Back to projects
Coverflex · 2026

AI Agent ROI Evaluation Framework

AI Skill Evaluator that measures the ROI of internal AI skills through labor time saved — IBM's first Hard ROI KPI — combining measured token cost, reliability, and risk into a single Keep / Optimize / Kill verdict.

How it works
Step 1 of 6Locate agent

Locate agent

Run the skill once end-to-end to capture its real token usage and wall-clock duration as the ground truth for every downstream calculation.

Highlights

  • End-to-end flow: run the skill once, collect user inputs, let the model compute score and ROI, output verdict.
  • Composite score weights Reliability 50%, Time Savings 30%, Risk 20% → Keep (>7), Optimize (4.5–7), Kill (<4.5).
ClaudeAI AgentsEvaluationPython
View framework

Want the deeper walk-through?

Happy to share dashboards, code and case decks on request.