Abridge

Rohit Mittal

@rohitdotmittal

vertical AI and rollups are the two hottest strategies in tech investing right now. every major VC fund is doing some version of this. worth looking at whats actually working and what isnt vertical AI startups are putting up real numbers: Harvey went from $0 to $100M ARR in three years. now valued at $8 billion after raising $750 million in 2025 alone. three funding rounds in one year. Abridge hit $117M ARR and is worth $5.3 billion. deployed in 150+ health systems including Mayo Clinic, Duke, and Johns Hopkins. EvenUp crossed $2 billion valuation serving 2000+ personal injury law firms. OpenEvidence claims 40% of US doctors use their platform and just raised at a $6 billion valuation investors have slightly different thesis but it points 2 main things: 1. vertical AI is solving what traditional SaaS couldnt. multimodal, language-heavy, service-heavy workflows that were too complex for the old generation of software 2. vertical AI can target $11 trillion in US labor spend versus $450 billion in traditional enterprise software. pricing on outcomes instead of seats changes what markets are venture scale overall, the TAM has expanded massively. the rollup strategy is more complicated: - General Catalyst has $1.5 billion allocated to their Creation Fund for AI-enabled rollups. - Thrive Capital is backing Long Lake (18 HOA management acquisitions, $600M raised) and Crete (targeting $500M in accounting firm acquisitions). Khosla is in on similar plays; they recently announced a partnership with OpennAI to work with their companies the idea is to buy traditional service businesses, integrate AI, expand margins. General Catalyst incubated Crescendo which acquired PartnerHero and hit ~$90M in revenue by May 2025 but venture-backed rollups have historically struggled. Thrasio collapsed into Chapter 11 in 2024 after raising billions to roll up Amazon sellers. debt load became unsustainable. Perch and the other aggregators mostly didnt work either. Harvard Business Review found two thirds of rollups are value-neutral or worse. the AI angle is supposed to change this. but 42% of enterprise AI initiatives were discontinued in 2024 according to S&P Market Intelligence, up from 17% in 2023. theres still a gap between AI capability and enterprise adoption what separates the winners the vertical AI companies that are working have deep workflow integration, not surface-level features Abridge isnt just transcribing conversations. theyre embedded in Epic workflows, generating billing codes, handling prior authorization. their Contextual Reasoning Engine handles the full documentation stack. switching cost is real Harvey is training on firm-specific data, customizing to internal processes. 235 customers in 42 countries including most of the top 10 US law firms. Bloomberg notes traditional SaaS companies trade at 10x ARR. Harvey is at 50x Toast and Procore show the long-term vertical SaaS pattern. Procore took 10+ years to get to $10M ARR but less than 10 more to hit $1B. they still have <2% of US construction companies. Toast has 13% of US restaurants after more than a decade. these businesses compound slowly but the terminal value is significant the common thread: they become operating systems for their industries. payments, scheduling, compliance, financial services layered on top of core workflow software. Toast does lending now. Procore added payments in 2023 what isnt working wrapper startups are struggling. Jasper raised $100M+, hit ~$90M ARR, then had to pivot after ChatGPT launched similar capabilities. valuation cuts, executive turnover. still alive but a different company now ~90% of AI startups fail within five years by most estimates. 2024 saw 966 US startup shutdowns versus 769 in 2023, a 25.6% increase. https://t.co/ZUld9UK51Z filed for bankruptcy in 2025 after burning $445 million when it turned out their "AI" was actually hundreds of offshore developers companies without proprietary data or deep workflow integration are exposed every time OpenAI or Anthropic ships a new feature on the rollup side, PolyAI spent six months in 2019 exploring contact center acquisitions and walked away. their conclusion per a Fortune piece: "Business Process Outsourcing firms are not trusted to innovate, not rewarded for innovating, and not allowed to innovate." Concentrix gets cited as a BPO transformation success story but their EV/EBITDA multiple is still in low single digits despite AI products deployed to 1000+ customers where this goes vertical AI companies that own full workflows, have proprietary training data, and price on outcomes will likely do well. Abridge has 1.5M+ medical encounters training their models. thats hard to replicate the rollup thesis requires PE-level operational expertise. General Catalyst might have it given theyre hiring real operators and avoiding debt in early stages. but running M&A playbooks isnt a natural VC skill. Alpine Investors 2024 scorecard shows the deal volume and integration discipline required. most funds dont have that infrastructure the established vertical SaaS companies are actually well positioned. Procore, Toast, ServiceTitan have the customer relationships and workflow depth. AI makes them better. the question is whether new entrants can build AI-native platforms fast enough to displace them most vertical AI startups will fail like most startups fail. but the ones that work will be worth more than the horizontal SaaS winners from the last cycle. theyre going after bigger budgets with stickier products. Harvey and Abridge are in a different category than the average AI startup. the rollup strategy will mostly produce mediocre returns outside of a few well-resourced operators who actually know how to run these plays

Michael Mignano

@mignano

There is a critical need for new edge data to sufficiently push AI progress forward. And the most obvious sources have already been tapped. At @usv, we are committed to backing teams who are building products that unlock new stores of high quality data. A great example of this is our portfolio company, @AbridgeHQ, which has become the gold standard for medical transcription in clinical settings. Before Abridge, this data simply wasn’t being captured. A similar dynamic partially motivated me to lead’s @meetgranola seed when I was at Lightspeed. But perhaps the most interesting area for new edge data is in the Physical World, where data exists out of reach from the big labs. This post from my partner @rebeccakaden lays out the whole thesis and highlights some of the companies we believe hold the most interesting edge data today. What are some areas or companies we are missing?

Jay Rughani

@JayRughani

3 ai product roles at @a16z-backed infinite healthcare startups: 1. counsel health [series a] — sr product manager, consumer 2. prosper ai [series a] — ai product manager 3. abridge [series e] — product lead, ai/ml dm for intros @CounselHealth @ProsperAI_HQ @AbridgeHQ

Katie Kirsch

@katiekirsch

.@ShivdevRao raised a $300M series E for @AbridgeHQ last year and is hiring across the board 🙌 a few roles that stand out: - Head of Growth Marketing (for up to $275K) - Senior Event Marketer (for up to $180K) - Director of Data Science (for up to $325K) - ML Scientist (for up to $300K) - Head of User Research (for up to $263K) - Director of Product Management (for up to $290K) link to apply in thread:

Vishnu Ravi, MD

@vishnuravi

This weekend, Austin Schoeffler and I spent some time on one of the harder problems in US healthcare at the @AbridgeHQ x @AnthropicAI x @lightspeedvp hackathon: ED boarding. Boarding is the gap after a patient is admitted but before an inpatient bed opens up, one in six admitted ED patients waits 4+ hours. In that gap, the patient sits in the ED while two teams split responsibility and new data keeps arriving, each assuming the other has it. It isn't benign: in-hospital mortality runs ~80% higher past 12 hours of boarding, and long boards cost an estimated $0.9 - 1.3B a year. So we built BoardX, a team of supervised agents that keeps a boarding patient's story current, surfaces what matters against the admission plan, and help the team close the loop. We were lucky to place in the top 6 out of 115 projects and 900 applicants! Congratulations to all the teams and thanks to Chase Yarbrough, @mnedoszytko, @ShivdevRao, @GalymImanbayev, @pmaederyork for putting on such an exciting event!

Wiki Milczyńska, MD

@w_milczynska

I flew 5,500 miles from London for a 12-hour hackathon! We built an agentic system for chronic health management between appointments with @pavanghantasal3 at the @AbridgeHQ x @AnthropicAI x @lightspeedvp healthcare hackathon. Our agent monitors patients, surfaces meaningful change, flags it to the physician with a suggested action (or a reason to hold off). The fun part was that the clinicians and engineers who met that morning were shipping together by lunch. There was no ‘medical people’ vs ‘tech people’. Everyone did both! Met some seriously smart people and learnt a lot. Now back to building @MoyaHealth 🫡

Shiv Rao

Co-founder & CEO

Cardiologist and former UPMC faculty member who founded Abridge in 2018 to ease clinician documentation burden.

Zachary Lipton

Co-founder, CTO & Chief Scientific Officer

Machine-learning researcher and Carnegie Mellon professor who leads Abridge's AI science and engineering.