thru Voice

It is the first Tuesday of January and my reading list has two items on it. The first is TheStreet’s January piece on McDonald’s 2026 AI plans, framed around CIO Brian Rice’s promise that “technology solutions will alleviate the stress” on the crew juggling four order channels at once (TheStreet, January 2026, via Yahoo Finance; Newsweek, January 8, 2026). The second is Wendy’s most recent FreshAI update — the company blog post from May 2025 that quietly confirms FreshAI is now processing “tens of thousands of orders daily” in English and Spanish (Wendy’s corporate blog, May 2, 2025).

Read them in order and the McDonald’s news reads differently. The 2026 plan — Accuracy Scales weighing every bag, voice agents in the drive-thru, geofenced “Ready on Arrival,” multi-lane builds nationwide by 2027 — is not a McDonald’s roadmap. It is, almost line for line, the deployment Wendy’s has been running for thirty months. The only large-chain drive-thru voice AI program in the US that has not been paused, rolled back, or quietly de-emphasized is FreshAI. That is the program McDonald’s is now copying. And it is the case study any operator weighing a voice deployment in 2026 needs to read first.

Methodology

This is a public-record case study. I did not interview anyone at Wendy’s, McDonald’s, Google Cloud, or any franchisee. Every fact, deployment count, accuracy claim, and quote in this piece is sourced to a dated, public article — Wendy’s corporate communications, McDonald’s CIO statements, the company earnings calls, and the major trade press (Restaurant Dive, Restaurant Business Online, QSR Magazine, Fortune, Restaurant Technology News). Where the figures I have don’t match the framing the brief assumed, I say so in the body. The McDonald’s-as-Wendy’s-template framing is interpretation, not reported fact — McDonald’s has not publicly cited FreshAI as a model, and I cannot prove they are deliberately copying Wendy’s choices. What I can show is that the architecture, the partnership stack, and the rollout sequence McDonald’s announced for 2026 are the same architecture, stack, and sequence Wendy’s has been running since 2023. Read the interpretation as informed inference. The facts behind it are not.

Two figures I cleaned up while writing this. The brief I was working from cited “roughly 14,000” US McDonald’s locations and “~45,356” globally. The closer numbers from corporate reporting are 13,706 US units at the end of 2025 and roughly 43,000 globally (Restaurant Business Online, year-end 2025 store count). I use those.

What FreshAI actually ships

The Wendy’s deployment is easy to under-read. The headline number — 500-plus restaurants targeted by end of 2025 (Restaurant Dive, May 2025) — sounds small next to McDonald’s announced US footprint. What that headline hides is that Wendy’s has been compounding the program for thirty months without a public rollback, a viral failure cycle, or a franchisee revolt — three things every other large-chain deployment has lived through.

The timeline, end to end, is on the public record.

June 2023: FreshAI goes live at one company-operated restaurant in Columbus, Ohio. The technology is built on Google Cloud’s large language model with Wendy’s order data, menu data, and franchisee-specific customisation rules layered on top (Fortune, October 15, 2024).

Through 2024: The pilot expands to 36 company-operated locations across Ohio and Florida. Wendy’s CIO Matt Spessard tells Fortune the team spent “thousands of hours of crew and customer feedback gathering” inside the Columbus location before opening the second store. The team rebuilds the system to handle Wendy’s customisable menu, limited-time offers, and language nuance — including teaching the model that a customer asking for a “milkshake” almost certainly wants a Frosty (Fortune, October 15, 2024).

February 2025: Wendy’s Q4 2024 earnings call. CEO Kirk Tanner confirms FreshAI is live at roughly 100 locations and tells analysts the company plans to expand to 500–600 by end of 2025. Margin improvement of 80 basis points at company-operated restaurants is attributed in part to FreshAI-driven labor efficiency. Tanner’s quote on the call: “We have a lot of confidence. It’s definitely cutting edge.” And, on the system: “One thing I love about it, it continues to get better” (Restaurant Business Online, February 14, 2025).

February 2025, separately: Wendy’s announces tech capex stepping from $53M in 2023 to $100–110M in 2025, with FreshAI and digital menu boards as the headline workloads (Restaurant Technology News, February 19, 2025).

May 2025: Wendy’s corporate blog confirms FreshAI is over 160 restaurants, processing tens of thousands of orders daily, with Spanish-language capability newly in production (Wendy’s, May 2, 2025; Restaurant Dive, May 2, 2025). The blog post is unusual for the genre — Wendy’s claims some customers report the AI is “even better than interacting with a real person” because of “clear and consistent” delivery. I would normally read that as marketing. In the context of every other public deployment in the category, it is at least interesting.

That is what the FreshAI program actually ships: a Google Cloud LLM-based voice agent, English and Spanish, deployed and operated at a triple-digit and now climbing footprint, with stated check-size and margin improvements at company-operated stores, and zero reported franchisee revolt. The system reduces drive-thru service times in early testing by 22 seconds against the regional average (Fortune, October 15, 2024) — the single most-cited operating number in the program.

What it doesn’t ship — and this is worth naming clearly — is a public per-store capex disclosure. Neither Wendy’s nor Google Cloud has put a number on the line. I cannot tell you what FreshAI costs to install at a unit. The publicly disclosed company-wide tech capex of $100–110M in 2025 (Restaurant Technology News, February 19, 2025) covers digital menu boards, app investments, loyalty work, smart scales, and FreshAI together. Dividing it by the 500-unit target produces a wrong number; the line items aren’t separable from the outside. Any vendor or trade outlet quoting a per-restaurant FreshAI capex figure is back-solving from the same press release I am, and the figure isn’t there.

The McDonald’s restart looks like the Wendy’s deployment

McDonald’s spent five years and a high-profile public failure getting to where it now is — and where it now is, looking at the technology stack, is roughly where Wendy’s was at the end of 2024.

The McDonald’s voice-AI story is well documented and I have covered it before in the McDonald’s drive-thru case study. The short version: Apprente acquired 2019; McD Tech Labs sold to IBM in 2021; 100-restaurant pilot through 2023; viral TikTok failures and franchisee resistance through early 2024; partnership wound down by July 2024; Google Cloud partnership announced six months earlier (December 2023) and quietly built out through 2024 and 2025 on the equipment-monitoring and edge-compute side first. The voice agent never publicly restarted under McDonald’s brand. Until now.

The 2026 plan, announced through CIO Brian Rice’s media tour in early January, has four pieces.

Voice AI in the drive-thru. The chatbot ordering layer, with Google Cloud’s stack as the foundation. Rice did not name the vendor publicly. The architecture — Google Cloud LLM, edge compute via Google Distributed Cloud, fine-tuned on McDonald’s order history — is the same architecture Wendy’s bought from Google in 2023 (Newsweek, January 8, 2026; Yahoo/Tech, January 13, 2026).

Accuracy Scales. AI-powered scales weighing every bag at the handoff window. The system compares actual bag weight to expected weight for the ordered items and prompts the crew to correct missing or extra items before the customer drives away. Already deployed across “thousands” of drive-thrus, kiosks, and delivery channels in a dozen markets per Rice’s statement (Mashed, December 24, 2025; Yahoo/Tech, January 13, 2026).

“Ready on Arrival” geofencing. The mobile app pings the kitchen when the customer crosses the property line, so order assembly starts before the car reaches the speaker. Expanding to six markets including US, Japan, and the UK in 2026 (Mashed, December 24, 2025).

Multi-lane drive-thru rebuilds. Physical infrastructure — adding lanes — with completion targeted nationwide in 2027 (Newsweek, January 8, 2026).

Three of those four pieces — voice, scales, geofencing — have direct analogues in the Wendy’s program. Wendy’s hasn’t publicly announced an Accuracy Scales product, but it has been deploying “smart delivery scales” as part of the same 2025 tech capex cycle (Restaurant Dive, May 2, 2025). The functional category is identical. The fourth piece — physical multi-lane rebuilds — is the McDonald’s-specific addition, driven by McDonald’s drive-thru volume more than by the technology stack.

The roughly 13,700 US McDonald’s locations Rice has under his portfolio aren’t getting one product in 2026. They’re getting a four-part bundle that, on any vendor’s slide deck, would be described as a single “AI-enabled drive-thru” suite. And that suite is, technologically, the same suite Wendy’s has been running for thirty months at a smaller scale.

I want to flag what I am not claiming. I am not claiming McDonald’s licensed Wendy’s technology, or even that anyone at McDonald’s is using the words “the Wendy’s playbook” in the building. The Google Cloud partnership that underpins both deployments was signed by McDonald’s in December 2023 (per Google Cloud’s press release of that date), six months after Wendy’s pilot went live but before Wendy’s had publicly demonstrated FreshAI at scale. The deployments are siblings, not parent-and-child. What I am claiming is that, looking at the working set of public drive-thru voice AI programs in the US in January 2026, the Wendy’s deployment is the only one that has gone from pilot through scale-out without a public rollback — and that the McDonald’s 2026 plan, on the technical merits, is converging on the same operating model.

The accuracy and latency math, on the public record

This is the section every operator skips to. The honest version requires careful reading.

The 22-second time saving Wendy’s reported in 2024 (Fortune, October 15, 2024) is the most-circulated operating number in drive-thru voice AI. It is also a single-region, early-test result with a comparison baseline of “regional average” — not a controlled study. I would read it as directionally meaningful and not literal.

The 80-basis-point company-operated margin improvement Wendy’s attributed in part to FreshAI on the Q4 2024 call (Restaurant Business Online, February 14, 2025) is a real disclosed number — but the words “in part” carry weight. Same-store sales were down 2.8% in Q1 2025 (Restaurant Dive, May 2, 2025). The labor efficiency story is real. The “FreshAI is moving the comp” story is not yet on the public record.

The accuracy claim has no specific public percentage. Wendy’s says “order accuracy is significantly improved” in restaurants using the tools (Restaurant Dive, May 2, 2025); the company has not published a hard number. McDonald’s IBM-era target was 95% accuracy; the IBM system reportedly never cleared the mid-80s (per CNBC’s June 2024 coverage). Wendy’s hasn’t named a target or a current rate publicly. If anyone in trade press has, I haven’t found it.

The check-size lift is partially on the record. Wendy’s says FreshAI “consistently recommending additional menu items” has increased average check size (Wendy’s, May 2, 2025). The dollar value of the lift is undisclosed.

The customer-preference data is the awkward number in the room. A January 2025 YouGov survey found 55% of Americans said they’d prefer a human to take their drive-thru order; only 4% said they’d actively choose an AI chatbot (per The Takeout’s coverage, March 5, 2024, and similar polling reported through 2025). That is not evidence that voice AI is failing. It is evidence that the customer-preference headwind is real and that operating margins from the technology have to clear it.

The operator-side takeaway is uncomfortable. The public accuracy and latency record on FreshAI is better than every other deployment in the category — but the operating numbers that are actually public are softer than the vendor narrative suggests. There is enough to argue the program is working. There is not enough to argue, from the public record alone, that it has crossed the threshold of “obvious unit economics” the way drive-thru kiosks did a decade ago.

I think it has. I am explicitly flagging that as interpretation.

What “the only one not paused” actually means

The brief I was given framed Wendy’s as “the only large-chain drive-thru voice AI not paused or rolled back.” That is a clean line. The actual landscape is messier and worth getting right, because it is the landscape McDonald’s is launching into.

Programs that have publicly paused or wound down:

McDonald’s-IBM, July 2024 (Restaurant Business Online, June 2024; CNBC, June 17, 2024). The defining public failure of the category. McDonald’s-Google in 2026 is the restart, not a continuation.

Del Taco / Presto, originally reported as paused in 2024 (The Takeout, March 5, 2024) — although by late 2025, Del Taco was reported expanding voice AI across roughly 600 restaurants again (Canopy AI drive-thru census, October 14, 2025). The reversal is itself instructive: pause, regroup, resume on a different vendor.

Applebee’s, Chili’s, Red Lobster, on Presto’s digital ordering tablet product — adjacent category, same vendor story. Confirmed non-renewal in 2024.

Programs that are still active and not publicly paused, per the most recent industry census (Canopy AI drive-thru census, October 14, 2025):

Wendy’s FreshAI (160+ stores climbing to 500-600 in 2025), White Castle Julia (~100 stores, ~30% of footprint), Checkers & Rally’s with Hi Auto (~350 stores), Bojangles Bo-Linda (expanding across hundreds), Taco Bell (testing toward a 500-store rollout), Carl’s Jr. / Hardee’s with Presto Voice (hundreds of stores), Panda Express (~30 stores with SoundHound), and a long testing tail (KFC, Pizza Hut, Church’s, Krystal).

What the honest read of that list shows is that drive-thru voice AI in the US is not a one-survivor category. Multiple chains are running production deployments. What is true and worth saying clearly is that Wendy’s is the only one running at the scale, time horizon, and architectural-stack profile that maps directly onto what McDonald’s announced for 2026. The 22-second time-savings number, the Google Cloud LLM foundation, the English/Spanish language coverage, the integration with order history and customisation, the company-operated and franchise rollout in parallel — that is FreshAI’s profile, and that is the profile of the McDonald’s voice product being described in the January 2026 trade press.

The other chains’ deployments are real. They are not the comp for what McDonald’s is about to do. FreshAI is.

What this means for operators

Three operator readings, in order of how much I’d weight them.

One: voice AI in drive-thru is no longer a category question. The 2024 read — does voice AI work in QSR at all? — is closed. There are too many production deployments running too long with too few public failures for that to be a serious question in 2026. What replaces it is a per-vendor and per-architecture question: which stack works at which scale, at which per-unit cost, in which menu environment. The McDonald’s 2026 announcement, read against the FreshAI 30-month operating record, is the moment the category transitions from “is this real” to “which stack.” See the Voice Agent Maturity Curve for the framing — drive-thru voice AI is moving from Stage 3 to Stage 4 in real time, and the McDonald’s-Wendy’s convergence on Google Cloud is the proof of it.

Two: the per-store economics are still not public, and that matters. If you are running a non-QSR concept and a sales rep tells you that “Wendy’s is making it work at $X per store,” ask for the source. The Wendy’s $100-110M 2025 tech capex disclosure covers a half-dozen products together. The per-FreshAI-unit figure isn’t broken out. I have seen back-solved estimates floating in industry decks; the divisor they are using is the wrong number. Wait for an investor-relations breakout or a franchisee disclosure before pricing a deployment off it.

Three: the customer-preference headwind is real and your brand has to clear it. 55% of Americans would prefer a human ordering experience. That is the headwind every voice deployment is operating against. Wendy’s and McDonald’s have brand permission to run a slightly impersonal drive-thru experience; they have always run a slightly impersonal drive-thru experience. A mid-market casual operator running a phone-reservation agent against the same headwind does not have the same brand permission. The McDonald’s-Wendy’s evidence is highly relevant to drive-thru QSR and less directly applicable than vendors will tell you to hotel F&B, full-service casual, or fine-dining. I have made this point before in the hotel F&B voice-AI piece and the McDonald’s case study; it’s worth repeating here.

A falsifiable prediction

I want to close with a specific bet, because that is the only way operator essays compound.

The prediction: of the ~13,700 US McDonald’s restaurants, the first 1,500 voice-AI deployments in 2026 will be disproportionately concentrated in company-operated restaurants and franchisee groups in markets where FreshAI is already live — specifically Florida and Ohio, the two states where Wendy’s ran its 36-restaurant pilot through 2024. The reasoning: a Google Cloud delivery team that has spent two years tuning a Wendy’s voice agent against Floridian and Ohio drive-thru audio — accent profile, ambient noise, menu-modifier language — has a meaningful adaptation advantage when McDonald’s brings a similar workload up in the same physical regions. The same edge-compute hardware in the same Google Distributed Cloud region. The same regional model fine-tuning. The same Capgemini engineering teams (per March 2026 reporting on the McDonald’s-Capgemini expansion, referenced in my earlier McDonald’s case study).

If by December 2026 the bulk of McDonald’s US voice rollouts are in Texas, California, and the Pacific Northwest — markets where Wendy’s FreshAI has not been concentrated — my prediction is wrong, and the implication is that McDonald’s voice-AI rollout sequencing is being driven by something other than vendor delivery readiness (most likely franchisee group dynamics, or pure marketing-friendly market selection). If it’s concentrated in the Florida-Ohio corridor and the contiguous Southeast, my prediction is right, and the implication is that the McDonald’s deployment is operationally inheriting from the Wendy’s deployment in a measurable way — vendor stack, regional tuning, and edge-compute infrastructure.

I’ll re-check in twelve months. The prediction is on the record; the metric is the geography of the first 1,500 US installations as disclosed in McDonald’s 2026 earnings call commentary or trade press coverage. If it’s there to check, I’ll check it. If McDonald’s keeps the rollout geography private, the prediction is unfalsifiable and I’ll downgrade it.

— Naomi covers operators for TableTransfers. Tips: naomi@tabletransfers.com.