Provenance Manifesto
arrow_back Back to Blog

Article artifact

AI Adoption Maturity Model (SDLC-Oriented)

Source: artifacts/ai-adoption-maturity-model.md

Table of contents

expand_more

AI Adoption Maturity Model (SDLC-Oriented)

Overview

This model defines AI adoption across four dimensions:

  • Coverage (SDLC stages touched)
  • Autonomy (AI vs human execution)
  • Trust (quality and predictability)
  • Governance (traceability and control)

Stage 0 — Baseline (Pre-AI)

Purpose

Establish a control group for comparison.

Metrics

  • Lead Time = Time from ticket creation → production
  • Cycle Time = Time from work start → completion
  • Defect Rate = Defects / tickets
  • MTTR = Mean time to resolve incidents

Example

  • Lead Time: 24 days
  • Cycle Time: 10 days
  • Defect Rate: 0.3 per ticket

Stage 1 — Non-Degrading Augmentation

Goal

Increase AI usage without degrading delivery quality.

Metrics

AI Code Contribution %

AI_Code_% = (AI-generated LOC / Total LOC) * 100

Quality Parity Index

QPI = (Defect rate AI code) / (Defect rate human code)

Target: ~1.0 or lower

Review Rejection Rate

Rejection Rate = Rejected PRs / Total PRs (AI vs Human)

Example

  • AI Code: 45%
  • QPI: 0.95 (AI slightly better)
  • Rejection Rate: 12% AI vs 10% human

Stage 2 — Assisted Execution

Goal

AI contributes to execution tasks with human oversight.

Metrics

AI-Assisted Ticket %

AI_Ticket_% = (Tickets with AI assistance / Total tickets) * 100

Human Touchpoints per Ticket

Touches = Number of manual interventions per ticket

Time-to-Merge Reduction

TTM Reduction = (Baseline TTM - Current TTM) / Baseline TTM

Example

  • 60% tickets AI-assisted
  • Avg touches: 3 per ticket
  • TTM reduced by 20%

Stage 3 — Multi-Step Agentic Workflows

Goal

AI executes chained workflows across SDLC steps.

Metrics

Workflow Automation Depth

Depth = Number of SDLC steps executed by AI per ticket

Failure Rate

Failure Rate = Failed workflows / Total workflows

Override Frequency

Overrides = Human interventions / workflows

Example

  • Avg depth: 4 steps (code + test + PR + deploy draft)
  • Failure rate: 8%
  • Override: 25%

Stage 4 — Full SDLC Coverage

Goal

AI is present across all SDLC stages.

Coverage Areas

  • Business Analysis
  • Requirements
  • Architecture
  • Development
  • Testing
  • Release
  • Support artifacts

Metrics

Artifact Coverage %

Coverage = (AI-assisted artifacts / Total artifacts) * 100

Consistency Score

Consistency = Alignment between requirements, code, and tests (manual or automated scoring)

Example

  • Coverage: 70%
  • Consistency: High (traceable links between artifacts)

Stage 5 — Autonomous Delivery

Goal

AI executes end-to-end delivery with human oversight.

Metrics

Fully AI-Executed Tickets %

AI_E2E_% = (Tickets completed end-to-end by AI / Total tickets) * 100

Exception Rate

Exception Rate = Tickets requiring human intervention / AI tickets

Rollback Rate

Rollback Rate = Rollbacks / deployments

Example

  • 30% fully AI-executed
  • Exception rate: 15%
  • Rollback rate: 3%

Stage 6 — Governed Intelligence (Provenance-Aware)

Goal

All decisions are traceable and explainable.

Metrics

Decision Coverage %

Decision Coverage = Decisions with full metadata / Total decisions

Decision Reuse Rate

Reuse Rate = Reused decisions / Total decisions

Conflict Detection Rate

Conflict Rate = Conflicting decisions detected / Total decisions

Example

  • 80% decisions tracked
  • 25% reused
  • Conflicts detected early in 10% cases

Stage 7 — Self-Optimizing System

Goal

System improves itself using feedback loops.

Metrics

Decision Success Rate

Success Rate = Successful decisions / Total decisions

Skill Efficiency Score

Efficiency = Outcome quality / Cost (tokens + time)

Drift Detection

Measure deviation in model outputs over time

Example

  • Success rate improves from 70% → 85%
  • Efficiency increases 30% over quarter

Summary

Stages: 0. Baseline

  1. Safe Augmentation
  2. Assisted Execution
  3. Agentic Workflows
  4. Full SDLC Coverage
  5. Autonomous Delivery
  6. Governed Intelligence
  7. Self-Optimizing System