Start With a Wedge That Prints Value

Harvey

L2

Harvey's initial wedge was not "legal AI." It was a specific workflow inside a specific buyer segment: legal research, drafting, and due diligence document review for Big Law attorneys — litigators who spend hours on case review and M&A associates who process massive closing checklists. The founders chose this not because it was easy, but because it was the wedge where failure cost the most and where hourly rates made ROI almost instantly visible. At $1,000–$1,500/hour for a senior Big Law partner, a tool that saves two hours per week pays for itself in days. Proof was built through what would look reckless by conventional sales standards: a months-long free pilot of 3,500 Allen & Overy lawyers running 40,000 queries. But this was deliberate trust construction in a buyer community that views vendor relationships through the lens of professional liability. Simultaneously, founder Winston Weinberg ran what he called the PACER demo tactic — downloading prospects' actual recent federal court filings and running prompts that critiqued their own work: "I would basically download the last thing that they submitted to court. And then I would try to come up with prompts that were like, 'This is bad.' And because they're a litigator and I'm basically attacking something that they just wrote — they would instantly read the screen. It was risky because sometimes Harvey would hallucinate and then it would just be over. But the times that they got it right, it was over." (Weinberg, Long Strange Trip, January 2026.) The wedge was deliberately narrow in another dimension: Harvey sold individual lawyers first, not law firms. "We had less friction actually in the beginning because we weren't pitching to firms. We were pitching to lawyers — individual lawyers. And so their pain was: I don't want to do this particular piece of my job. Like, I don't want to go through tens of thousands of documents and do this massive closing checklist." This B2C2B motion gave Harvey production usage and real feedback before the enterprise procurement cycle began. Platform expansion came only after the wedge was deeply proven. Product arc: 2022–2023 as a document assistant for individuals → 2024 as a workflow platform (Vault, Knowledge, Workflows) → 2025 as an agentic system with 25,000+ custom agents and 400,000+ daily agentic queries. Revenue composition shifted from 96% law firms (early 2025) to 58% law firms / 42% Fortune 100 corporates (Q4 2025) — the law firm relationships actively seeded the corporate expansion. The platform came after the wedge, not before.

Sierra

L3

Sierra's wedge was customer service containment: deploying an AI agent that could handle inbound customer contacts end-to-end — answering questions, processing requests, escalating only what genuinely required a human — at a fraction of the $13/contact cost of human-staffed call centers. The value was not marginal improvement. It was an order-of-magnitude cost reduction: $13/contact → under $1/contact, with containment rates of 70–90% for sophisticated implementers. This made the ROI case visible in days, not quarters. Proof was structured deliberately through a paid design partner program (November 2023 – February 2024) with six enterprises, all paying 10–20% of total contract value upfront. WeightWatchers achieved 70% containment and 4.5/5 CSAT in the first week of operation. SiriusXM used their design partnership to build Harmony, an agent managing subscription retention. The 100% conversion rate from design partner to paying customer was not accidental — upfront payment pre-selected motivated buyers, and the co-build process produced a product shaped specifically to each partner's problems. Program head Logan Randolph: "We told partners upfront: 'We'll give you dedicated engineers, direct access to our founders, and our cell phone numbers. But in return, we need real investment from you — payment, access to your systems, and weekly meetings to get candid feedback.'" The wedge was narrow in an additional structural way: Sierra explicitly targeted companies with acute CX operational strain, not "AI tourists." The screening mechanism was the 10–20% upfront payment — it forced prospects to demonstrate budget authority and urgency before any product was built. Sierra screened out the exploratory tier entirely. Platform expansion moved outward from chat containment to voice (launched October 2024; voice overtook text as primary channel by September 2025 — eleven months post-launch), then to email, WhatsApp, and omnichannel. Within accounts, the motion deepened from containment into adjacent CX workflows: subscription retention, loan conversion (Rocket Mortgage: 4x faster), NPS improvement (SoFi: +33 points). Voice was not a founding feature — SiriusXM demanded it during the design partner phase, and it became the second design partnership. The platform emerged from customer pull, not from a founding vision.

Decagon

L3

Decagon's wedge was the precise use case where enterprise AI could produce ROI that was impossible to dispute: customer support ticket resolution at scale. Not a chat assistant overlay, not a FAQ bot — a system that could execute multi-step workflows, integrate with backend APIs (Zendesk, Salesforce, Stripe, internal databases), and deflect 70–95% of inbound tickets without human intervention. The distinction from first-generation chatbots was deliberate: Decagon entered specifically by replacing failed chatbot deployments, not by competing against doing nothing. The discovery methodology was itself the first proof of the wedge. Co-founders Zhang and Sreenivas ran 100+ structured interviews before writing a line of product code, asking directly "How much would you pay for this?" rather than "Is this interesting?" The signal they found: "People were like, yes, if you can deploy this thing, I will sign a $150,000 check immediately, right? And this happened repeatedly." (Sreenivas, PMF Show, January 2026.) They tested other departments — sales, operations — and found conditional, delayed WTP. Support gave them immediate, specific, large commitment signals. Proof was structured via a fixed 4-week pilot with pricing agreed before the pilot started. Success metrics — deflection rate and CSAT — were defined at kickoff. This structure eliminated post-pilot renegotiation and created urgency for both sides. The Duolingo English Test case illustrates the speed possible: a prior vendor had worked for a full year and failed to get chat live; Decagon went live in one month and achieved 80% chat deflection immediately. Rippling deployed with 75+ custom routing tags across 12+ product lines. Notion ran a formal five-vendor RFP; Decagon won and delivered 2x deflection improvement and 34% faster resolution. Horizontal expansion across verticals came after the core support automation was proven. Voice (Voice 2.0, sub-400ms p95 latency) opened phone-based support volumes that chatbots had never competed for. Proactive agents (March 2026) expanded from reactive support automation into customer relationship initiation. Each expansion layer was built on top of the demonstrated wedge, not in parallel with it.

Abridge

L3

Abridge's wedge was ambient AI clinical documentation: a system that listened to doctor-patient conversations and automatically generated a draft clinical note, reducing the 2+ hours per day physicians spent on charting after encounters. The problem had a precise national price tag — the AMA estimated physician burnout costs U.S. healthcare $4.6 billion annually in turnover alone, with replacement cost per physician at $800K–$1.3M. This meant Abridge did not have to sell the existence of the problem. They had to demonstrate ROI faster and with higher confidence than competitors. The wedge was proven through a structured pilot-to-enterprise motion: 1–3 months, 15–160 clinicians, with quantified outcomes in burnout, time saved, and note quality. The outcomes were extraordinary and consistent: Seattle Children's achieved 79% documentation effort reduction; Sutter Health CDO Laura Wilt reported mid-March start to mid-April go-live with 100+ clinicians across all specialties; Lee Health saw 86% of clinicians doing less after-hours work. Shiv Rao's self-description is precise: "We're automating well over 91, 92 percent of the note. We're saving people two to three hours a day." (Rao, NerdMDs GenAI Series, August 2024.) UNC Health CMIO David McSwain described a physician who had written a resignation letter — and after using Abridge, chose not to submit it. These outcomes are not 20% improvements; they are life-changing for the individuals and economically decisive for the health systems. Epic naming Abridge its first "Pal" in August 2023 was the wedge-to-platform acceleration event. It reduced implementation from months to 2 weeks and made Abridge the default path for Epic-using health systems (38–42% of U.S. hospitals). Clinician count grew from ~8,000 pre-partnership to 60,000+ by late 2024 — a 7.5x increase in 18 months that is not achievable through direct sales alone. Platform expansion came only after the scribe wedge was proven at scale: outpatient notes → emergency department → inpatient → nursing → order generation (Contextual Reasoning Engine) → revenue cycle management. Rao explicitly reframed the product: "Providers are compensated for the care that they document, not the care that they deliver...we're a revenue cycle company, along with the other things we do." (Rao, No Priors ep108, April 2025.) This repositioning speaks to CFOs, not just CMOs — converting the product from a clinician welfare investment into a revenue capture tool.

Glean

L2

Glean's initial wedge was deceptively simple: enterprise search. Not an AI assistant, not an agent platform — a system that let employees find any document, Slack message, code file, or SaaS record across 100+ connected company apps, with correct per-user permission enforcement. Every knowledge worker searches for information multiple times per day. No behavior change required. The product worked on Day 1. This is the wedge that founder Arvind Jain, a 10-year Google Search veteran, chose deliberately — not because it was glamorous, but because it generated irrefutable adoption data (queries per day, daily active users) that became both the retention argument and the expansion trigger. Glean spent three years (2019–2021) in stealth validation before public launch, building 40 design partner customers from Jain's Silicon Valley network. The proof mechanism was engineered into the product: search generates clean, real-time adoption metrics. When Glean's customer success team brought data showing 80% adoption in 90 days and 5 queries/day to an executive sponsor, the expansion conversation was self-proving. The typical arc: $60K departmental pilot → "If I can't find it on Glean, it doesn't exist" user behavior → CSM triggers company-wide rollout → $300–500K+ contract within 9 months. The hidden depth of the wedge was not the search interface itself but the permissions layer underneath it — a system that enforces each user's exact access rights across all connected apps in real time. This took 3–4 years of engineering and became an insurmountable competitive moat when LLMs arrived in 2023. Jain: "Glean built comprehensive data infrastructure before leveraging LLMs — deep integrations with Salesforce, Confluence, Jira, plus governance layers and knowledge graphs. When LLMs emerged, this foundation positioned Glean to excel at RAG better than competitors." (Sequoia Training Data podcast, 2024.) While competitors scrambled to build enterprise AI from scratch after ChatGPT, Glean had 3+ years of production deployments already running. Platform expansion followed as a deliberate Act Two, articulated by Jain explicitly: Act One was enterprise search (the wedge); Act Two was AI applications and agentic reasoning built on top of Act One's infrastructure. Glean Agents launched February 2025. The $1M+ contract segment grew 3x in one fiscal year. Company-wide deployments doubled year-over- year. The category label officially became "Work AI" in September 2024 — after, not before, the wedge was scaled to $100M ARR.

Moveworks

L2

Moveworks chose the IT helpdesk wedge not by accident but through deliberate strategic selection. IT ticket resolution met every criterion simultaneously: universal pain (every enterprise employee hits IT friction daily), exact measurability (ticket volume, mean time to resolution, and deflection rate are all automatically logged), a clear budget owner (CIO/IT Director), and an objectively terrible status quo (portal-based ticketing that employees universally hated). The pre-existing data — 250M+ historical IT issues across the market — meant Moveworks could train before any single customer signed on, arriving with 99% entity recognition from Day 1. What distinguishes Moveworks from every other company in this study is the discipline of their pre-launch period: 3 full years in stealth (2016–2019) building proof before going to market. When they announced their Series A in April 2019, they already had paying enterprise customers with documented autonomous resolution rates of 25–40%. They did not launch to prove the concept — they launched to scale a proven one. The synthesis memo captures this precisely: "Most AI companies launch to prove the concept. This is the single most important factor." CVS achieved 50% reduction in support chats within one month. Nutanix was live in 7 weeks. These timelines were not exceptional — they were engineered into the product. The commercial structure reinforced the wedge clarity. Moveworks charged a flat per- employee fee — never per-ticket — which aligned incentives (Moveworks earns more as automation improves) and made the Forrester TEI math unambiguous: 256% ROI, payback under 1 year for a 30,000-employee organization. Sub-1-year payback made the financial case trivially easy to approve in enterprise procurement. The CEO stated the discipline explicitly: "If you can't deliver value over and over again, procurement comes to you after a year and says you didn't deliver." Expansion from IT into HR, Finance, and Facilities came only after 90%+ whole-org IT deployment rates were established — a signal that the wedge was fully captured before horizontal expansion began. Creator Studio (2023) then opened arbitrary use cases. The acquisition by ServiceNow for $2.85B in December 2025 — at ~20x ARR — was framed explicitly as a strategic purchase of an AI asset that ServiceNow could not build internally. This is the ultimate validation of wedge clarity: the incumbent bought the winner because the wedge had become an infrastructure layer they needed.

Hebbia

L3

Hebbia's initial wedge was private equity due diligence document analysis — and not as a broad description. The specific task was: running parallel analytical queries across data rooms containing thousands of files per deal, surfacing risks, covenants, and material facts that junior analysts previously spent 80-hour weeks reviewing manually. Founder George Sivulka chose this wedge because it had properties that no other enterprise AI use case could match simultaneously: extreme document volume per transaction, extreme cost of error, extreme willingness to pay (Bloomberg Terminal at $10K/seat was already a budgeted line item), and a buyer community so concentrated that winning 9 of the 10 largest US PE megafunds in year one represented near-complete top-tier market penetration. The proof threshold was qualitatively different from the other companies in this study. Sivulka's framing: "Finance is the slowest moving, most lethargic Leviathan. It's the worst possible customer base to go after unless you're providing outsized alpha or real value, in which case, the minute that there's something real, finance moves faster than any other industry." (20VC, January 2025.) The value Hebbia delivered was labor replacement, not productivity assistance — tasks that required 2–3 hours now completed in 2–3 minutes. Oak Hill Advisors (T. Rowe Price umbrella, $108B AUM) published a 6x ROI figure with 75% reduction in review times. Permira's CTO called it "by far the most advanced tool on the market that we've seen." Provident Healthcare Partners: "It has allowed us to expand capacity without adding headcount." The SVB crisis in March 2023 produced the most compressed proof event in this study: Hebbia helped PE clients map their entire portfolio's banking exposure across thousands of documents within hours. In a trust-scarce, NDA-heavy community, this event spread internally through informal peer networks faster than any marketing campaign could. ARR grew 11x in calendar year 2023 alone ($900K → $10M). Expansion was strictly sequenced. The PE due diligence wedge was proven before entering hedge funds, then investment banking, then consulting (McKinsey), then legal (Orrick), then government (US Air Force). Within accounts, the architecture was land on 3–5 Professional seats ($10K/seat) with a specific team, prove ROI, then proliferate Lite seats ($3K/seat) across the firm. NRR >200% is the result — contracts roughly double within 12 months without a new sales cycle.