The August Earnings Gauntlet: What Restaurant-Tech Investors Are Actually Watching

I printed the August earnings calendar Wednesday morning and circled eight dates in red pen. By the time I got to PFG on the 13th I’d used most of the margin of the page for arrows pointing between names — Toast to Olo, Olo to DoorDash, DoorDash to Sweetgreen, Sweetgreen back to Cava — because the prints aren’t really independent events. They’re the same vote, taken eight times in eleven sessions, on whether the AI capex restaurant-tech sank into 2024 and the first half of 2025 is starting to convert into operating leverage. The sell-side notes are going to lead with the headline beats and misses. That is not the read. The read is three lines on three different income statements, and you can see all three move in real time across this calendar.

Bottom line up front: net revenue margin at Toast and Olo, ARR per location at Toast and Olo (and implicitly at SevenRooms inside DoorDash), and the year-over-year delta in restaurant-level labor cost at Sweetgreen, Cava, and Dutch Bros. Everything else — same-store sales, take rate, GAAP-to-non-GAAP bridges, even guidance — is downstream of those three. If you only have twenty minutes per print, spend them there.

The calendar, and why it matters in this order

The order matters because it front-loads the platforms. Olo reports Monday August 4 — and for the first time in five years, there will be no earnings call, because Thoma Bravo’s take-private has the company in customary blackout. That is its own data point, which I’ll come back to. Toast and Marriott print Tuesday the 5th. DoorDash and Dutch Bros on the 6th. Then Thursday the 7th is the heaviest single session of the month: Sweetgreen, Yelp, US Foods, and RBI all in one tape. Cava follows on the 12th and PFG closes the restaurant-tech-adjacent block on the 13th.

What’s useful about that sequencing is that the platform reads come first and the operator reads come second. By the time Sweetgreen and Cava print on the 7th and 12th, the market will already have Toast’s gross margin and Olo’s release in hand. If the platforms guided down on subscription gross margin — that’s the AI capex tell — and then Sweetgreen or Cava posts a step-up in labor productivity, you’ll know the cost is going one place and the savings are going somewhere else. That’s the asymmetry investors should be sizing.

Line one: net revenue margin (Toast, Olo)

For Toast, net revenue margin is the ratio of GAAP revenue less payment processing costs less hardware cost of goods less third-party platform fees, divided by gross payment volume. It is the closest thing the company reports to a SaaS gross margin on the payments-attached business, and it has been the single best leading indicator of operating leverage Toast has produced. Q1 2025 ran roughly 56–58 bps, depending on how you treat hardware subsidies. Q2 consensus is calling for a couple of basis points of expansion. I think the more interesting question is whether ToastIQ-related cost — the conversational AI assistant that’s been in pilot since 2024 — is sitting in operating expense or already migrated to cost of revenue. If it’s in cost of revenue and the line still expanded, that’s a real signal. If margin held flat because the AI cost is parked in opex below the gross-margin line, the print looks better than it should and the Q3 setup gets harder.

I went deeper on the ToastIQ pilot in a forthcoming May piece on what the Sous Chef pilot taught Toast about what restaurant AI actually has to be — the short version is that the chat box was the experiment and the upsell prompt at order entry was the product. The expense question matters because the upsell prompt is the part with the obvious payback. It runs at order entry, it lifts average order volume in pilot data, and it scales linearly with GPV. The chat assistant is the part where the unit economics are still being written.

For Olo, the calculation is simpler and the print is going to be opaque on purpose. Thoma Bravo is in the middle of taking the company private at $10.25 per share. The deal closes in the second half of the year. There will be no call, no guide, no Q&A. The 10-Q will land and that’s the whole information set. What investors should mark is the implicit ARR per location — total subscription revenue, divided by the average active location count, annualized. Q1 2025 ran roughly $5,400 per location per year, against a fleet of about 87,000 locations. If that number expanded materially in Q2, the Thoma Bravo bid looks underpriced relative to a public-market multiple on the same KPI. If it compressed, the bid looks fair. Either way, the silence around the print is the most expensive call Olo has skipped in five years, because the buyer is the one writing the check.

Line two: ARR per location (Toast, Olo, and the SevenRooms tell at DoorDash)

ARR per location is the cleanest test of whether restaurant-tech is winning attach — selling more products into the existing footprint — versus just adding logos. The reason it matters this quarter is that the public comps have been adding logos at roughly the same clip for eighteen months, and the operating-leverage story increasingly depends on the same store buying more SKUs. Toast doesn’t report ARR per location directly, but you can back into it from disclosed location count, subscription revenue, and the SaaS-attach disclosures management has been incrementally giving on calls.

Olo, again, won’t say much. But the DoorDash print on the 6th will have a different problem: SevenRooms closed inside the quarter, and the comparability of the marketplace gross margin line is going to be ugly. The question I want answered is whether DoorDash discloses SevenRooms subscription revenue as a separate line — or buries it inside “platform services.” If it’s separate, you can compute the SevenRooms ARR per booked-cover. If it’s buried, it’s because the unit economics aren’t yet where management wants them to be public. I wrote about the deal in an upcoming May piece on what the SevenRooms acquisition actually buys DoorDash — and the punchline there is that DoorDash isn’t a delivery company anymore. The August print is the first chance to see that thesis in segment-reported numbers. I wouldn’t bet on the disclosure being clean. The market will give them one quarter of grace.

Line three: labor-cost trajectory (Sweetgreen, Cava, Dutch Bros)

This is the line that decides whether the AI capex on the platform side is actually showing up on the operator P&L. The mechanism is straightforward: if Toast’s upsell prompt, Olo’s order-management automation, and DoorDash’s drive-thru and back-of-house tooling are working, the operators using those tools should be turning more revenue with the same — or fewer — labor hours. Restaurant-level labor as a percentage of restaurant revenue is the cleanest read.

Sweetgreen on the 7th is the most interesting because the Infinite Kitchen ramp is well past the proof-of-concept stage. Q1 2025 disclosed labor cost as a percentage of revenue around 28%, against a fleet where roughly 12% of stores have Infinite Kitchen. If Q2 prints with that percentage moving down a hundred basis points while the IK-enabled store count creeps up, the conversion math gets interesting fast. Cava on the 12th doesn’t have the equivalent hardware bet, but it has been quietly modernizing labor scheduling, and the comparison between Cava’s labor line trajectory and Sweetgreen’s is going to be the second-derivative read on whether the productivity gains are coming from the hardware or from the scheduling software. Dutch Bros on the 6th is the wild card — the throughput model is so different that the labor line behaves differently from quarter to quarter, but a clean print there would suggest the gains are not specific to the salad-bowl format.

What the prints don’t tell you, and where the M&A read shows up

A few things to flag. First, none of these companies are going to volunteer the AI capex number directly. You will have to triangulate it off R&D expense, capitalized software in the cash flow statement, and qualitative commentary on the call. Second, the Olo blackout means the most useful platform print of the month will be the quietest, and that’s a feature, not a bug — the buyer is the one paying for the silence. Third, the operator prints on the 7th and 12th are going to be read against expectations set by the platform prints earlier in the week. If Toast guides cautiously on Tuesday and Sweetgreen prints a clean labor number on Thursday, the gap between platform multiple and operator multiple narrows.

The M&A consequence I’m watching is whether any of the operators — Sweetgreen, Cava, or a third party I won’t name on a calendar piece — start sounding like buyers on the call. The deeper integration play I covered in a forthcoming May Bottom Line on the DoorDash/SevenRooms read-through is the template. Operators who don’t own their reservation funnel, their order funnel, or their loyalty layer are going to find themselves either renting from DoorDash or buying their own. The August prints will tell you which way the operators are leaning. Tone on the call matters more than the numbers here. If a CFO uses the word “in-housing” or “vertical” in scripted remarks, you’ll know.

The headline beats and misses are going to dominate the day-of coverage. I would not optimize for that. Pull the three lines, lay them across the eight prints, and the picture you get is whether the AI capex of the last eighteen months is — finally — starting to pay back. That is the only question worth asking this month.

— Marcus edits The Bottom Line for TableTransfers. Tips: [email protected].