GenAI Divide: 95% of all pilots fail [pdf] (mlq.ai)

Project NANDA’s preliminary “GenAI Divide” report (Jan–Jun 2025), based on a review of 300+ public initiatives, 52 structured interviews and 153 senior-leader surveys, finds that despite $30–40B in enterprise GenAI spending, roughly 95% of pilots produce no measurable P&L impact. Adoption is high—over 80% have explored tools like ChatGPT/Copilot and ~40% report deployments—but value is largely limited to individual productivity. Custom, enterprise-grade systems fare much worse: 60% were evaluated, 20% reached pilot, and only 5% reached production. The divide isn’t driven by model quality or regulation but by approach: most systems are brittle, don’t retain feedback, fail to adapt to context, and don’t integrate with day-to-day workflows. Technically, the report highlights a critical “learning gap”: scalable value comes from learning-capable systems that integrate feedback loops, process-specific customization, and tight workflow integration. Organizations that succeed prioritize business-outcome metrics, expect systems to improve over time, and often leverage external partners (who have ~2x the success rate of internal builds). Sector disruption is uneven—Tech and Media show clear structural change while seven other sectors show minimal impact. Early measurable gains include reduced BPO/agency spend and selective workforce effects in support, engineering and admin, suggesting targeted, adaptive GenAI can deliver real ROI when designed around context and continuous learning.