Skip to main content

​
How it works

The AI efficiency index is a composite metric that combines four AI-related metrics into a single 0-100 score. It helps you understand how effectively your team is leveraging AI tools by balancing output, adoption, cost, and code quality. A higher score indicates better overall AI tool usage - high output and adoption, reasonable costs, and low code churn.

​
Calculation

The AI efficiency index uses percentile-based normalization to combine metrics with different scales. Each component is converted to a percentile rank (0-100) based on global benchmarks, then combined with weighted scoring. Formula:

​
Component metrics

These are the default weights.

​
Percentile normalization

Since these metrics have different scales (percentages, dollars, lines of code), we convert each to a percentile rank before combining them:
  1. Your value is compared against global benchmark data from all organizations
  2. A percentile rank (0-100) indicates where you fall in that distribution
  3. For “lower is better” metrics (cost and turnover), the percentile is inverted so that higher always means better
Example: If your AI cost is in the 20th percentile (low cost = good), it becomes the 80th percentile rank in the efficiency calculation.

​
Technical details

AI output volume
  • Source: Code output (PR estimates) from merged PRs that is attributed to AI tools
  • Normalized per day for fair comparison across time periods
AI usage percentage
  • Formula: AI code output / Total code output
  • Displayed as 0-100%
AI cost per user
  • Source: Aggregated spending from connected AI tools (Claude Code, Cursor, GitHub Copilot, etc.)
  • Normalized per month
Code turnover rate
  • Measures how much code is changed or deleted within 90 days of being merged
  • See code turnover for details
All four component metrics must be present to calculate the efficiency index. If any component is missing for a given time period, that data point is skipped.This ensures the score is always based on complete data rather than partial calculations.
The AI efficiency index pulls data from:
  • GitHub, GitLab, and Bitbucket for code output and turnover
  • Connected AI tools for usage and cost data (Claude Code, Cursor, GitHub Copilot, Windsurf, etc.)
Last modified on October 17, 2026