Toast's AI Strategy: How the Restaurant Platform Learns What Operators Need in 2026

TL;DR#

Toast's AI strategy centers on ToastIQ, an intelligence layer launched in May 2025 that turns the operating system behind roughly 164,000 restaurant locations into a conversational, proactive co-pilot for time-starved operators. By the end of 2025 Toast had added a record 30,000 net locations, crossed $2.0 billion in annualized recurring revenue, and processed $195.1 billion in gross payment volume — a data moat few restaurant-tech companies can match. ToastIQ and its "Sous Chef"-style voice agents draw on millions of transactions to surface upsells, automate marketing, and answer plain-language questions, and Toast President Steve Fredette frames the core problem bluntly: operators sit on "a mountain of data" but have "limited time to sift through it." That same time-starvation is exactly why Toast's qualitative understanding of why an operator struggles, downgrades, or churns lags its operational telemetry — restaurant owners rarely fill out surveys, and usage data plus sparse NPS can show what happened but not why. Conversational AI customer research, where an AI interviewer meets a busy operator in a two-minute spoken exchange, closes that gap. This is the layer Perspective AI builds: AI-first customer research that captures the "why," because it cannot start with a web form.

What is Toast's AI strategy?#

Toast's AI strategy is to embed an intelligence layer — branded ToastIQ — directly into the restaurant operating system it already runs for roughly 164,000 locations, so that AI surfaces recommendations, automates routine work, and answers operators' questions in plain language without leaving the platform. The strategy is "data-first": because Toast captures point-of-sale, payments, payroll, online ordering, and guest data in one stack, its AI can act on real operational signal rather than disconnected exports, positioning Toast as an operator's "right hand" rather than another dashboard to check.

That single-platform data advantage is the spine of everything below. It is also, as we'll see, the strategy's blind spot: telemetry is rich, but the reasons behind operator behavior are thin.

Toast at a glance: the scale behind the AI#

Toast operates one of the largest single-vertical software-and-payments platforms in the United States, and that scale is what makes its AI bets credible. According to Toast's full-year 2025 results, the company finished the year at approximately 164,000 locations — up 22% year over year after adding a record 30,000 net new locations — with annualized recurring revenue above $2.0 billion (up 26%) and gross payment volume of $195.1 billion (up 23%). CEO and co-founder Aman Narang summarized the year as "a strong year for Toast, adding a record 30,000 net locations, growing recurring gross profit 33%, and delivering Adjusted EBITDA margins of 34%."

Five numbers frame the AI opportunity:

• ~164,000 locations running on a single connected platform as of year-end 2025.
• 30,000 net new locations added in 2025 — the company's fastest expansion year on record.
• $195.1 billion in gross payment volume processed across those locations in 2025.
• $2.0 billion+ in annualized recurring revenue, growing 26% year over year.
• 34% Adjusted EBITDA margins, signaling a platform with the cash to invest aggressively in AI.

A platform this large generates an enormous behavioral exhaust — every order, refund, void, shift swap, and menu edit is captured. The strategic question is what Toast can learn from it, and where the data stops short. The same single-platform pattern shows up across the SMB-software world; we map it in our look at how Block and Square turned a seller ecosystem into a research surface and how Bill.com automates finance for small businesses.

How does ToastIQ work?#

ToastIQ works by sitting natively inside Toast's platform and converting raw operational data into timely prompts, personalized recommendations, automated workflows, and conversational answers — so operators get insight pushed to them rather than having to go digging. Toast introduced ToastIQ on May 1, 2025 as an intelligence engine drawing on millions of transactions and interactions, and expanded it in October 2025 with a full conversational AI assistant.

The assistant gives operators three things: a personalized "For you" feed of timely recommendations; the ability to ask complex questions about the business in plain language and get tailored advice back; and the ability to take action — updating a menu, editing a shift, modifying an item — directly inside the conversation. The first ToastIQ features targeted concrete operator pain: Menu Upsells, Shift at a Glance, Digital Chits, an AI Marketing Assistant, and Advertising. Toast also rolled out voice-AI agents (the "Sous Chef" lineage of conversational assistants) that, per company-reported results covered by PYMNTS, lifted average order volume by about 6% through AI-driven upsells.

The framing from leadership is telling. Toast President and co-founder Steve Fredette described the design rationale this way: operators "have access to a mountain of data that can help them make informed business decisions — but limited time to sift through it." ToastIQ, he said, "goes beyond simply answering questions. It provides proactive, personalized recommendations." That diagnosis — abundant data, scarce time — is correct. It also reveals the boundary of a pure telemetry-and-prompts strategy, which we'll get to next.

Toast Sous Chef AI and the conversational turn#

Toast Sous Chef AI is the conversational, voice-and-text agent lineage inside ToastIQ that lets operators and guests interact with the platform through natural language rather than menus and reports. The shift matters because it signals Toast's recognition that the interface to restaurant software is becoming a conversation — a guest asking a voice agent to add an item, or an operator asking "which menu items had the worst margin last week?" and getting an answer in seconds instead of pulling a report.

One operator, Romain Bernus of Le Petit Four Bakery, told Restaurant Technology News that "pulling reports was just too slow when we're so busy. Now, complex reports take five seconds," adding that "Toast IQ changes everything for small businesses like ours." Notably, the same coverage cited Toast research finding that over 81% of restaurant operators agree they will use more AI in the future — strong demand signal for a conversational layer.

Here is the strategic insight Toast has half-grasped: if a conversation is the right interface for operating a restaurant, a conversation is also the right interface for understanding the operator. Toast uses conversational AI to help operators act. The unexploited frontier is using conversational AI to learn from operators — the discipline of conversational customer research, which we explore in why conversations beat surveys for real customer research.

The restaurant operator: the hardest customer to research#

The restaurant operator is one of the hardest B2B audiences to research, because they are time-poor, frequently on their feet during business hours, and structurally unlikely to complete a survey. A restaurant owner working a double shift will not stop to fill out a 12-question NPS form about their POS provider, and Toast's own messaging — "limited time to sift through it" — concedes the constraint.

This is where Toast's data-first strategy runs into a wall. Usage telemetry is fantastic at answering what: this location's order volume dropped 18% month over month, that operator hasn't opened the marketing module in 40 days, this account's payments volume is trending toward a downgrade. But telemetry is silent on why. Did order volume drop because of a new competitor across the street, a kitchen manager who quit, a delivery-app dispute, a pricing change the operator regrets, or a feature in Toast that quietly stopped fitting their workflow? Each of those demands a completely different response — and usage data can't tell them apart.

The conventional fallback is survey-and-NPS listening. But response rates make that fragile: external digital surveys in 2025 typically landed between 5% and 30%, and customer-feedback fatigue is rising as people receive more invitations than ever, a shift we trace in the death of the annual customer survey. For an audience as time-starved as restaurant operators, you skew toward the low end — and the operators who do respond are systematically the least representative (the very happy and the very angry). A 9% NPS response rate from a self-selected slice is not a voice-of-customer program; it's an anecdote with a number attached. We unpack why scores miss the reasoning in the complete guide to voice-of-customer programs in 2026.

Where forms and NPS bottleneck Toast's understanding#

Forms and NPS bottleneck Toast's understanding because they capture a number or a checkbox but not the narrative, and the narrative is where churn risk, expansion appetite, and product-fit problems actually live. For a platform whose churn — per Bessemer Venture Partners' analysis — is driven almost entirely by restaurants going out of business rather than switching to a competitor, the highest-value research question isn't "rate us 1–10." It's "walk me through the last 90 days — what got harder, where did we help, and what nearly made you reconsider?" No form field captures that.

Three concrete failure modes recur across SMB-platform research:

1. The why behind a downgrade is invisible. Telemetry shows a location moving from a premium tier to basic. The reason — cash crunch, a feature they never adopted, a competitor's promo, a bad support experience — is exactly the thing a dropdown can't hold.
2. Pre-churn signals arrive too late. By the time payment volume craters, the operator has often already decided. The early, qualitative tremors ("I'm thinking about closing the second location") never surface because no one asked in a low-effort way.
3. Feature feedback is flattened. A star rating on a new module tells you sentiment polarity; it doesn't tell you the operator turned it off because it added two taps to a workflow they run 300 times a day.

This is the same pattern we documented across SaaS leaders — see how Salesforce approaches conversational customer discovery and how Snowflake handles product discovery on a data cloud. The deeper argument — that an AI-first company cannot reduce customers to schema fields — is laid out in why AI-first customer research cannot start with a web form.

What conversational AI customer research unlocks for a platform like Toast#

Conversational AI customer research unlocks the "why" at survey-impossible scale by replacing the static form with an AI interviewer that talks to operators in their own words, follows up on vague answers, and runs hundreds of interviews simultaneously — turning a population that won't fill out forms into one that will answer a two-minute spoken question. Instead of a 12-field survey emailed to 164,000 operators (and ignored by most), an AI interviewer can ask one open question — "What almost made you reconsider Toast this quarter?" — and probe the answer the way a skilled human researcher would.

For a Toast-scale platform, four use cases stand out:

• Churn and downgrade diagnosis. When telemetry flags a downgrade risk, trigger a short AI interviewer conversation to capture the operator's actual reasoning while the decision is still reversible.
• Onboarding friction discovery. New-location activation is where Toast adds 30,000 accounts a year; a conversational intake flow surfaces where setup confuses operators far better than a completion-rate metric.
• Feature validation before build. Before shipping the next ToastIQ capability, run discovery interviews with operators to validate the problem — the practice behind the complete guide to product-market-fit research in 2026.
• Always-on voice of customer. A continuous cadence of lightweight conversations beats an annual survey blast, an approach detailed in how to build a voice-of-customer program from scratch.

The qualitative depth that AI interviews capture is precisely what NPS and usage data leave on the table; we contrast the two approaches in moving beyond NPS scores and in our roundup of the customer research stack modern teams actually use.

Toast's AI vs. conversational customer research: complementary layers#

Toast's operational AI and conversational customer research are complementary, not competing — one acts on behavior, the other explains it. The table below maps where each layer wins, and why a platform needs both.

The pattern holds across verticals. We've traced the same operational-AI-plus-conversational-research split at consumer-facing platforms in our Shopify customer research case study and in how Stripe researches 4 million businesses. Companies built for customer-experience and product teams can wire the research layer directly into their workflows — see what we build for CX teams and for product teams.

Frequently Asked Questions#

What is ToastIQ and when did it launch?#

ToastIQ is Toast's AI intelligence layer, launched on May 1, 2025, that delivers timely prompts, personalized recommendations, automated workflows, and a conversational assistant inside the Toast restaurant platform. It expanded in October 2025 with a full conversational AI assistant that lets operators ask plain-language questions and take actions like editing menus or shifts. ToastIQ draws on data from roughly 164,000 restaurant locations.

Is Toast Sous Chef the same as ToastIQ?#

Sous Chef is the conversational and voice-agent lineage that now sits within the broader ToastIQ ecosystem, not a separate product. "Sous Chef" originally described Toast's conversational AI assistant, and its voice agents have been credited with lifting restaurants' average order volume by about 6% through AI-driven upsells. Toast has consolidated these capabilities under the ToastIQ intelligence brand.

How does Toast understand why a restaurant operator churns?#

Toast primarily infers churn risk from usage and payment telemetry, supplemented by NPS-style guest and operator feedback. This shows what changed — a downgrade, a volume drop — but not the underlying reason, because time-starved operators rarely complete surveys and usage data can't distinguish a cash crunch from a workflow problem. Conversational AI interviews close that gap by capturing the operator's reasoning directly.

Why are restaurant operators so hard to research with surveys?#

Restaurant operators are hard to survey because they are time-poor, frequently working during business hours, and experiencing rising feedback fatigue. External digital surveys in 2025 typically saw response rates between 5% and 30%, and busy SMB audiences skew toward the low end. The operators who do respond tend to be the least representative — the very satisfied or the very frustrated — which biases the resulting data.

How does conversational AI customer research differ from Toast's operational AI?#

Conversational AI customer research explains why customers behave as they do, while operational AI like ToastIQ acts on what they're doing. ToastIQ converts transaction data into recommendations and actions; conversational research uses an AI interviewer to ask open questions, follow up on vague answers, and capture decision drivers, churn reasoning, and unmet needs that telemetry and NPS scores cannot reveal. The two layers are complementary.

Can conversational AI interviews reach a busy SMB audience like restaurant operators?#

Yes — a short AI-led interview reaches busy operators far better than a multi-field survey because it asks one open question, takes about two minutes, and meets people in conversation rather than a form. The AI interviewer follows up automatically, so a single low-effort prompt yields the depth a human researcher would get, and hundreds of interviews can run simultaneously without adding research headcount.

Conclusion#

Toast's AI strategy is a masterclass in operational intelligence: a single platform spanning roughly 164,000 locations and $195.1 billion in payment volume, with ToastIQ and Sous Chef-style agents turning that data into proactive recommendations operators actually use. But the strategy's strength exposes its gap. Telemetry and sparse NPS tell Toast what its time-starved operators do; they struggle to explain why an operator downgrades, hesitates, or quietly closes a second location. Steve Fredette's own diagnosis — a mountain of data, but no time to sift it — applies just as much to listening as it does to operating.

Conversational AI customer research is the missing layer. When the right interface for running a restaurant is a conversation, the right interface for understanding the operator is a conversation too. That is the wager behind Perspective AI: AI interviewers that talk to your customers in their own words, follow up on the messy "it depends," and run at the scale of a survey while delivering the depth of a one-on-one — because AI-first customer research, like a great restaurant, cannot start with a form. Start a study or explore how it works to hear the "why" your dashboards can't show you.