Worked example
Demo AI Adoption Audit
A worked example of an Autonomy Index audit for an 850-person higher education group operating across the UK, Ireland and Australia. Illustrative, not a real client engagement.
Northstar Learning Group
- Sector
- Higher education and professional learning
- Size
- 850 employees
- Regions
- UK, Ireland, Australia
- AI stack in use
- Copilot, ChatGPT Team, Gemini, Notion AI, Intercom AI, Gong AI
Executive summary
AI activity is broad but shallow. Adoption sits at 68% across the five workflows in scope, but the portfolio Autonomy Index score is 2.4 out of 4. Most workflows are assisting or embedded, not owning steps. Two workflows are creating measurable value and are safe to scale. One workflow carries material risk today and should be contained within thirty days. Governance readiness is 42 out of 100: use cases, ownership and oversight are not yet documented at the level a board can rely on.
Adoption depth
68%
Weighted across 5 workflows in scope
Autonomy Index score
2.4 / 4
Portfolio average across workflows
Output quality
3 of 5
Workflows meeting quality bar at current autonomy
Governance readiness
42 / 100
Inventory, oversight, documentation, review
AI use-case inventory
Five workflows are in scope. Each has a named team, a live tool stack, an autonomy level, an adoption depth score, an output quality band, a risk band and a recommended action.
Student Services
Student support ticket triage
Intercom AI, ChatGPT Team
- Autonomy
- L3 · Owning steps
- Adoption
- 82%
- Output quality
- Acceptable
- Risk
- Elevated
Oversight: Auto-reply on Tier 1; human review on escalation only
Verdict: AI owns first response and categorisation. Escalation logic is inconsistent between UK and AU teams and sensitive cases (safeguarding, fees appeals) are not always routed to a human reviewer.
Marketing
Marketing content drafting
ChatGPT Team, Gemini
- Autonomy
- L2 · Embedded
- Adoption
- 74%
- Output quality
- Uneven
- Risk
- Moderate
Oversight: Editor review before publish; no shared prompt or brand library
Verdict: AI is embedded in first-draft workflow but every team maintains its own prompts. Output quality varies by author, not by tool. Brand and claims review happens late.
Enrolments / B2B Sales
Sales call summaries and CRM follow-up
Gong AI, Copilot
- Autonomy
- L3 · Owning steps
- Adoption
- 88%
- Output quality
- Strong
- Risk
- Low
Oversight: Rep confirms summary; auto-logged to CRM
Verdict: Highest-value workflow in the audit. AI reliably summarises calls, drafts follow-up and writes CRM notes. Reps trust the output and cycle time has dropped materially.
People, Legal, Ops
Internal policy drafting
ChatGPT Team
- Autonomy
- L2 · Embedded
- Adoption
- 41%
- Output quality
- Weak
- Risk
- High
Oversight: Ad-hoc; drafts sometimes shipped without legal review
Verdict: Individual contributors are drafting policy with AI and circulating without a documented review path. Two drafts referenced UK-only clauses in AU-facing documents. High reputational and regulatory exposure.
Exec / Strategy
Board reporting summaries
Microsoft Copilot
- Autonomy
- L2 · Embedded
- Adoption
- 55%
- Output quality
- Acceptable
- Risk
- Moderate
Oversight: Chief of Staff edits before board pack
Verdict: AI compresses source docs into board-ready narrative but source attribution is inconsistent. Numbers in the narrative do not always reconcile to the underlying data pack.
Risk and quality findings
High
Internal policy drafting
Internal policy drafting has no documented review path.
Pause unreviewed AI-drafted policy circulation. Introduce a two-step legal + regional review.
Elevated
Student support ticket triage
Safeguarding and fees appeals not consistently routed to human reviewers.
Add mandatory human handoff rules on sensitive intents. Log every override.
Moderate
Board reporting summaries
Board narrative does not reconcile to underlying data.
Redesign the workflow so Copilot summarises from a signed-off numbers pack, not raw sources.
Moderate
Marketing content drafting
No shared prompt or brand claims library for marketing.
Ship a governed prompt library with a claims and regulated-copy checklist.
Governance readiness
Northstar has the elements of a governance layer but they are not yet connected. Ownership sits with individual managers, not workflow owners. Documentation exists in Notion but is not indexed or reviewed. Oversight is real for sales and student support and effectively absent for policy drafting. Before adding new AI tools, the group needs a single inventory, named owners per workflow and a documented review cadence: the operational evidence layer, not a legal compliance product.
Board questions, answered
Which workflows has AI actually changed?
Sales call summaries (L3, Scale) and student support triage (L3, Operationalise). Everything else is L2 or below.
Where are teams only experimenting?
Internal policy drafting and board reporting. Both show individual use without a workflow owner or review path.
Which workflows are ready to scale?
Sales call summaries and CRM follow-up. Adoption 88%, quality Strong, risk Low, oversight documented.
Where is human oversight required?
Safeguarding and fees intents in student support, all policy drafting, and any board narrative that carries numbers.
Where are output quality and risk concerns highest?
Internal policy drafting (High). Student support escalation logic (Elevated). Board narrative reconciliation (Moderate).
What evidence can we show to the board?
Use-case inventory, per-workflow autonomy score, risk register with named owners, oversight log and the 90-day roadmap in this report.
Should we buy more AI tools or redesign workflows first?
Redesign first. Two of the five workflows in scope will produce more value from redesign than from any new tool purchase in the next two quarters.
Ninety-day roadmap
Days 0 to 30
Contain and inventory
- Publish the AI use-case inventory across UK, IE and AU.
- Pause unreviewed AI-drafted policy circulation. Route through People + Legal.
- Introduce mandatory human handoff for safeguarding, fees and appeals intents in Intercom.
- Assign a named owner for each of the 5 workflows in scope.
Days 31 to 60
Operationalise the working workflows
- Formalise the Gong sales summary workflow as the reference standard. Document it as a model card.
- Ship a governed marketing prompt and claims library. Retire personal prompts.
- Redesign board reporting so Copilot summarises from a signed-off numbers pack.
- Stand up a lightweight AI review cadence (monthly) with the Exec sponsor.
Days 61 to 90
Scale with evidence
- Scale Intercom triage with the new escalation rules; measure quality weekly.
- Publish a board-ready AI evidence pack: inventory, autonomy scores, risk register, oversight log.
- Decision point: redesign two more workflows before buying additional AI tools.
- Move from audit to standing quarterly Autonomy Index review.
What this demo shows
This is the shape of a real Autonomy Index audit: a use-case inventory, per-workflow autonomy scores, a risk and quality read, a governance readiness view and a prioritised ninety-day roadmap. In a real engagement, every number is grounded in interviews, tool telemetry and workflow walkthroughs, and each workflow ships with a named owner and a model-card-style record.
Related
- AI Adoption Audit
- AI Governance Readiness
- Workflow Autonomy Framework
- AI Adoption Metrics
- AI adoption FAQ
Northstar Learning Group is an illustrative example created to demonstrate the format of an Autonomy Index AI Adoption Audit. It is not a real client engagement.
Ready to measure what AI actually changed?
Take the Autonomy Index™ Assessment and receive an executive report covering workflow ownership, capacity released, AI debt and prioritised recommendations.