11.4 Audit Plan, Team & Working Documents

Key Takeaways

  • The audit plan converts objectives, scope, and criteria into a time-bound operational schedule covering dates, sites, team, methods, interviews, and logistics—and must be communicated to the auditee in time for preparation.
  • Team selection for AIMS audits requires management-system audit competence plus sufficient AI-domain understanding (or technical experts) matched to in-scope systems, risks, and technologies.
  • Responsibilities must be assigned clearly: team leader ownership of the plan and conclusions, auditors for process areas, technical experts for explanation without taking audit decisions, and guides for logistics—not for answering for process owners.
  • Working documents (checklists, sampling plans, interview guides, evidence logs) should be tailored to AI artifacts—impact assessments, SoA lines, model/system documentation, data lineage, monitoring logs, human-oversight records, supplier controls—without becoming rigid scripts that ignore unexpected evidence.
  • Plans and working documents are living tools: controlled updates are allowed when scope or logistics change; secrecy that blindsides the auditee on logistics is poor practice, while revealing every audit test question in advance can undermine sampling integrity if misused.
Last updated: August 2026

11.4 Audit Plan, Team & Working Documents

Quick Answer: When feasibility, objectives/scope/criteria, and Stage 1 readiness are in place, operationalize the engagement. Produce an audit plan, select a competent team for the AI domains in scope, assign responsibilities, prepare working documents tailored to AIMS evidence, and communicate the plan so the auditee can provide access. ISO 19011 Clauses on planning and working documents apply; certification bodies add ISO/IEC 17021-1 plan expectations.

This section is the last preparation gate before opening meetings and evidence collection (Chapter 12). A brilliant Stage 1 report is wasted if Stage 2 starts with no plan, the wrong skill mix, and a generic ISO 9001 checklist with “AI” written in the header.


Audit Plan Contents

An audit plan is the team’s operational schedule and agreement framework for a specific audit (or Stage). Typical contents:

Plan elementAIMS-oriented detail
ObjectivesCertification Stage 2, surveillance focus, internal objective, etc.
CriteriaISO/IEC 42001, SoA-applicable controls, organizational/legal requirements agreed
ScopeEntities, sites, AI systems/processes, exclusions
Dates and timesStage days, time zones for remote interviews
Sites / platformsOffices, cloud environments, SOC-like monitoring rooms, vendor meetings
Audit teamNames, roles, contact points
MethodsInterviews, observation, document/record review, remote screen-share of registries
Schedule of activitiesProcess areas mapped to times and auditors
LogisticsRooms, remote tools, security clearance, interpreters, NDAs
Sampling approach (high level)Which systems/classes get deep dives
Report timing & languageDraft/final expectations
Confidentiality & safetyHandling of model IP, personal data in prompts, restricted labs

Illustrative multi-day flow: opening and context → risk/impact → lifecycle and data deep dive → deployment/monitoring/oversight/third parties → performance evaluation/improvement → closing. Adjust to risk and size; keep traceability from plan lines to criteria and systems. Prefer secure read-only registry/dashboard access over screenshots alone; pair process and technical owners; book third-party sessions when foundation-model controls are critical; buffer for access failures.


Selecting Competent Team Members for AI Domains

Competence = ability to apply knowledge and skills to achieve intended results. For AIMS audits, combine:

  1. Management system audit competence — ISO 19011 process, evidence, findings, impartiality
  2. ISO/IEC 42001 knowledge — clauses, Annex A intent, SoA logic
  3. AI domain competence — enough to understand evidence in scope (ML lifecycle, data issues, generative systems, human oversight, evaluation metrics)
  4. Sector competence where needed — medical, financial, safety-critical contexts
Team roleFunctionCaution
Audit team leaderPlan ownership, direction, consolidation, client communicationMust have lead competence; cannot outsource conclusions
AuditorCollect evidence in assigned areasNeeds enough AI literacy for assigned systems
Technical expertProvide specific AI/sector knowledge to the teamDoes not act as auditor of record; no independent findings without auditor evaluation
Translator / interpreterLanguage supportMust not filter or “improve” auditee answers
GuideLogistics, introductions, escortsMust not answer for process owners or steer sampling
ObserverTraining or CB oversightNo interference with audit

Match skills to systems (e.g., generative chatbots need prompt/filter/transparency literacy; credit models need model-risk and drift evidence literacy). If the only auditor is an ISMS specialist with zero AI literacy, add a technical expert or reassign. Preparation also includes impartiality conflict checks for anyone who recently implemented the AIMS or systems under review.


Assigning Responsibilities

Document assignments in the plan or a team briefing note:

  • Who leads opening/closing meetings
  • Who owns each process cluster (e.g., Clause 6 risk/impact vs. A.6 lifecycle vs. A.10 third parties)
  • Who deep-dives which AI systems
  • Who consolidates findings each day
  • Who manages evidence logging and confidentiality of AI artifacts
  • Escalation path if access is blocked mid-audit

Team briefing before site work should cover Stage 1 concerns, high-risk systems, cultural notes, remote tool rules, and “what good evidence looks like” for AI (not only signed policies).


Working Documents Tailored to AI Artifacts

Working documents guide and record audit activities. They are tools, not prisons. ISO 19011 warns against checklists that prevent additional investigation.

Core working document types

DocumentAIMS tailoring ideas
Clause/control checklistMap 4–10 + SoA controls to example evidence questions
AI system sampling planPopulation (inventory), risk ranking, sample size/rationale, backups if systems unavailable
Interview guidesRole-based questions for executives, model owners, data stewards, human reviewers, procurement
Observation guidesWhat to watch in human-in-the-loop queues, release meetings, incident bridges
Evidence/request logsTrace requests for model cards, lineage extracts, monitoring exports
Finding drafts worksheetsCriteria → evidence → evaluation → classification
Time sheets / attendanceSupport later credential hour claims where relevant

Build guides around impact assessments, SoA vs. procedures, model/system docs, data quality evidence, evaluation results, deployment approvals, monitoring/incident records, human-oversight logs, transparency artifacts, and third-party due diligence. A Y/N “do you have an AI policy?” checklist misses effectiveness—pair presence with implementation probes. For multi-system sampling, record stratification (risk, lifecycle, tech, site), mandatory deep dives, medium-risk sample size/rationale, and fallbacks if a system is deprecated. Judgmental sampling is fine when reasoned and recorded.


Communicating the Plan to the Auditee

Communicate the plan with enough lead time for the auditee to:

  • Book process owners and executives
  • Arrange access to tools and rooms
  • Notify third parties if joint sessions are needed
  • Prepare logistics without manufacturing fake records overnight (you cannot fully prevent last-minute document creation, but surprise full schedules make chaos more likely)

What to share vs. hold back

Usually shareUse professional judgment
Objectives, scope, criteria, dates, team names/rolesExact detailed checklist questions that would allow gaming every sample
High-level agenda by process areaConfidential risk notes for the team only
Logistics and access needsInternal team disagreements
Request for key documents/records availabilityPersonal opinions about individuals

Confirm receipt and invite questions. Update the plan when the auditee reveals a site outage or a critical system owner is on leave—document the change.

Final gate: feasibility/access still holds; objectives/scope/criteria match Stage 1; plan acknowledged; competence and impartiality confirmed; working documents controlled; confidentiality briefed; sampling plan ready. Then the engagement is prepared to conduct—opening meetings and evidence collection succeed only if this preparation was real.

Test Your Knowledge

Which set of elements belongs in a typical ISO/IEC 42001 audit plan?

A
B
C
D
Test Your Knowledge

What is the correct role of a technical expert on an AIMS audit team?

A
B
C
D
Test Your Knowledge

Why should working documents for an AIMS audit be tailored to AI artifacts rather than only generic facilities checklists?

A
B
C
D
Test Your Knowledge

When communicating the audit plan to the auditee, which practice is most professional?

A
B
C
D