CTO Toolkit

AI Engineering Scorecard

Four measurement categories for engineering organisations where humans and agents build software together.

How to use this

Review quarterly alongside the Engineering Health Scorecard. DORA remains the spine of Flow and Quality; the last two categories exist because an organisation can improve DORA numbers while spending more to get less.

Flow — where does work wait?

Signal Score (1–5) Evidence
Lead time for changes ____ ____________________
Cycle time from ready to merged ____ ____________________
Review time and review queue depth ____ ____________________
Change size distribution ____ ____________________
Time spent waiting on decisions rather than implementation ____ ____________________

Quality — what does speed cost?

Signal Score (1–5) Evidence
Change failure rate ____ ____________________
Escaped defects reaching customers ____ ____________________
Rework rate on merged changes ____ ____________________
Security findings by severity and age ____ ____________________
Rollback rate ____ ____________________

AI effectiveness — is delegation working?

Signal Score (1–5) Evidence
Agent task completion rate without human takeover ____ ____________________
Human intervention rate per delegated task ____ ____________________
Retry rate before an acceptable result ____ ____________________
Acceptance without rework ____ ____________________
Defects attributable to agent-generated changes ____ ____________________
Verification failure rate on agent output ____ ____________________

Economics — did this create value?

Signal Score (1–5) Evidence
Cost per accepted change ____ ____________________
Inference and tooling spend against delivery outcomes ____ ____________________
Engineering hours reclaimed for architecture and difficult problems ____ ____________________
Rework avoided ____ ____________________
Time to value for the customer ____ ____________________

Anti-patterns

  • Celebrating volume of generated code as an outcome
  • Reporting adoption metrics — seats, prompts, tool usage — as engineering performance
  • Measuring individuals rather than the system
  • Improving Flow while Quality and Economics quietly degrade

This is an original, generic framework for reuse. It is not proprietary employer material.

← All toolkit resources

Related: Engineering Playbook