Toast IQ at the Front of the House: How Toast's First Conversational AI Cohort Is Actually Performing

The office sits above the prep kitchen at a fourteen-unit casual-dining group in the Mid-Atlantic, and at 9:47 on a Tuesday morning the only sound is the espresso machine downstairs and the low whine of an old desktop fan. The director of operations — call her R., because her CFO has not signed off on this story being told under her real name yet — pulls up a chair next to me and types a question into what looks like a chat window.

“Which three locations had the worst voids over the last 14 days, and what was the most common reason code?”

The response takes about eight seconds. There is a brief animation that suggests the system is “thinking,” which I have learned to read as “fetching from three or four tables and then asking a language model to write a sentence.” A small paragraph appears, naming the three units, ranking them by void rate against same-period-last-year, and noting that the top reason code at two of the three is “training” — which in this group’s configuration means a manager voided an item because a server rang it incorrectly.

R. nods, copies the answer into a Slack DM to her regional manager, and goes back to her email. Then she looks at me, half-amused.

“That used to be a Friday afternoon,” she says. “Now it’s a sentence.”

This is Toast IQ. It launched at Toast’s annual conference in September, it was already a talking point on the company’s Q3 earnings call on November 4, and as of this week — December 2, 2025 — it is the single most-asked-about line item in every operator phone call I take. The pitch is straightforward enough: a conversational AI layer that sits on top of the data Toast already collects, lets you ask plain-English questions of your operations, and — in its more ambitious framings — eventually closes the loop by acting on what it finds.

The contrarian thesis I am going to argue, after a week inside R.’s deployment and a separate site visit at a six-unit pizzeria group that has had it longer, is this: Toast IQ is a Q1 2026 buying decision for multi-unit operators on Toast already, and a Q3 2026 wait-and-see for everyone else. The reasons are not the ones Toast is putting in the press release. They are about what conversational AI actually replaces in the operator workflow versus what it merely augments — and the line between those two outcomes is sharper than the marketing suggests.

What Toast actually shipped, and what it didn’t

Let’s anchor on what is on the record. On November 4, Toast reported Q3 2025 results: ARR of $2.0 billion, up 30% year-over-year, and a net add of approximately 7,500 locations in the quarter, bringing the installed base to roughly 156,000. Those are not small numbers. To put them in operator terms: roughly one in eight casual-dining and quick-service locations in the United States now runs on Toast, and the pace of net adds has not slowed despite the maturation of the category.

The Q3 release talks about Toast IQ as a strategic priority but does not — and this is important — break out IQ-specific monetization or attach rates. There is no SKU-level disclosure of how many of those 156,000 locations have actually turned on a Toast IQ capability. CFO Elena Gomez referenced it as part of the “next chapter” of the platform on the call. CMO Kelly Esten, talking about the Alinea selection and the wider product narrative, has been the more visible internal voice on what IQ means: a conversational front-end that, in her framing, is supposed to make the deep operational data Toast collects legible to a human running three restaurants without a BI analyst.

Then, two weeks ago, Toast announced something more concrete: 120 new R&D jobs in Dublin, IDA Ireland-supported, over a three-year horizon, explicitly framed as an AI engineering investment. Dublin is not a random pick. It is where Toast already has an EMEA hub, where Stripe and Workday have proven that AI/ML talent can be hired at competitive rates relative to Boston or the Bay, and where the time-zone overlap with both U.S. coasts and European customers is workable for a 24-hour engineering org. The press release is light on what those engineers will build, but the read between the lines, from people I’ve talked to who work adjacent to that team, is that Dublin gets the model-evaluation, agentic-workflow, and natural-language-to-SQL workstreams. The roadmap items, in other words, that determine whether Toast IQ becomes the front of the house or stays a side window.

Restaurant Technology News framed all of this as the opening of a “next phase” of competition — and they’re right, in the sense that the field of POS-plus-platform vendors who have all said “AI” out loud in the last twelve months is now going to be forced to ship something that operators can actually touch. SpotOn, Square, Lightspeed, and Olo have all been making noises in this direction. Toast has the distribution and the data depth to set the floor. The question for operators is not whether Toast will eventually win on capability — they probably will — but whether the version available today changes the buying decision today.

After a week inside, my view is: for some operators, yes. For others, the right move is to wait two quarters.

What “conversational AI” actually replaces in R.’s week

R.’s operation runs fourteen units across three states, all on Toast. She has been a customer since 2019 — early enough that she remembers the version of Toast Reports that ran on Flash, and old enough as a Toast-stack veteran that she has, by her own count, three separate dashboards she built in Toast’s earlier analytics tooling that nobody on her team uses anymore.

When Toast IQ rolled out to her account in late October as part of an early-cohort program, she did what every experienced operator does with new technology: she gave it to the people most likely to break it. In her case, that meant her three regional managers and the GM of her highest-volume unit.

I asked her, on Monday morning, to walk me through what she had stopped doing in November because of Toast IQ. Her answer surprised me, because it was not about the things Toast tends to put in the demo. The demo, if you’ve sat through one, leans heavily on labor scheduling and menu mix — the headline-friendly capabilities. R.’s actual list:

First, the Monday morning “what happened over the weekend” loop. She used to spend ninety minutes pulling pivot tables out of three different reports — voids, comps, average ticket by daypart, and one custom report her old analyst built that compared bar-to-food ratio against the same weekend last year. She now asks IQ four questions in sequence and gets answers in under three minutes. The pivot tables still exist; she has not deleted them. But she opens them only when IQ gives her an answer she wants to interrogate at the row level.

Second, the weekly comp-and-void review with each unit’s GM. The previous workflow was: GM emails her a PDF, she opens it, asks two or three follow-up questions, GM digs in, replies the next day. Now the GM and she are in the same Slack thread, both with IQ access, and the back-and-forth happens inside thirty minutes on Tuesday afternoon.

Third — and this is the one she said matters most — the ad-hoc “I have a hunch” question. Last Wednesday, she suspected one of her units was over-staffing the bar relative to bar sales during the 4-to-6 happy hour window. She used to need to remember to ask her ops analyst, who used to need to write a query against Toast’s labor data and the sales data and join them on time-of-day. She typed the hunch into IQ. She had the answer — yes, two of three Tuesdays in November the unit was overstaffed by approximately 0.8 FTE in that window — before she had finished her coffee. She forwarded it to the GM. Bar labor at that unit dropped by an hour the following Tuesday.

This is the pattern I have now seen at five separate operators over the last six weeks: Toast IQ replaces, with surprising completeness, the kind of work that previously required a junior analyst or a patient operator with two hours and a spreadsheet. It does not replace, and is not pretending to replace, the manager judgment that follows. It compresses the find-the-question and fetch-the-data parts of the workflow into roughly the same act.

That compression is the actual value. The conversational interface is the delivery mechanism, not the magic. The magic, such as it is, is that Toast has the data already — sales, labor, menu, customer, sometimes inventory — sitting in a coherent schema, and the LLM-on-top finally makes that schema accessible without SQL. For operators who already pay Toast to be their system of record, this is meaningful. For operators who have their data spread across a POS, a separate scheduling tool, a separate inventory tool, and a separate CRM, IQ would only be one-fourth as useful, because it would only see one-fourth of the picture.

This is also why the question of who Toast IQ is for, in Q1 2026, is so much narrower than the keynote implied.

The Q1 2026 buying case (for some operators)

If you are a multi-unit Toast operator with at least four locations, with most of your operating data already inside the Toast platform, and with a manager-and-up team that has been quietly drowning in dashboards for the last two years, Toast IQ in its current form is a Q1 2026 buying decision. The math, in my read, is fairly clean.

The replacement target is not headcount. Nobody R. employs is going to lose their job because of IQ. The replacement target is hours of high-judgment people doing low-judgment retrieval. R. estimates, conservatively, that IQ has freed up four to six hours a week of her own time and another two to three hours a week from each of her three regional managers. Call that fifteen hours a week of senior-operator capacity that is now going to things other than pivot tables. If you value that capacity at even a modest $75 an hour, fully loaded, you are talking about something in the neighborhood of $60,000 a year of capacity per fourteen-unit group — before you count any actual operational decisions made better or faster because the data showed up on time.

That is the upside case. The downside case is real and worth naming. Three concerns recurred across every operator I talked to:

The first is the hallucination risk on the long tail of questions. IQ is good at the common questions — voids, comps, mix, labor cost, sales-by-daypart — because Toast has spent two years tuning the underlying schemas and the prompt scaffolding for exactly those queries. It is, in the operators’ experience, less reliable on weirder questions: things like “compare wine attach rate on tables of four versus tables of two over the last 60 days.” Sometimes it answers brilliantly. Sometimes it answers confidently and wrongly, in a way that requires a domain expert to catch. R. has trained her regional managers to spot-check anything that surprises them by running the equivalent native report. The cost of that habit is real; the benefit is that they catch the occasional bad answer before it propagates.

The second is the action-loop gap. Today, IQ is read-mostly. It answers questions, it sometimes proposes actions (“you appear to be overstaffed in bar labor during the 4-to-6 window at unit X”), but it does not take the action. The closed-loop version — the thing where IQ moves a shift, edits a schedule, kills a menu item, comps a check — is implied by the roadmap, hinted at in the Dublin investment, and largely absent from the product today. For operators who want to delegate operating decisions to the agent, this is the gap. For operators who want a faster advisor, it is irrelevant.

The third is the lock-in deepening. Toast already has, by any reasonable measure, the deepest data footprint in the restaurant POS category. IQ deepens it further, because once your regionals have spent a quarter operating through the conversational interface, the switching cost is not just data migration — it’s workflow migration. You are not just changing systems; you are retraining your team to find the question in a different shape. That cost is real, and it is going to be a factor in any future renegotiation of Toast’s take rate.

For a multi-unit Toast operator, all three concerns are manageable. The hallucination risk is mitigated by spot-checks. The action-loop gap is fine if you are buying an advisor. The lock-in is a known cost of a relationship you were not, realistically, going to walk away from. The buying case in Q1 holds.

This is the context in which a forthcoming May piece will revisit IQ — by then the agentic version, the one with the action loop, will be more legible, and the lock-in question will sharpen. But for now, in early December, the read is: if you fit the profile above, you are not waiting. The upside-to-risk ratio is too clean.

The Q3 2026 wait-and-see case (for almost everyone else)

The harder argument, and the one I have been making to operators on the phone for the last two weeks, is the wait-and-see argument for everyone outside that profile.

If you are a single-unit operator on Toast, IQ is interesting but the ROI is thin. Most of the workflows it compresses — the Monday-morning loop, the weekly review with the GM — collapse to “you do them yourself in less time anyway” at single-unit scale. The conversational interface is pleasant. It is not transformative.

If you are a multi-unit operator but you are not on Toast, the wait-and-see argument is stronger still. The decision you are actually being asked to make is not “should I adopt Toast IQ?” It is “should I migrate my POS, my scheduling, my reporting, and my data foundation to Toast in order to unlock IQ?” That is a fundamentally different decision, with two-to-three quarters of implementation pain, and it deserves the kind of scrutiny that an upcoming desk review will eventually bring to it. For now, the version of IQ I have seen does not by itself justify a platform migration. It justifies adopting IQ once you are already on the platform, which is a much weaker argument for the migration itself.

If you are a multi-unit operator on a mixed stack — Toast plus a separate scheduling tool, plus a separate inventory tool — the wait-and-see is about IQ actually integrating with the other systems you care about. Today it reasons over what Toast sees. It does not reach into Restaurant365, Crunchtime, 7shifts, Olo, or DoorDash unless those flow through Toast. The roadmap says that will improve. The Dublin investment is, I think, one of the bets that it will. But it is not better yet.

There is a fourth case worth naming, which is the case for operators who care most about front-of-house guest experience. That is where Toast has been increasingly vocal about its state-of-the-industry framing — the argument that operators in 2025 are being squeezed on both labor and on guest expectations, and that AI is one of the few levers that can move both. The pitch is real. The current version of IQ, however, is largely a back-of-house tool. The conversational interface sits on operating data; the guest-facing AI capabilities — Sous Chef, Menu Upsells, the kiosk and online-ordering personalization layer — are separate product lines under the broader IQ umbrella. The unification is implied by the brand. It is not yet built. For an operator whose primary pain is guest experience, the version of IQ that matters most is the one that ships in mid-2026, not the one available today.

What the Dublin investment actually signals

Spend a few hours talking to operators about Toast IQ, and you eventually circle back to the question of credibility of roadmap. Operators have been burned, repeatedly, by POS vendors promising AI capabilities that arrive eighteen months late, with half the demo features, and with an asterisk about the pricing.

The Dublin investment changes the credibility calculus, in my read, in two specific ways.

The first is that 120 R&D engineers, hired over three years, is a real commitment. It is not a slide. It is a hiring plan with Irish-government tax structure underneath it, which is a thing that companies do not do casually. The IDA Ireland framing of the deal — three-year horizon, AI focus — is the sort of thing that gets audited, both by IDA and by the local Irish press. Toast cannot quietly walk it back. That makes the implied roadmap — agentic IQ, deeper integrations, model-evaluation infrastructure — more believable than a typical conference keynote.

The second is geography. By placing the AI R&D in Dublin rather than Boston, Toast is — explicitly and implicitly — buying access to a talent pool that the major U.S.-only AI shops have not fully tapped, and one that has a stronger applied-ML and platform-engineering culture than the model-research-only labs. The kind of engineer who builds a robust natural-language-to-SQL system, with the evaluation harness and the failure-mode telemetry that operators actually need, is not the same engineer who is being bid up at OpenAI or Anthropic. Toast is hiring for the former. That is, I think, the right hire to make.

The risk in the Dublin announcement is that R&D investments are by definition slow. Three years is a long time in this category. The competitors are not going to wait. SpotOn, Square, Lightspeed, and Olo will all ship something IQ-shaped in 2026. The window where Toast can set the floor for what operators expect from a “POS-plus-AI” platform is, by my read, roughly six to nine months. That makes Q1 2026 — the window in which operators are doing budget reviews, renewals, and migration decisions for the year — disproportionately important.

Mark interpretation

Let me close with the operator-level read, because at TableTransfers that’s what we owe you.

Mark: Toast IQ is a real product, with a real adoption curve, deployed by a vendor with the credibility and the engineering investment to make the roadmap believable over a two-to-three-year horizon. The Q3 financials underwrite the investment thesis; the Dublin announcement underwrites the roadmap; the operators I have talked to underwrite the day-one utility for a specific operator profile. None of those three pillars is fake. None of them is fully baked, either.

Interpretation, by operator type:

If you are a multi-unit operator on Toast with three or more locations and most of your data already inside the platform: turn IQ on this quarter, give it to your regionals and your highest-volume GMs first, and budget for the agentic version to land in mid-2026. Train your team to spot-check the long-tail answers. Track the hours saved; that’s your renewal-negotiation leverage in 2027.

If you are a single-unit Toast operator: try it, especially if it’s bundled. Don’t change your buying decision on it. Revisit at the agentic launch.

If you are a multi-unit operator not on Toast: do not let IQ be the reason you migrate. Let the migration question be answered on its own merits — payments economics, hardware, take-rate, integrations — and then adopt IQ once you’re inside. The version of IQ available today is not strong enough, by itself, to justify the migration cost.

If you are a multi-unit operator on a mixed stack with Toast as one of several systems: wait for the Q2 2026 integration roadmap. The version of IQ that reads across your full stack is the one that matters for you, and it is not the version shipping today.

The contrarian piece of all this — and the reason I have used the phrase “Q1 2026 buying decision or a Q3 2026 wait-and-see” three times now in conversations with operators this week — is that the keynote made it sound like IQ is a horizontal capability available to everyone. It is not. It is a vertical capability deeply tied to depth of data inside the Toast platform. For operators with that depth, it is one of the strongest product introductions of 2025. For operators without it, it is a reason to watch Toast more closely, not a reason to act yet.

Back in R.’s office on Tuesday morning, after she sends her regional manager the Slack message about the void rates, she looks at the IQ window for another second and then turns back to her email. The conversation took eleven seconds. The pivot table it replaced used to take ninety minutes. Multiply that across fourteen units, three regionals, and a year of Tuesday mornings, and you start to understand why she has already told her CFO that the IQ line item, whatever Toast eventually charges for it, is staying in the 2026 budget.

That is the buying decision in one operator’s office. Across the industry, the same decision is being made in roughly 156,000 conversations between now and Q1. The shape of those conversations — and how many of them end the way R.’s did — is going to set the AI floor for the restaurant category for the next three years.

— Priya covers operators for TableTransfers. Tips: [email protected].