Why it Matters

Impact assessment is critical for allocating capital effectively, but it's resource-intensive and inconsistent. AI could solve both problems—if investors can navigate the pitfalls.

The Big Picture

Better Society Capital (BSC), a UK impact fund-of-funds on Bluemark's top-quartile leaderboard, ran an experiment pitting AI against their deal teams on impact scoring. The results challenge assumptions about where humans and machines add value in impact investing.

The Experiment

BSC tested Perplexity Deep Research against their deal leads across 45 venture portfolio companies, scoring each on three Impact Frontiers dimensions: who benefits, depth of impact, and contribution. Both assessors worked independently, then a Venture Impact Lead reviewed discrepancies to create "final" scores.

By the Numbers

• AI matched final scores 81% of the time

• Human deal leads matched final scores 64% of the time

• AI processed 50+ sources per assessment vs. human capacity constraints

• Time savings: AI assessments took 15-20 minutes vs. 15 minutes to 2+ hours for humans

• Hallucination rate: 10-20% of AI assessments included factual errors

Where AI Excelled

Machines brought three key advantages. First, information depth—AI routinely analyzed 50+ sources and conducted 50+ tasks per assessment, impossible for time-constrained humans. Second, consistency—while human assessments varied based on familiarity, mood, and experience, AI applied frameworks uniformly. Third, accuracy—no typos or input errors that humans commonly make.

Where AI Failed

Two critical weaknesses emerged. Hallucinations plagued 10-20% of assessments, with AI confidently stating incorrect facts that required 2-3 follow-up prompts to correct. Data dependency meant early-stage companies with limited public information generated more AI errors.

The Hidden Benefits

Beyond speed and accuracy, AI delivered unexpected value. It created comprehensive documentation**—detailed rationales with references instead of human teams' typical 1-3 sentence explanations. More importantly, AI exposed framework gaps. When assessing affordability, AI sometimes used absolute measures (£10/month is unaffordable at X income) and sometimes relative ones (cheaper than alternatives = affordable), revealing BSC hadn't clearly defined their criteria.

What's Next

BSC is redesigning their process around AI-first assessment. Deal leads would receive pre-generated AI analyses for review rather than starting from scratch—dramatically reducing admin while enabling deeper strategic focus. They're also experimenting with additional data sources like LinkedIn, reviews, and proprietary databases.

Go Deeper

• Impact Frontiers' five dimensions framework

• Bluemark's Impact Management verification methodology

• Operating Principles of Impact Management best practices

• Stanford Social Innovation Review's AI and impact measurement resources