What could reasonably be done?
Built from current capabilities, business tasks, firm context, and practical constraints.
Methodology
Two independent engines. One honest comparison.
The index separates practical AI potential from observed adoption. Actual behavior cannot inflate or suppress the estimate of what is possible.
What is credibly observed?
Built from external adoption evidence, normalization, evidence grades, and uncertainty.
Current AI capability
Stable capability identities receive dated assessments of maturity, affordability, reliability, integration difficulty, oversight, and risk.
Business tasks
Capabilities are mapped to concrete work rather than broad claims about replacing jobs or industries.
Practicality
Transparent weights and safety caps convert capability-task mappings into a reproducible task score.
Industry baseline / Possible
Task scores roll into business functions and then a representative industry archetype.
Actual adoption evidence
Government surveys, credible studies, and other observations are normalized without entering the Possible calculation.
AI Utilization Gap
A gap is calculated only when Possible and Actual measure compatible populations, periods, constructs, and scales.
Evidence confidence
Grades and confidence communicate directness, source quality, uncertainty, and limitations without false precision.
Versioning over time
Capabilities, evidence, methods, and scores are timestamped so the index can change transparently as AI changes.
Non-negotiable rules
Missing is a finding, not a zero.
- Missing adoption evidence remains missing unless a separately labeled model is introduced.
- Modeled estimates and observed values are never presented as the same thing.
- Evidence quality may vary by country, industry, population, and measured concept.
- Every score must be reproducible from dated inputs and a declared score version.
- Scores change as capabilities, costs, reliability, and evidence change.