Similarweb Cashes the Check — Then Stops Talking About the Thing That Wrote It
Revenue reaccelerates to 15.6%, GAAP profit arrives for the first time, and NRR finally inflects — just as the LLM-training and GEO/AEO language that built the narrative goes quiet.
SMWB · Earnings Call · 2026-11-17
The quarter the story turned into arithmetic
For five straight calls, Similarweb sold a promise: AI would make its web-traffic data indispensable, and the money would follow. On 17 November the money showed up in the P&L. Revenue of $83 million, up 15.6% year-over-year, was the largest sequential dollar add in company history. Non-GAAP operating profit of $10.1 million produced a roughly 12% margin against $4.5 million a year earlier, and GAAP diluted EPS printed positive $0.04 — the first non-negative GAAP number in the company's public life. Normalized free cash flow of $10.5 million (about 13% margin) came alongside $80.2 million of cash, no debt, and an untouched $75 million line.
That combination — revenue accelerating and margin accelerating — is what management has spent a year promising while delivering one or the other. The keyword board reflects the shift in register. The company's top theme this quarter is big opportunity, with company history sitting near the top of the list on the back of phrases like "highest operating margin in company history," and cash generation now a headline topic rather than a footnote. A software company that has spent twelve quarters talking about products is suddenly talking about the scoreboard.
The AI mix is the engine behind it: “AI-related revenues ... reached almost 20% of revenues in the third quarter. This is up from 11% at the end of the fourth quarter of 2025” — Ran Vered, Chief Financial Officer · 2026-11-17. AI deals accounted for roughly two-thirds of ARR added, the second consecutive quarter in which AI drove the majority of new ARR.
What management stopped saying
Here is the tension worth pinning down. Last quarter's keyword set was littered with the mechanics of selling training data: use case at the very top, then Large Enterprise, LLM training, post-training, large language models, agents in production. This quarter, almost none of that language survives. The vocabulary has been replaced by proprietary data, "AI infrastructures," "capable model," and upmarket.
Gone too is the Gen AI visibility / GEO-AEO pitch that dominated the February call, when Chief Business Officer Maoz Lakovski made the case at length:
we are a data company. We have a meaningful data mode ... this is one thing that we are unique from any other player in this market, and we think we're going to win.
Lakovski does not speak at all in the current Q&A. AI Studio, MCP connectors, and ChatGPT integration — all top-15 company keywords as recently as Q1 2026 — have dropped out entirely, and the term ChatGPT registers as a momentum decliner in the intervening quarters. What replaced all of it is a blunter, more defensible pitch, which Or Offer framed as the reason the data is valuable at all:
AI is only as powerful as the data behind it. Models can reason, but they cannot see the digital world on their own. AI need context.
That reframing is a retreat from a red-ocean product narrative to a data-licensing narrative — and it is the more honest one, because it is where the gross margin actually is.
A promise from August, paid in November
NRR had been the analyst community's punching bag for a year. On the August call, Offer explained the mechanics of the fix: “because we know that this quarter NRR was very, very high, we already know that the next quarter's NRR will continue to increase” — Or Offer, CEO and Co-Founder · 2026-08-12. It arrived, and earlier than the model assumed: 104% overall, the best in three years, and 114% for customers above $100,000 in ARR. Management explicitly conceded it "came through earlier and in greater magnitude than expected" — rare phrasing from a team that has spent four quarters defending a trailing metric. The improvement in NRR is now a realized fact rather than a forecast.
The guidance band confesses more than the guide
In February, new CFO Ran Vered widened the revenue range to $10 million precisely because the big deals were unforecastable: “we widened the range to $10 million just because we know -- we see these fills in the pipeline but the timing of them to land is not that clear” — Ran Vered · 2026-02-18. This quarter's Q4 revenue guide is a $2 million band ($83.9–85.9 million), and so is the full-year range ($318–320 million). The lumpiness has been converted into contracted backlog: RPO of $334 million, up 25%, with 69% recognizable inside twelve months, deferred revenue up 17% to $123 million, and ARR contracted under multiyear contracts at 66% of ARR versus 58% a year ago. Offer's framing has hardened accordingly: “all of the deals are ARR and yearly commitment and most of them or if not all of them are multiyear commitment” — Or Offer, Co-Founder and former CEO · 2026-11-17.
The other confession is in the customer count. Accounts above $25,000 of ARR shrank year-over-year, from 1,828 to 1,790 — a deliberate surrender of small inbound deals in favour of depth. Average account value in that cohort rose 25% to $158,000; the 474 customers above $100,000 pay $464,000 on average, up 27%. Those two cohorts are now 90% and 70% of ARR respectively. And on the concentration question that has hung over the whole AI thesis, Vered was unusually flat: “We don't have any single customer that is above 10%” — Ran Vered, Chief Financial Officer · 2026-11-17.
The tape is fighting the narrative
There is no usable price series for this $685 million market-cap name in our context, so the market's verdict on this print is a blank — a real constraint given that a history of violent two-way moves is documented in prior calls. What can be read is the category. The tape is actively de-rating AI-adjacent infrastructure over the last ninety days: high bandwidth memory down nearly 29% across fourteen tickers, co-packaged optics down 27% across thirteen, data-center capacity down 29%. Meanwhile the one phrase Similarweb is leaning hardest into — being a trusted source of truth for machine consumption — is among the tape's better ninety-day advancers (AMPL, KVYO, FROG). Similarweb sits on the software side of that divide, not the capex side, which is precisely the argument bulls would make.
The strategic question now belongs to Michael Akkerman, two weeks into the job after nearly twenty years of founder stewardship. His mandate is arithmetic, not narrative: turn six-figure accounts into seven, and seven into eight, with “Headcount is staying flat” — Or Offer, Co-Founder and former CEO · 2026-11-17 as the operating constraint. The NIQ joint product ships around December with a revenue and cash uplift promised for 2027, and the market-research pipeline is described as strong. If the 15.6% growth, 12% margin, and first GAAP profit are the new baseline rather than a quarter of unusually good conversion, the story has genuinely changed. If AI mix stalls near 20% and the SMB cohort keeps bleeding, this was a well-timed exit.