SoundHound's Restaurant Story Is Real: MOD, Habit, Red Lobster, Torchy's All Live

I have spent eighteen months marking SoundHound’s quarterly disclosures with a yellow highlighter and a question mark. The slides have always been confident — drive-thru voice agents at scale, automotive deals, agentic platform, named “tier-one QSR brands” — but the named-brand list lagged the narrative. Today it didn’t. The Q3 release that hit the wire Thursday morning names MOD Pizza, Habit Burger, Red Lobster, and Torchy’s Tacos as live deployments alongside the already-disclosed Chipotle and Casey’s. Franchise wins at Firehouse Subs, Five Guys, and McAllister’s Deli are listed in the same breath.

The contrarian read is the simple one: the restaurant voice-AI story SoundHound has been telling is real, and the disclosure file finally backs it.

What the print actually shows

Revenue of $42 million for the quarter, up 68% year-over-year, with the full-year outlook raised to $165–180 million. $269 million in cash, zero debt. The 8-K earnings exhibit filed the same morning reproduces the named-customer paragraph verbatim, which matters: a press release can puff; a Reg FD-compliant filing in front of the SEC reads tighter and tracks closer to what an auditor will sign.

A few things to mark. First, the chain list is heterogeneous — MOD and Habit are corporate-heavy footprints, Red Lobster is mid-recovery and trying to operate at lower labor cost, Torchy’s is digitally aggressive growth, Chipotle and Casey’s were already named publicly. The mix tells you the technology is being trialed across operating models, not just at the easiest one. Second, the franchise wins — Firehouse Subs, Five Guys, McAllister’s — matter more than headlines usually credit. Franchise systems are the hardest place to push a vendor down: the franchisor has to qualify the stack, the franchisees have to agree to pay, and the unit economics have to survive a labor-cost defense at the local level. Three franchise systems closing in a single quarter is a real signal.

Third — and this is the one I keep coming back to — Peet’s Coffee has expanded the Employee Assist agent to all of its corporate locations. Employee Assist is the back-of-house voice agent that answers staff questions about procedures, recipes, and policy. It is not the drive-thru product. The fact that an existing customer rolled the second product across the corporate footprint is the kind of disclosure that vendors typically only get when the first product is actually working.

Why I’m reading it as a moat, not a moment

The bear case on SoundHound has always been that voice-ordering is a feature, not a company — that Google, Amazon, OpenAI, or a vertical-AI startup will commodify the stack. That case isn’t dead. But the operator-side moat is wider than the technology moat, and it builds quarter by quarter. Each named chain is a months-long integration with a POS, a kitchen display system, a menu structure that changes daily, and a fraud-and-coupon logic layer that has to behave at the speaker post. The reason you don’t see a dozen vendors with this customer list is not because the models are uniquely good — it’s because the operator-integration cost is brutal, and the trust-building is even worse.

The Motley Fool earnings transcript from the day after the print makes the same point in management’s voice — large-customer AI penetration was a small share of installed base three years ago and is now a majority — but I’d weight the SEC filing higher than the call.

What the operator side has to watch:

• Order-accuracy disclosure. SoundHound publishes order-completion metrics for some customers but not all. The named-chain list raises the stakes on getting an accuracy number per concept.
• Labor model. Voice agents redistribute labor cost rather than eliminate it. Habit and MOD will be the cleanest read on whether the in-store labor freed by the speaker-post agent gets reinvested in throughput or harvested.
• Franchise unit economics. Firehouse, Five Guys, and McAllister’s will be the test of whether a franchisee — paying the SaaS fee out of unit-level P&L — sticks past the trial.

What I’m marking down

A forthcoming TableTransfers framework piece on voice-agent maturity puts SoundHound’s stack in the “operator-grade” tier rather than the “lab demo” tier — which is the right read coming out of this print. The argument I made privately to a couple of operators last week is the one to make publicly now: if you are a multi-unit operator and your POS vendor is selling you on a drive-thru voice product that does not have a named-chain disclosure list of this length, ask why.

This is not a victory lap for SoundHound — the stock is volatile, the customer concentration is real, and the integration backlog is a forward risk. But the gap between the company’s narrative and its disclosure has closed materially. The Q3 print is the first one I’ve read where the named customers carry the slide rather than the slide carrying the customers.

The restaurant voice-AI story is real. The operator question is no longer whether it’s working. It’s which chains have the leverage to negotiate the terms before the moat hardens.

— Maya covers restaurant tech for TableTransfers. Tips: [email protected].