Methodology

How the Autonomy Index measures workflow-level AI adoption.

The methodology moves from scoping one workflow to evidence, scoring, diagnosis, recommendations and reassessment. Evidence comes from the workflow itself, not from activity dashboards.

Written by Christine Barnett, founder of the Autonomy Index™. Last reviewed .

What the methodology measures

The methodology measures whether AI has changed a defined workflow, how much work it reliably completes, where human intervention remains, whether output quality holds and whether governance and business evidence support the current design. It does not treat usage volume or an autonomy score as proof of value.

The process at a glance

  1. 01

    Scope workflow

    One clearly bounded business workflow becomes the unit of assessment.

  2. 02

    Collect evidence

    Leadership input, frontline behavioural evidence and operational evidence.

  3. 03

    Score eight dimensions

    Four weighted pillars, eight formal dimensions, one 0 to 100 score.

  4. 04

    Diagnose constraints

    Primary constraint, Theatre Gap, AI Debt and Quality Matrix placement.

  5. 05

    Redesign

    Prioritised recommendations, ownership, controls and evidence discipline.

  6. 06

    Reassess

    Re-score after implementation to show movement, not activity.

1. Scope the workflow

Select one clearly bounded business workflow. It is the unit of assessment. Tools, licences and platforms are context, not the object of measurement.

2. Map the workflow

Map tasks, decisions, systems, handovers, owners and exception paths end to end, including what happens when something goes wrong.

3. Collect evidence

  • Leadership evidence on intent, investment, objectives and stated outcomes.
  • Frontline behavioural evidence on how the work is actually completed day to day.
  • Operational evidence: artefacts, sample outputs, reviewer feedback, exception rates, cycle-time change.

4. Evaluate the workflow measures

Adoption consistency, feature depth, workflow ownership, human intervention and output quality are evaluated as workflow measures. They are evidence inputs into scoring, and they are reported back to leaders. They are not the formal dimensions.

5. Score four pillars and eight dimensions

Scoring uses the four weighted pillars and eight formal dimensions set out on The Index: Leadership & Strategy at 20%, Workflow & Operations at 30%, Behaviour & Trust at 30% and Governance & Risk at 20%.

20% weighting

Leadership & Strategy

  • Leadership Alignment: Does the stated AI narrative match operational reality?
  • Strategic Integration: Is AI connected to objectives, budget and decision making?

30% weighting

Workflow & Operations

  • Workflow Integration: Is AI embedded in the workflow or bolted alongside it?
  • Operational Dependency: Would the workflow degrade if the AI were removed?

30% weighting

Behaviour & Trust

  • Behavioural Adoption: Do people use it by default or only when prompted?
  • Operational Trust & Override: How often is output overridden, and on what basis?

20% weighting

Governance & Risk

  • Governance Maturity: Who owns the output, and how is it controlled?
  • Risk & Output Assurance: How are errors detected, escalated and corrected?

6. Assign the score and Autonomy Maturity Band

Dimension scores roll up into a single 0 to 100 Autonomy Index score, which places the workflow in one of five Autonomy Maturity Bands.

  • 0 to 20: Level 1, AI Theatre, visible AI activity with no measurable change to how the workflow runs.
  • 21 to 40: Level 2, Fragmented Pilots, real capability exists in pockets, while the surrounding workflow stays fragmented.
  • 41 to 60: Level 3, Functional Adoption, aI is used consistently inside defined steps of the workflow with reliable output.
  • 61 to 80: Level 4, Operational Integration, the workflow is redesigned around AI, with ownership, controls and evidence in place.
  • 81 to 100: Level 5, Operational Autonomy, aI reliably runs the workflow end to end under governance, monitoring and escalation.

7. Assign the Workflow Ownership Stage

The five-level Autonomy Maturity Band describes the overall maturity of the assessed workflow or organisation. The four-stage Workflow Ownership scale describes how much of a specific workflow AI reliably completes. They are different outputs and are always labelled separately.

  • Ownership Stage 1, Individual Assistance, aI helps a person complete a task. The workflow itself is unchanged.
  • Ownership Stage 2, Repeatable AI-Assisted Tasks, specific steps are consistently completed with AI inside the workflow.
  • Ownership Stage 3, Supervised Workflow Ownership, aI runs the workflow end to end with human review at defined checkpoints.
  • Ownership Stage 4, Trusted End-to-End Autonomy, aI owns the workflow reliably, with governance, monitoring and escalation.

8. Diagnose

Identify the primary constraint, the Theatre Gap between stated and operational adoption, AI debt, and placement on the Autonomy Quality Matrix™.

9. Produce prioritised recommendations

Each assessment produces a prioritised set of actions: scale, redesign, supervise or stop. Recommendations are workflow-specific and tied to observed evidence.

10. Reassess

Reassess after implementation, after material workflow change, or at a defined cadence, typically two to four times per year. Trends matter more than single scores.

Scoring principles

  • Score workflows, not tools.
  • Reward reliable ownership, not activity.
  • Penalise autonomy that outpaces output quality.
  • Surface AI debt where investment does not move ownership forward.

Detailed weightings within each dimension remain proprietary. The principles, pillars, dimensions, bands and outputs above govern how any score is produced.

How to interpret the results

Interpret the score, maturity band and Workflow Ownership Stage together with output quality, human intervention and the underlying evidence. More autonomy is not automatically better. The appropriate target depends on the workflow, the consequence of error, the need for human judgement and the strength of governance and escalation.

Trends after a comparable reassessment are more useful than a single score. A change in scope, baseline or evidence quality can affect comparability and should be documented.

Evidence standard

Claims about capacity released, cycle time, quality, cost or business impact require a relevant baseline and evidence that can be traced to the workflow. Where override, exception or rework data is not recorded, the result identifies a measurement gap rather than assuming a rate. Self-reported evidence should be checked against operational artefacts wherever possible.

Limitations

The methodology depends on the quality of workflow evidence. It does not evaluate AI models, and it is not a substitute for governance, security or compliance processes. It is a measurement framework, not a certification.

Examples and related evidence

Apply the methodology to one workflow

Use the existing assessment to identify the workflow's maturity, ownership and priority constraints.