AI Capability Maturity Assessment · Part III. Running an assessment

Toolkit: survey, interviews, workshops and evidence

25 minute read · Version 1.0.0

In this chapter

Everything an assessor needs to gather evidence: the staff survey (25 statements), interview guides for 9 roles, five workshops and the evidence request (41 items). Each downloads as an editable Word document.

Download the toolkit in Word
Staff survey
Interview guides by role
Workshops
Evidence request

Download the toolkit in Word

Each instrument downloads as an editable Word document, generated in your browser from the same model as this page. Choose the edition for the markings and wording.

Edition Australian Government edition Organisation edition

Staff survey

An anonymous staff survey measures how AI practice reaches people: actual use, training, rules, trust and reporting. It surfaces unapproved tools and unreported near misses that interviews miss.

Administration: Open to all staff in scope for 7 to 10 days. Anonymous; no free-text question asks for names. Use a survey tool the organisation already approves.
Scale: 1 Strongly disagree; 2 Disagree; 3 Neither agree nor disagree; 4 Agree; 5 Strongly agree
Don’t know: Don’t know. Excluded from rates; its share is reported per item.

About you

No.	Question	Answer type
P1	Which business area do you work in?	single select (organisation’s list)
P2	Which best describes your role?	single select: frontline or service delivery; specialist or professional; manager; senior leader; technology or data
P3	How often do you use AI tools in your work?	single select: daily; a few times a week; weekly; monthly or less; never
P4	Which AI tools do you use for work? Include features built into software you already use.	multi select (organisation’s approved list plus ‘public tools such as chatbots’, ‘features inside other software’, ‘other’)
P5	Have you completed the organisation’s training on using AI responsibly?	single select: yes; no; not sure

Statements

Each statement maps to the criteria it informs. Negative statements are reverse-scored.

No.	Statement	Criteria	Polarity
S01	I know what our organisation wants to achieve with AI.	1.1	Positive
S02	I know what our organisation has decided not to use AI for.	1.1, 1.2	Positive
S03	Leaders in my area talk about AI use and its risks in a way that helps me.	1.3, 8.4	Positive
S04	I know who to ask for a decision about using AI in my work.	2.1, 2.5	Positive
S05	If my team wanted to use AI for a new task, I know the process we would follow.	7.1, 3.3	Positive
S06	I know which AI tools I am allowed to use for work.	5.3, 8.3	Positive
S07	The approved AI tools meet my needs well enough that I do not need other tools.	5.3	Positive
S08	In the last three months I have used an AI tool for work that I was not sure was approved.	5.3	Negative
S09	I know what information I must not enter into AI tools.	4.3, 8.3	Positive
S10	I know when work I have done with AI needs to be saved as a record.	4.5	Positive
S11	The AI training I have received prepared me to use AI tools safely.	8.1	Positive
S12	I check AI outputs for accuracy before I rely on them or pass them on.	8.5, 8.1	Positive
S13	I can recognise when an AI tool should not be used for a task.	8.1	Positive
S14	I sometimes rely on AI outputs without checking them because there is not enough time.	8.5, 6.2	Negative
S15	If an AI tool produced something wrong, biased or unsafe, I know how to report it.	3.5, 8.5	Positive
S16	I would feel safe reporting an AI mistake, including one I made myself.	8.5	Positive
S17	When problems with AI are reported here, something changes as a result.	8.5, 3.5	Positive
S18	When I am asked to approve or review an AI output or action, I have enough information and time to make a real decision.	6.2	Positive
S19	I know when I must tell a customer or member of the public that AI was involved in their matter.	3.4	Positive
S20	The information our AI tools draw on is current and correct.	4.2, 4.1	Positive
S21	There is someone I can go to for help with AI tools in my area.	8.4	Positive
S22	AI has made a measurable difference to the quality or speed of my work.	7.4	Positive
S23	I understand how AI is changing my role, and I have had a say in how it is used in my team.	8.4, 8.2	Positive
S24	I have the skills I need for the AI-related parts of my role.	8.2	Positive
S25	I know where to find guidance on using AI at work.	8.3, 8.4	Positive

Free text

No.	Question
F1	Where could AI help most in your area?
F2	What stops you using AI more, or more safely?
F3	Would you be interested in being an AI champion for your area? (yes, maybe, no)

Calculation

Agreement rate: For a positive item: (Agree + Strongly agree) divided by responses excluding Don’t know.
Concern rate: For a negative item: (Agree + Strongly agree) divided by responses excluding Don’t know. A negative item contributes (1 minus concernRate) to signals.
Criterion signal: The mean of the item contributions mapped to the criterion, as a percentage.
Validity: Results can adjust or flag scores only when responses reach n >= min(100, max(20, 15 percent of staff in scope)). Below that they are reported as indicative and trigger no caps.
Anonymity: No breakdown is reported for any group with fewer than 10 respondents.

Rules

Rule	What it does
No uplift	Survey results never raise a criterion score. They can corroborate it, flag it or cap it.
Corroborate	A criterion signal of 70 percent or more is cited as corroborating evidence for that criterion’s score.
Not reaching staff	If a criterion is scored 3 or above and its signal is below 50 percent, flag ‘practice not reaching staff’. Moderation must review the flag; if the practice is confirmed as not operating for staff, reduce the score to the highest level whose pass test is still met.
Shadow ai	If the concern rate on S08 is 25 percent or more, criterion 5.3 cannot score above 3, and its Level 4 pass test (detection working) is treated as not met.
Unchecked reliance	If the concern rate on S14 is 30 percent or more, flag criteria 6.2 and 8.5 for moderation and report ‘unchecked reliance on AI outputs’ as a finding.
Awareness gap	If Don’t know exceeds 40 percent on any item, report an awareness gap for its criteria.
Pass test inputs	Criterion 8.3: Level 3 needs the mean agreement on S06 and S09 of at least 70 percent; Level 4 needs at least 85 percent. Criterion 8.5: Level 3 needs agreement on S15 and S16 of at least 60 percent each.
Area divergence	If any business area’s signal for a criterion differs from the organisation’s by 30 percentage points or more (groups of 10 or more only), report it as a finding naming the area.

Interview guides by role

Interviews run 30 to 60 minutes with two assessors: one leads, one codes each substantive answer to a criterion and an evidence confidence as it is given. Interviews test and extend what the documents show; they do not replace them. Run them in weeks 2 and 3, after the evidence request has been issued.

Common opening

No.	Question	Criteria
O1	Where is AI being used in your area today, including features that arrived inside products you already use?	2.3, 5.4
O2	Who would you say is accountable for that use, and would they say so too?	2.1
O3	What would worry you most if an auditor, a regulator or a journalist asked about it tomorrow?	3.1, 3.4

Common closing

Of what you have told us, which two points could you show us evidence for, and who holds it?
Who else should we speak to?
Is there anything we did not ask that we should have?

Executive sponsor or Chief AI Officer 60 minutes · 11 questions

Purpose: Test direction, appetite, authority, investment discipline and the executive’s use of evidence.
When: Week 2, first interviews; repeat for 30 minutes in week 5 for playback.
Opening: We are assessing capability across eight domains, not auditing people. We will ask for evidence behind the important points so the result is defensible.

No.	Question	Criteria
ES1	What is the organisation’s ambition for AI, and what has it decided not to do?	1.1
ES2	How is the AI risk appetite set, and who decides a case at the margin?	1.2
ES3	What authority does the AI lead hold in writing, and when was it last used to pause or refuse something?	1.3
ES4	How is AI investment decided, and what stops pilots continuing without a decision?	1.4, 7.3
ES5	What AI evidence do you see, how often, and what did it last change?	1.5
ES6	How are use-case owners chosen, and how are they held to account?	2.1
ES7	What does the governance committee decide, and what has it refused or made conditional?	2.2
ES8	If internal audit examined AI governance next quarter, what would it find?	2.4
ES9	Which AI benefits can you evidence today, and which have fallen short?	7.4
ES10	What would make you confident to let an AI system act without approving each action?	6.2, 5.5
ES11	How would you respond to the first person who reported an embarrassing AI mistake?	8.5

Close: Ask for the strategy, appetite, delegation instrument and last two executive papers on AI if not already provided.

CIO or CTO 60 minutes · 10 questions

Purpose: Test the platform, tool approval, vendor AI, observability, records capability and whole-of-life cost.
When: Week 2.
Opening: We want to understand where AI runs, how it is provisioned and how you would see, stop and reverse it.

No.	Question	Criteria
CT1	Where do your AI workloads run today, and which of them run outside the approved platform?	5.1
CT2	How is a new AI service provisioned, logged and monitored, and how long does that take?	5.1, 5.5
CT3	Which of your largest vendors switched on AI features in the last year, and how did you find out?	5.4, 2.3
CT4	How long does a request for a new AI tool take, from request to decision?	5.3
CT5	What can you see about AI behaviour in production, beyond uptime?	5.5
CT6	How would you stop an AI system mid-task, and what happens to work in flight?	5.5
CT7	How is code written with AI assistance assured before release?	6.4
CT8	Who pays for monitoring, assurance and retirement once an AI system is in production?	1.4
CT9	Which shared AI components do you maintain, and how often are they reused?	7.5
CT10	Could you place a legal hold on AI content, such as chat histories, today?	4.5

Close: Ask for the reference architecture, platform assessment, approved tool list and logging configuration.

CISO or security lead 60 minutes · 9 questions

Purpose: Test AI-specific security controls, information handling, vendor containment, agent identity and incident readiness.
When: Week 2.
Opening: We are interested in what is implemented and evidenced for AI specifically, not general security posture.

No.	Question	Criteria
SE1	Which AI tools and platforms are approved, and how do you detect use outside them?	5.3
SE2	Which AI-specific threats have you tested for, and what did you find?	5.2
SE3	How is your AI control mapping maintained when the control framework changes?	5.2
SE4	How is the sensitivity ceiling of each AI tool enforced, technically?	4.3
SE5	How are vendor AI capabilities assessed and contained, including plugins and connectors?	5.4
SE6	Do agents act under their own identities, and how are their permissions limited?	5.5, 5.2
SE7	Walk me through the first hour of an AI incident that exposed information or harmed people.	3.5
SE8	Show me the last design review you signed off for an AI system.	6.5
SE9	Which AI controls do you test yourselves, and which do you rely on owners for?	2.4

Close: Ask for the AI threat model, control mapping, detection reports and the incident runbook.

Chief Data Officer or data owner 45 minutes · 8 questions

Purpose: Test data quality, shared meaning, knowledge curation, data flows and decision provenance.
When: Week 2 or 3.
Opening: We are looking at the data and knowledge your AI depends on, and whether you could reconstruct what it did.

No.	Question	Criteria
DO1	For the data feeding your highest-value AI use, what is its measured quality and who owns it?	4.1
DO2	Has anyone tested whether that data represents the people the AI makes or informs decisions about?	4.1
DO3	Which definitions do your AI systems depend on, and who owns them?	4.2
DO4	Which content do assistants retrieve from, and how is superseded content removed?	4.2
DO5	Which data flows into AI services leave your environment, and where are they processed?	4.3
DO6	Could we reconstruct, for a decision last month, the model, version and sources involved?	4.4
DO7	Where is data the binding constraint on an AI ambition?	7.1, 4.1
DO8	Which analytics or machine learning models are in operational use, and are they on the register?	2.3

Close: Ask for quality scorecards, the glossary, the knowledge source inventory and sample decision records.

Legal, privacy and records lead 60 minutes · 9 questions

Purpose: Test impact assessment, automated-decision transparency, contestability, records and contractual protections.
When: Week 2 or 3.
Opening: We want to understand which AI touches people and personal information, and how you would answer for it.

No.	Question	Criteria
LP1	Which AI uses touch personal information, and how are they assessed?	3.1
LP2	Which decisions about people does AI contribute to, and how would someone affected find out?	3.4
LP3	How ready is the privacy policy to explain automated decisions from 10 December 2026?	3.4
LP4	How does a person affected by an AI-assisted decision get a human to review it?	3.4, 6.2
LP5	Pick a decision from last month: can we reconstruct how AI contributed to it?	4.4
LP6	Where are AI chat histories and generated content kept, for how long, and who decided?	4.5
LP7	Where does legal advice enter the AI lifecycle today?	3.3
LP8	Do your contracts stop vendors training their models on your data?	5.4
LP9	What complaints, disputes or information requests have involved AI?	3.4, 4.5

Close: Ask for the privacy policy draft, the automated-decision inventory, the records mapping and standard AI contract clauses.

Risk, compliance and internal audit 45 minutes · 9 questions

Purpose: Test the link to enterprise risk, the lines of assurance, tiering consistency and the reliability of reported figures.
When: Week 3.
Opening: We want to know what has been independently tested and what is taken on trust.

No.	Question	Criteria
RA1	How does the AI risk appetite connect to the enterprise risk framework, and where are AI risks reported?	1.2
RA2	Who does what in each line of assurance for AI?	2.4
RA3	When will internal audit next look at AI, and against which criteria?	2.4
RA4	Show me two similar use cases. Do their tiers match?	3.2
RA5	Show me a gate decision that was ‘no’ or ‘not yet’.	3.3
RA6	How many AI near misses were recorded last quarter?	3.5
RA7	Where do the AI dashboard figures come from, and have they been tested?	1.5
RA8	Has anyone independent checked the benefit figures?	7.4
RA9	Show me an escalation from the last six months and how it was resolved.	2.5

Close: Ask for the assurance map, the internal audit plan and any AI audit or review reports.

Use-case owner or business lead 45 minutes · 10 questions

Purpose: Test how one real use case was chosen, assessed, gated, overseen and measured.
When: Weeks 2 and 3; interview at least three owners, including one high-tier use if any exists.
Opening: We will walk through your use case from idea to today. There are no wrong answers; gaps are what we are here to find.

No.	Question	Criteria
UO1	What does the AI you use actually do, and what would happen if it were wrong?	3.1
UO2	Did you know you are the accountable owner? What does that require of you?	2.1
UO3	What changed in the design because of the impact assessment?	3.1
UO4	What tier is the use case, and who assigned it?	3.2
UO5	Which gates did it pass, and what evidence did you provide?	3.3, 7.2
UO6	Which actions can the system take without asking anyone?	6.2
UO7	How do you know it is still behaving as intended?	5.5
UO8	What was the baseline before AI, and what does the measure show now?	7.4
UO9	What result would have stopped the pilot?	7.3
UO10	What preparation and support did you get for the owner role, and what was missing?	8.2

Close: Ask for the use case’s register entry, assessment, gate records and benefit measures.

Engineering or architecture lead 60 minutes · 10 questions

Purpose: Test specification, enforcement of autonomy, evaluation, code assurance, failure behaviour and agent security in practice.
When: Week 3; include a short screen-share of configuration where possible.
Opening: We are interested in what the systems enforce, not what the documents say. Showing us is better than telling us.

No.	Question	Criteria
EN1	Show me the specification for one AI component. What is it not allowed to do?	6.1
EN2	Where is the approval requirement enforced, and could an agent bypass it?	6.2
EN3	When the model provider last updated the model, what tests ran before you accepted it?	6.3
EN4	Which systems ran in shadow before acting, and what did the comparison show?	6.3
EN5	How do you know which production changes were written with AI assistance?	6.4
EN6	What does the system do when it fails or is unsure?	6.5
EN7	How do your agents handle instructions embedded in documents or web pages they read?	5.2
EN8	How would you stop this agent mid-task?	5.5
EN9	How are decision records protected from alteration?	4.4
EN10	What did the last use case reuse from earlier work?	7.5

Close: Ask for evaluation suites and results, merge gate configuration and the approval configuration.

Frontline staff (individuals or a focus group of six to eight) 45 minutes · 10 questions

Purpose: Test whether strategy, rules, tools, training and reporting reach the people who use AI every day.
When: Weeks 2 and 3; at least two groups from different business areas. Run without managers present.
Opening: This is anonymous. We are testing whether the organisation has given you what you need, not testing you.

No.	Question	Criteria
FL1	Which AI tools do you use for work, including features built into other software?	2.3, 5.3
FL2	Which tools are you allowed to use, and have you used others? Why?	5.3
FL3	What information are you not allowed to put into an AI tool?	4.3, 8.3
FL4	What did your AI training teach you to check before relying on an output?	8.1
FL5	Have you seen an AI tool get something wrong at work? What did you do?	8.5, 3.5
FL6	When you approve or review AI work, how much time and information do you have?	6.2
FL7	Who do you go to when you are stuck with an AI tool?	8.4
FL8	Has an AI tool given you an answer based on an old policy or document?	4.2
FL9	What do you say when a customer or member of the public asks whether AI was involved?	3.4
FL10	If you had an idea for using AI in your work, where would you take it?	7.1

Close: Thank participants and explain how findings will be shared back.

Workshops

The workshops use the maturity model as a strategic planning instrument, not only a scoring tool. Leaders work through the model layer by layer (foundations, then ways of working, then practice), compare the current profile with the descriptors, choose targets per domain and criterion, and select initiatives from the library. Every workshop records its decisions in the assessment record so the generated documents can cite them.

W1 Kickoff and ambition sounding

Week 1 · 120 minutes · Executive sponsor or Chief AI Officer; CIO or CTO; CISO or security lead; Chief Data Officer or data owner; Legal, privacy and records lead; Risk, compliance and internal audit; People and capability (HR and learning); AI lead or AI centre of excellence; Internal communications

Inputs

Scope note (enterprise or function; edition; what counts as AI)
Stakeholder map
Draft evidence request with owners
Pre-read: the five levels and the eight domain descriptors
Timeline and interview schedule

Outputs

Confirmed scope and edition
Evidence request issued with owners and due dates
Interview schedule
Survey launch date and staff message
Ambition hypotheses per domain (recorded, not scored)

Minutes	Agenda item
10	Purpose, outcomes and ground rules: score on evidence; absence of a document is a finding, not a failure.
20	The model: levels, domains, layers, pass tests and how scores are calculated.
30	Ambition sounding: each leader places the organisation against the domain descriptors privately, then the group compares placements. Differences of two levels or more become hypotheses for the assessment to test.
20	Scope: business units, edition, the AI definition used, and an amnesty for undisclosed AI use found during discovery.
20	Evidence request walk-through: items, owners, formats and due dates.
10	Interview plan, survey launch date and communications to staff.
10	Risks to the assessment and how they will be handled.

W2 Evidence review and lifecycle walk-through

Week 3 · 150 minutes · AI lead or AI centre of excellence; CISO or security lead; Legal, privacy and records lead; Chief Data Officer or data owner; CIO or CTO; Use-case owner or business lead; Engineering or architecture lead

Inputs

Evidence log with items received and outstanding
Register export
One real use case selected in advance (preferably the highest-tier one)

Outputs

Observations coded to criteria in D2, D3, D4, D6 and D7
List of unregistered AI found
Follow-up evidence actions

Minutes	Agenda item
15	Evidence status by domain: received, outstanding, confirmed absent.
75	Walk the chosen use case from idea to today against the lifecycle: intake, tiering, assessment, specification, testing, gates, approvals, monitoring, records. At each step, show the record or note its absence.
30	Discovery: compare the register with procurement records, software features and survey tool use; list unregistered AI.
20	Agree follow-up evidence requests and owners.
10	Close.

W3 Scoring moderation

Week 4 · 180 minutes · AI lead or AI centre of excellence; Risk, compliance and internal audit. The two assessors and a quality reviewer run the session. Evidence providers join for their domains in 30-minute slots.

Inputs

Both assessors’ independent scores with confidence and evidence IDs
Evidence log
Survey results and flags
Calibration examples

Outputs

Agreed score, confidence, evidence IDs and reason for all 40 criteria
Moderation log
Domain and overall results computed
Open evidence actions

Minutes	Agenda item
15	Calibrate: review two calibration examples together before scoring.
120	Domain by domain: largest differences first; read pass tests from the lowest unmet level; apply caps; agree score, confidence and reason.
25	Review survey flags and the cap rules they trigger.
20	Quality reviewer sample check and open evidence actions.

W4 Target state and roadmap planning

Week 5, after playback of validated scores · 210 minutes · Executive sponsor or Chief AI Officer; CIO or CTO; CISO or security lead; Chief Data Officer or data owner; Legal, privacy and records lead; Risk, compliance and internal audit; People and capability (HR and learning); Chief Financial Officer or finance lead; Use-case owner or business lead

Inputs

Validated maturity profile (overall, radar, heat map)
Gap analysis against a draft ambition preset
Applicable obligations and dates
Candidate initiatives selected by the roadmap logic
Capacity estimate per horizon

Outputs

Domain targets with ambition and time frame
Criterion target overrides with reasons
Prioritised initiative list with horizon and owner
First 90 days plan
Roadmap governance cadence and re-assessment date

Minutes	Agenda item
15	The profile in one page: overall band, shape, strongest and weakest domains, foundation flag.
20	Choose an ambition preset and time frame; confirm obligations that set floors.
45	Foundations layer (D1, D2, D8): read the next level’s descriptors, agree domain targets, lower or raise criterion targets where justified.
30	Ways of working layer (D3, D7): same method.
30	Practice layer (D4, D5, D6): same method; check the layer rule.
15	Break.
35	Review the generated roadmap: must-dos, quick wins in the first 90 days, dependencies, capacity per horizon; move or drop initiatives with reasons.
20	Name owners for every first-horizon initiative and agree the governance cadence.

W5 Read-out and decisions

Week 6 · 90 minutes · Board or accountable authority; Executive sponsor or Chief AI Officer; CIO or CTO; Risk, compliance and internal audit; Chief Financial Officer or finance lead

Inputs

Findings Report
AI Strategy and Roadmap
Governance and Operating Model recommendations
Decision paper listing the decisions sought

Outputs

Recorded decisions
Endorsed or amended targets
Funded first horizon
Re-assessment date
Staff communication plan

Minutes	Agenda item
15	Headline result, foundation flag and the three findings that matter most.
20	Risk exposure and obligations due.
25	Strategy, target state and roadmap, with the first 90 days in detail.
20	Decisions: endorse targets, fund the first horizon, establish or confirm forums and roles, set the re-assessment date.
10	Close and communications to staff.

Evidence request

Issue in week 1. Where an item does not exist, say so; its absence is recorded as evidence and is not chased. Provide documents in their current version; provide records for the last 12 months unless stated.

1. Strategy and leadership

No.	Item	Owner	Format	Criteria
ER-1-01	AI strategy or statement of intent, with version history and endorsement record	Executive sponsor or Chief AI Officer	Document (PDF or Word) and minute	1.1
ER-1-02	All-staff communication of the AI strategy	Internal communications	Email, intranet page or briefing pack	1.1
ER-1-03	AI risk-appetite statement and tier criteria, with endorsement record	Risk, compliance and internal audit	Document and minute	1.2, 3.2
ER-1-04	Appointment and delegation instrument for the Chief AI Officer or AI lead; budget line	Executive sponsor or Chief AI Officer	Document; finance extract	1.3
ER-1-05	AI portfolio register and recent business cases, including whole-of-life costs	Chief Financial Officer or finance lead	Spreadsheet export; documents	1.4, 7.1
ER-1-06	Executive and board papers and minutes on AI for the last 12 months; dashboard definition	Executive sponsor or Chief AI Officer	Minutes and papers (PDF)	1.5

2. Governance and operating model

No.	Item	Owner	Format	Criteria
ER-2-01	Designations of accountable officials or owners; owner acknowledgements and attestations	AI lead or AI centre of excellence	Register extract; documents	2.1
ER-2-02	AI governance committee terms of reference, membership, last six months of minutes, decision log and conditions tracker	AI lead or AI centre of excellence	Documents; log export	2.2
ER-2-03	Use-case register export with field definitions; last discovery sweep method and results; reconciliation records	AI lead or AI centre of excellence	CSV or spreadsheet; documents	2.3
ER-2-04	AI assurance map; internal audit plan; any AI audit reports and findings trackers	Risk, compliance and internal audit	Documents	2.4
ER-2-05	AI RACI, decision categories and escalation log	AI lead or AI centre of excellence	Document; log export	2.5

3. Risk, assurance and transparency

No.	Item	Owner	Format	Criteria
ER-3-01	Impact assessment method and completed assessments for in-scope use cases; refresh schedule; mitigation tracker	AI lead or AI centre of excellence	Documents; tracker export	3.1
ER-3-02	Tiering records with rationale and assessor; any calibration exercise	Risk, compliance and internal audit	Register extract; document	3.2
ER-3-03	AI lifecycle standard; gate decision records for use cases in the last 12 months; retirement records	AI lead or AI centre of excellence	Documents; records	3.3, 7.2
ER-3-04	Published AI transparency statement and review history; privacy policy section on automated decisions; inventory of AI-assisted decisions about people; human review procedure; review-request log	Legal, privacy and records lead	Web links; documents; log export	3.4
ER-3-05	AI incident definition, severity scale and runbook; incident and near-miss register; rehearsal record; post-incident reviews	CISO or security lead	Documents; register export	3.5

4. Data, knowledge and semantics

No.	Item	Owner	Format	Criteria
ER-4-01	Data quality scorecards, thresholds and owners for data domains feeding AI	Chief Data Officer or data owner	Reports; dashboard export	4.1
ER-4-02	Glossary or ontology for priority domains; inventory of knowledge sources used by AI with owners and currency rules	Chief Data Officer or data owner	Documents; inventory export	4.2
ER-4-03	Handling rules for AI; data flow maps for in-scope use cases; label, data loss prevention or masking configuration	CISO or security lead	Documents; configuration screenshots or exports	4.3
ER-4-04	Decision-record standard and five sample records per in-scope system; logging and tamper-evidence configuration	Engineering or architecture lead	Document; record samples; configuration	4.4
ER-4-05	Records policy and mapping of AI content to retention classes; retention configuration in AI tools; legal hold procedure and test	Records manager	Documents; configuration screenshots	4.5

5. Platform, architecture and security

No.	Item	Owner	Format	Criteria
ER-5-01	AI reference architecture, platform security assessment and residency register for AI services	CIO or CTO	Documents	5.1
ER-5-02	AI usage policy, AI threat model, control mapping and AI security test reports	CISO or security lead	Documents	5.2
ER-5-03	Approved AI tool and model list with uses and ceilings; request process; unapproved-use detection reports; provider model change log	CISO or security lead	Document; reports	5.3
ER-5-04	AI procurement checklist and standard contract clauses; completed checklists for recent procurements; embedded feature assessments; vendor change notifications	Procurement and contract management	Documents; records	5.4
ER-5-05	Monitoring dashboards and behaviour measures; halt and rollback procedures; rehearsal records; agent identity and permission records	CIO or CTO	Screenshots or exports; documents	5.5

6. Design and engineering practice

No.	Item	Owner	Format	Criteria
ER-6-01	Specification template and completed specifications for in-scope components	Engineering or architecture lead	Documents	6.1
ER-6-02	Autonomy and approval matrix per system; approval records; approval and override rate reports; platform approval configuration	Engineering or architecture lead	Documents; configuration; reports	6.2
ER-6-03	Evaluation suites, pass criteria and results; shadow or simulation reports; pipeline configuration	Engineering or architecture lead	Reports; configuration	6.3
ER-6-04	Policy for AI-generated code and content; merge gate and scanning configuration; provenance records	Engineering or architecture lead	Document; configuration	6.4
ER-6-05	Secure-and-safe design checklist and signed-off reviews	CISO or security lead	Documents	6.5

7. Use-case delivery and value

No.	Item	Owner	Format	Criteria
ER-7-01	Intake form or workflow; intake records; prioritisation minutes; deferred and refused log	AI lead or AI centre of excellence	Documents; records	7.1
ER-7-02	Register stage history and evidence links for three sample use cases	AI lead or AI centre of excellence	Register extract	7.2
ER-7-03	Pilot charters and graduation decisions	AI lead or AI centre of excellence	Documents	7.3
ER-7-04	Benefits register with baselines; benefit reports to the executive	Chief Financial Officer or finance lead	Spreadsheet; reports	7.4
ER-7-05	Pattern catalogue with owners; reuse records	Engineering or architecture lead	Document or catalogue export	7.5

8. People, capability and culture

No.	Item	Owner	Format	Criteria
ER-8-01	AI training curriculum; completion reports by business area; effectiveness measures	People and capability (HR and learning)	Documents; learning system export	8.1
ER-8-02	AI capability map, gap analysis, sourcing plan and role-based learning paths	People and capability (HR and learning)	Documents	8.2
ER-8-03	Acceptable-use policy, acknowledgement records and any breach handling records	CISO or security lead	Document; records	8.3
ER-8-04	Change plan, champion list with coverage and time allocation, adoption measures by area	People and capability (HR and learning)	Documents; usage reports	8.4
ER-8-05	AI concern channel description (staff and public), reports with responses, communications of changes made	AI lead or AI centre of excellence	Documents; records	8.5

This assessment is not legal advice and is not a certification. Obligations and dates were checked against official sources on 15 October 2026; confirm them against the source instruments, which can change.