Analysis
How AI return is measured, and where published studies disagree.
Each piece reads published research against other published research and sets out where methods, definitions, samples, or timeframes differ. Every piece lists its sources with publication dates.
Published
Piece
Sources
Aug 20, 2026
Why the same ROI number means opposite things
IDC's $3.70 and MIT's 95% failure rate are not contradictory. They are measuring different cohorts, on different timelines, answering different questions.
2 sources cited
Aug 5, 2026
Why most pilots stall before P&L impact
Why AI ROI is so hard to prove to the board: Deloitte's lengthening payback periods and MIT's genAI divide describe the same gap from two directions — the distance between a working pilot and a number the P&L actually notices.
2 sources cited
Jul 22, 2026
What finance teams should count as AI return
IBM's KPI framing and Wharton's productivity metrics point at the same fix: define the unit of return before the pilot starts, not after.
2 sources cited