How this research was conducted — sources, evidence tiers, and analytical framework
This page explains how the research was conducted, what types of evidence were collected, and how claims were rated for causal confidence. Understanding the methodology is necessary for correctly interpreting the growth laws and company profiles.
This benchmark covers 13–16 AI and AI-adjacent companies that achieved exceptional sales-led growth between 2018 and 2026. The research was conducted in three structured phases over several months, with a third-pass upgrade (V3) applying causal-confidence rating to all major claims. The goal throughout was to distinguish what actually caused hypergrowth from what founders describe as the cause — two different things. V3 added explicit evidence-tier rating, mechanism classification, and counterfactual analysis to separate foundational conditions from operational accelerators from compounding mechanisms. A second synthesis layer — Six Sales Laws — was derived in April 2026 by cross-referencing the same primary sources for tactical commercial patterns: how the first deals were run, how demos were structured, how pilots were paid and structured, and how individual practitioners pulled enterprise procurement. The Sales Laws describe the commercial motion; the Growth Laws describe the structural scaling logic. Together they form a complete picture of how this cohort sold and scaled.
Primary source collection across all 13–16 benchmark companies. The aim was to get as close as possible to direct verbatim evidence — founder statements in their own words, operator confirmations from VP-level executives, and customer evidence from named case studies and public review platforms.
Podcast and video interviews were downloaded using yt-dlp and transcribed using OpenAI Whisper (base.en model, CPU inference). Transcriptions were stored as raw text files and then extracted into structured source files with explicit VERBATIM and PARAPHRASE/INFERENCE labels. Audio extraction was prioritized because it captures the unedited, unfiltered voice of the speaker — no editorial rewrites, no PR cleanup, no selective quotation. Founders and operators say things in long-form podcasts that they would never put in a press release. Sources extracted via audio: 20+ podcast episodes across Decagon, Legora, Harvey, Abridge, Gong, Wiz, Sierra, Ramp, Glean, Moveworks, Hebbia, Writer, and Intercom/Fin. Notable examples include the Escape Velocity Ep. 04 (Jesse Zhang, Decagon), 20VC Jan 2026 (Max Junestrand, Legora), No Priors (Bret Taylor, Sierra), and WMIF 2024 (Shiv Rao, Abridge).
Note: Whisper base.en may mishear proper nouns, company names, and specialized terminology. All transcriptions were reviewed for accuracy. Named figures, statistics, and quoted verbatims were cross-referenced with other available sources where possible.
Company websites, blog posts, investor announcements, G2 reviews, analyst reports (Sacra, The Information, Forbes), LinkedIn posts, and trade press interviews were collected and indexed. These sources are reliable for named facts (funding rounds, ARR milestones, customer names) but carry higher narrative-bias risk than audio transcriptions for qualitative claims about strategy and causation.
Statements from non-founder executives — VP Sales, CRO, Head of CS, RevOps — are treated separately from founder statements. Operators confirm or contradict founder narratives based on their daily operational reality rather than investor storytelling incentives. Where operator-confirmed evidence was found for a claim, it upgraded the claim from L1 to L2 in the evidence tier system. Notable operator sources extracted: Evan Cassidy (Decagon VP Sales), Patrick Forquer (Legora SVP Revenue), John Haddock (Harvey CBO — 2 blog posts + trade press), Reggie Marable (Sierra Head of Global Sales — audio extraction), Trish Cagliostro (Wiz VP Channels — audio extraction), Chris Orlob and Udi Ledergor (Gong GTM executives).
Named customer verbatims, KLAS reports, G2 reviews, and case studies with specific metric attributions. Customer evidence is the strongest confirmation of value claims — a customer who publicly states "we saved 8 hours per lawyer per week using Harvey" is providing third-party causal confirmation of an ROI claim. Notable customer sources: KLAS reports for Abridge, named law firm case studies for Harvey and Legora (LPHS, Vinge, Deutsche Telekom), Decagon customer ROI documentation (Bilt, Rippling), Wiz enterprise customer references.
Every major claim in the synthesis files and growth laws is rated against this five-tier evidence hierarchy. Claims at L1 carry narrative bias risk; claims at L4–L5 have the highest causal confidence.
Claim comes from the founding team (CEO, co-founder) speaking in their own interest — fundraising context, press interviews, investor updates. High narrative-bias risk. Founders have incentives to tell a coherent, positive story about what caused their success. L1 claims are treated as hypotheses, not conclusions.
Claim is confirmed or elaborated by a non-founder executive (VP Sales, CRO, Head of CS, RevOps) who operates the described process daily. Operators have lower narrative-incentive than founders and more direct operational knowledge. L2 substantially reduces narrative-bias risk.
Claim is confirmed by a named customer who has independently experienced the described outcome. Customer ROI claims with specific metrics and attribution are the strongest form of value evidence. Named customer ≠ anonymous review.
Claim includes explicit documentation of what happened (or what the founder predicted would happen) without the described intervention. Counterfactual evidence is the closest to causal proof available in retrospective business research.
The same pattern appears independently in 3 or more companies with no coordination between them. Multi-company convergence is the strongest causal signal available across a benchmark set — it separates universal structural patterns from company-specific or founder-specific choices.
Claims are also classified by mechanism type — whether they describe a pre-condition that had to exist before growth started (F), an accelerator that sped up growth once started (A), or a compounding mechanism that built structural defensibility over time (C). This classification prevents mistaking accelerators for preconditions, which is the most common error in applying playbooks from one company to another.
A pre-condition that had to exist before the growth playbook could work. Foundational conditions are not strategies — they are gates. Missing one is a ceiling on growth rate, not a missing lever. Most foundational conditions are non-replicable at will (founding team domain access, investor network, AI capability threshold crossing, exogenous market catalyst).
An operational choice or motion that sped up growth once the foundational conditions were in place. Accelerators are the "playbook" — the replicable strategies. The highest-confidence accelerators appear in 10+ of 13 companies and have L2+ evidence: paid pilot structure, WTP discovery filter, founder-led sales phase, non-black-box product design.
A structural property that became harder to displace over time as the company used it. Compounding mechanisms are the source of durable competitive advantage. They include: customer reference networks (every new logo makes the next easier), AOP/workflow libraries (accumulated product configuration per customer), domain-expert CS teams (relationships that competitors cannot buy), and category-defining content flywheels.
Evidence coverage by company and evidence zone — reflects the actual depth of the research corpus, not a quality rating of the company. Each cell rates whether strong, partial, or no primary evidence was found for that zone. Use this to calibrate which company profiles are load-bearing vs. which should be treated as illustrative only.
Hover over column headers for zone definitions.
| Company | Founder Voice | Operator / Sales | Customer Evidence | Pricing & Packaging | GTM Motion | Deployment / Impl | Outcome / ROI | 2024–2026 |
|---|---|---|---|---|---|---|---|---|
| Harvey | ||||||||
| Abridge | ||||||||
| Gong | ||||||||
| Decagon | ||||||||
| Legora | ||||||||
| Intercom / Fin | ||||||||
| Ramp | ||||||||
| Wiz | ||||||||
| Deel | ||||||||
| Sierra | ||||||||
| Glean | ||||||||
| Moveworks | ||||||||
| Hebbia | ||||||||
| Writer | ||||||||
| Incident.io | ||||||||
| Listen Labs | ||||||||
| Cognition AI (Devin) |