R² = 0.11 Between ARR Scale and Revenue Multiple: A regression of revenue multiple versus ARR scale in public markets shows little correlation, with an R-squared of 0.11. Even when including outliers like Adobe and Salesforce, the correlation drops to 0.02, suggesting no meaningful “size premium.”

3 Toxic Consequences of ARR Obsession: Dan Gray outlines three outcomes of overemphasizing ARR: companies stay private longer, post-IPO performance declines, and capital flows toward rapid scalability over genuine innovation. This dynamic limits public investor participation and concentrates gains in private markets.

Billions in ARR Reflect Selection Bias, Not Causation: Larger companies often show stronger metrics across dimensions, but the post attributes this to selection bias rather than causation. Businesses with durable growth and strong unit economics naturally “grow out” of smaller buckets, while weaker companies remain stuck, creating the illusion that scale drives quality.

1,057 Funds Analyzed, Fund I Leads at Median and Top 5%: The analysis covers 1,057 North American VC funds from 2000 to 2020 vintages. Fund I vehicles outperform not only at the median but also across the top quartile, top decile, and 95th percentile when measured by TVPI, surpassing Funds II through V.

Fund I 2x to 4x Smaller Than Later Funds: Median Fund I size is 2x smaller than Fund II, 3x smaller than Funds III and IV, and 4x smaller than Fund V. Smaller fund sizes reduce reliance on extreme outliers and allow ownership discipline, concentrated bets, and capital-efficient deployment to drive returns.

Low DPI vs NAV, Capital Concentration in 2025: Although 2025 showed healthy absolute exit dollars, DPI as a percentage of NAV remains near historic lows. With LP NAV exposure close to all-time highs, capital has flowed disproportionately to established managers, even as first-time funds show stronger benchmark performance.

2/3 of Costs Are People, 3-Part WHO Scorecard: Butler notes that roughly two-thirds of a software company’s expenses are people, yet hiring is often reactive. He highlights the WHO model, built around a three-part scorecard: a one-sentence role vision, core competencies, and 6 to 12 month success criteria. He also emphasizes hiring A players from the top 10 percent who thrive with autonomy rather than micromanagement.

30 Customer Interviews, 50% Beta Threshold: On product-market fit, Butler argues founders should speak with at least 30 potential customers before building. If fewer than 50 percent convert into beta agreements, development should not proceed. He also points to the early adopter ceiling at roughly 7 percent of the market, warning that traction there does not equal mainstream demand.

40% Revenue on S&M, 40% Future Decline: Drawing from experience at Concur, where over 40 percent of revenue went to sales and marketing, the discussion stresses that execution beats product superiority. Butler expects AI-driven account-based marketing to reduce sales and marketing spend by up to 40 percent while increasing personalization. He also frames culture as the fourth pillar, arguing that reward systems and leadership behavior determine whether values are real or decorative.

10,562 Startups Analyzed, ~50% Reach Series B in Strong Cohorts: Across the dataset, roughly half of Series A companies in strong market cohorts eventually raise a Series B. The remainder either get acquired, operate without additional capital, or shut down after extended runway pressure.

2020 Cohort: 20% in 1 Year, 40% in 2 Years: Startups that raised in 2020 benefited from favorable market conditions. About 20% reached Series B within one year and roughly 40% within two years, reflecting abundant capital and faster fundraising cycles.

2022 vs 2024: 8% vs 11% in Year 1: The 2022 cohort faced tougher conditions, with only 8% advancing to Series B within one year and about 10 to 12% within two years. By contrast, 2024 shows early signs of stabilization, with 11% reaching Series B within a year and projections pointing toward roughly 20% by year two.

$700K ARR per Employee, $500K+ New Bar: Top AI-native startups are reportedly reaching about $700K ARR per employee, far above historical SaaS norms. Lemkin suggests investors now look for a path to at least $500K ARR per employee, reflecting structurally leaner teams and AI-driven leverage.

Major Releases Every 60 Days vs Quarterly Cycles: AI-native companies are shipping major agent updates roughly every 60 days, while many traditional vendors remain on quarterly release cycles. The post argues that slower cadence limits competitiveness in a market where iteration speed compounds advantage.

2x-10x AI Upsells, 2-3 Day Hybrid vs 5 Day RTO: Traditional vendors attempting 2x to 10x AI upsells face resistance when value is incremental. At the same time, Lemkin observes many incumbents operating on 2 to 3 day hybrid schedules, while AI-native teams often choose full 5 day in-office collaboration, reflecting higher urgency and execution intensity.

6/6 Developers Train on Chat Data by Default: After Anthropic switched from opt-in to opt-out in September 2025, all six frontier developers now train their AI models on user chat inputs and outputs by default. Enterprise users are opted out, creating a two-tiered privacy system where businesses receive greater data protections than the hundreds of millions of individual consumers.

700M ChatGPT Users Affected by Default Training: With OpenAI's ChatGPT alone reaching 700 million users as of August 2025, the scale of default data collection for model training is staggering. Most users never change defaults, and OpenAI frames its opt-out as a social benefit—"Improve the model for everyone"—a framing researchers characterize as guilt-based persuasion to align user behavior with company interests.

28 Policy Documents Analyzed, Massive Fragmentation: Across the six developers, the researchers coded 28 separate documents including privacy policies, sub-policies, and FAQs. OpenAI alone required six documents to capture its full data practices, with material facts disclosed not in the main privacy policy but scattered across branching sub-policies, making it practically impossible for consumers to understand how their data is used.