G0 — Minimal Governance
Use where work is very low risk.
Examples: brainstorming; formatting; internal drafting; summarisation.
Typical controls: acceptable-use rules; normal user responsibility.
Free tool 01
Work Classification Worksheet
Purpose: Use this worksheet to decide how a task or activity should be performed.
Step 1 — Define the work
Task / activity
Business outcomeWhy does this work exist?
Current performerHuman / AI / Automation / Mixed
Step 2 — Ask whether the work should exist
If we designed this process today, would we deliberately create this task?
☐ Yes ☐ No ☐ Unsure
If No, classify as Eliminate. If Yes or Unsure, continue.
Step 3 — Assess the work characteristics
| Characteristic | Low / one end | High / other end | Assessment |
|---|---|---|---|
| Frequency | Rare | Frequent | |
| Predictability | Unpredictable | Predictable | |
| Judgment required | Low | High | |
| Consequence of error | Low | High | |
| Reversibility | Easy | Difficult | |
| Data sensitivity | Low | High | |
| Business context required | Limited | Deep | |
| Rule clarity | Clear | Ambiguous | |
| Human relationship | Limited | Important | |
| Failure visibility | Easy to detect | Difficult to detect | |
| Recovery difficulty | Easy | Difficult | |
| Evidence requirement | Low | High |
No single characteristic determines the answer. Consider the whole work profile.
Step 4 — Choose the work class
- ☐ Eliminate — the work should no longer exist.
- ☐ Human-led — human judgment or authority is central.
- ☐ AI-assisted — AI supports a human who remains directly responsible.
- ☐ AI-executed — AI performs bounded work under human authority.
- ☐ Deterministic automation — normal software or rules should perform the work.
Step 5 — Record the reasoning
Why is this classification appropriate?
What could make the classification change later?
Free tool 02
Lightweight Intent Contract
Purpose: Use this before material Human-AI work begins. The Intent Contract prevents the AI and human from solving different problems.
Desired OutcomeWhat should be achieved?
Definition of DoneWhat evidence will show that the outcome has been achieved?
In ScopeWhat work is authorised?
Out of ScopeWhat is explicitly excluded?
ConstraintsWhat must remain true? Examples: budget limit; systems that must not be changed; customer impact; legal/security boundaries; time boundary.
Human AuthorityWho owns the outcome or material decisions?
AI RoleWhat may AI recommend, analyse and execute?
Stop ConditionsWhen must AI stop and return control?
Escalation DestinationWho receives control when a stop condition occurs?
Free tool 03
Human-AI Responsibility Map
Purpose: Use this to prevent unclear accountability.
| Responsibility | Assigned role / system |
|---|---|
| Process Owner | |
| Recommender | |
| Executor | |
| Approver | |
| Monitor | |
| Escalation Owner | |
| Knowledge Owner | |
| Risk / Control Owner, if required |
Key questions
Who owns the outcome?
What may AI recommend?
What may AI execute?
What requires human approval?
What operates under standing approval?
Who receives exceptions?
Responsibility check
- ☐ The process has a human owner.
- ☐ Recommendation and approval are not confused.
- ☐ AI execution boundaries are known.
- ☐ Human review is meaningful.
- ☐ Escalation has a real destination.
- ☐ Knowledge has an owner.
Free tool 04
Governance Level Quick Guide
G1 — Standard Operational Governance
Use where work is repeatable, low risk and easy to correct.
Examples: routine categorisation; internal knowledge search; low-risk routing; standard reporting.
Typical controls: Process Owner; standing approval; monitoring; exception handling.
G2 — Controlled Governance
Use where work may create material operational, customer, financial or data impact.
Examples: customer-impacting actions; privileged access; production change preparation; sensitive information processing.
Typical controls: execution boundary; evidence requirements; stop conditions; approval where required; monitoring; escalation; durable record; recovery consideration.
G3 — Enhanced Governance
Use where consequences are high or difficult to reverse.
Examples: destructive actions; regulated decisions; major financial actions; safety-critical decisions; significant employment decisions.
Typical controls: explicit human authority; separation of duties; strong evidence; independent validation where appropriate; recovery/fallback; detailed audit trail.
Selected levelG0 / G1 / G2 / G3
Why?
Governance friction check
- What risk does it reduce?
- Is that risk material?
- Does the control actually reduce it?
- Is there a simpler control?
- Is it duplicated elsewhere?
- Would monitoring work better than approval?
Free tool 05
Human Takeover Quick Check
Purpose: Use this to check whether a critical AI-enabled process remains operable without the AI.
Understanding
- ☐ A competent human can explain the process.
- ☐ The desired outcome is understood.
- ☐ Authority and responsibility are known.
Access
- ☐ Required systems can be accessed without the AI.
- ☐ Required credentials or credential references are known.
- ☐ Required data remains available.
Documentation
- ☐ Current procedures exist.
- ☐ Procedures are usable at the operational level.
- ☐ Runbooks exist where needed.
- ☐ Important knowledge is not trapped only in chat.
Execution
- ☐ A human can perform critical steps.
- ☐ A human knows how to recognise success.
- ☐ A human knows how to recognise failure.
- ☐ Escalation paths are known.
Recovery
- ☐ Recovery or fallback is understood.
- ☐ Previous safe state can be restored where needed.
Readiness classification
Evidence of readiness: Documentation, access and procedures exist. ☐ Yes ☐ No
Demonstrated capability: A competent human has successfully performed or tested the takeover. ☐ Yes ☐ No ☐ Not required for this risk level
Free tool 06
Basic HAIOF Self-Assessment Checklist
Rate each statement from 0 to 5.
0 Not established1 Ad hoc2 Emerging3 Defined4 Managed5 Adaptive
Work Design
- We examine whether work should exist before automating it.
- We distinguish AI from deterministic automation.
- Important workflows have been deliberately classified.
Score: ___ / 5
Human Authority
- Material AI-enabled processes have clear human owners.
- Approval authority is explicit.
- AI execution boundaries are understood.
Score: ___ / 5
Knowledge
- Important knowledge has owners.
- Authoritative sources are known.
- Outdated knowledge is retired or clearly marked.
Score: ___ / 5
Governance
- Governance is proportional to risk.
- Stop conditions and escalation paths exist.
- Controls are reviewed for unnecessary friction.
Score: ___ / 5
Human-AI Collaboration
- AI is used where it improves the work.
- Human review is meaningful rather than symbolic.
- AI may challenge instructions when material evidence changes.
Score: ___ / 5
Human Independence
- Critical workflows can continue without the AI.
- Required access and procedures exist.
- Takeover capability is tested where appropriate.
Score: ___ / 5
Measurement
- AI initiatives have clear intended outcomes.
- End-to-end performance is measured.
- Human review, rework and cost are included.
Score: ___ / 5
AI Operations
- AI-enabled workflows have operational owners.
- Failures, drift and exceptions are monitored.
- Changes to AI behaviour are controlled where material.
Score: ___ / 5
Do not rely only on the average score. A serious weakness in one domain can matter more than strong scores elsewhere.
Reflection
Strongest area
Weakest area
Most material risk
Biggest source of friction
First workflow to examine
Practitioner resources · Preview only
Deeper records remain practitioner-only.
These may be named and described publicly, but the complete reusable templates are not published.
Decision Register
Decision, authority, status, affected workflows and review trigger.
Knowledge Register
Source, ownership, authority, sensitivity, freshness and retirement.
Knowledge Debt Register
Missing, unreliable, conflicting or poorly maintained knowledge.
Governance Record
Governance level, boundaries, evidence, stop conditions and recovery.
AI Operating Model Decision Matrix
Supports deliberate operating-model choices.
Evaluation Plan
Cases, prohibited actions, thresholds and acceptance evidence.
Pilot Scorecard
Baseline, outcomes, human effort, cost, quality and risk.
Incident Record
Failure, impact, recovery, evidence and corrective action.
AI Change Record
Material change, expected effect, verification and rollback.
KPI Record
Outcome, quality, effort, cost and risk measures.
Transformation Backlog
Evidence-backed priorities and ownership.