Chapter 15 of Evolve establishes that responsibility and authority for AI governance roles must be explicitly assigned and communicated (International Organization for Standardization, 2023), and notes that Appendix B points to a clause-by-clause mapping of ISO/IEC 42001:2023 to the product operating model (Hodgson, 2026, Chapter 15). Appendix B identifies that the Sprint cadence, investment hypothesis, and Product Leadership Trinity together satisfy ISO 42001 Clauses 5, 6, 8, 9, and 10 — for organisations that need to demonstrate compliance and govern effectively (Hodgson, 2026, Appendix B). This page is that mapping.
In Brief
- The four structural commitments in Evolve map directly to ISO 42001’s leadership requirements (Clause 5) — each is an executive action that changes the conditions under which teams operate, not a team-level design decision.
- The Sprint Review and Retrospective satisfy Clauses 9 and 10’s monitoring, audit, and improvement requirements at a fortnightly cadence — more frequently than the standard specifies.
- The operating model’s product lifecycle stages (Introduction, Growth, Maturity, Decline) are the strongest single alignment with Annex A.8’s AI system lifecycle controls.
- Data governance is the most significant structural gap — Annex A.7 and the data quality requirements in Clause 8 have no direct counterpart in the operating model and must be built separately.
- The irreversibility boundary — the human Product Owner’s release decision — is the primary operational control mechanism and satisfies ISO 42001’s human oversight requirement at the point of highest consequence.
ISO/IEC 42001:2023 — Artificial Intelligence — Management System (2023 edition) specifies the requirements an AI management system must satisfy. The product operating model described in Evolve is the structural mechanism through which those requirements are most directly satisfied in practice. The clause-by-clause mapping covers each operative clause (4 through 10) and the key Annex A control categories, identifying where the operating model directly satisfies the standard and where implementation gaps typically remain. It is a readiness mapping tool for the practitioner preparing for an assurance review — not a compliance claim for Evolve.
Clause 4 — Context of the Organisation
ISO/IEC 42001:2023 requires the organisation to understand its internal and external context, identify the interested parties whose needs bear on the AI management system, determine the scope of that system, and establish the system itself (International Organization for Standardization, 2023).
The product operating model addresses context directly through the distinction it establishes between customers — the people and teams who consume a product and whose experience generates demand signals — and stakeholders, who create and impose governance obligations on the product and its value stream (Hodgson, 2026). ISO 42001 uses structurally similar logic: interested parties include both users and those with legal, regulatory, or governance interests in the AI system’s behaviour. The operating model requires the Product Manager to maintain a clear working understanding of both, and to route their different signals through different mechanisms — customer needs through Voice of Customer forums; stakeholder obligations through explicit governance constraints encoded in the team’s governing instructions.
When compliance requirements and user needs arrive through the same channel, the Product Manager has no structural basis for distinguishing a governance constraint from a demand signal. One is a boundary within which the product must operate; the other is evidence of what the product should become. An organisation that routes both through the same working group will find that compliance crowds out discovery — or, worse, that discovery proceeds without the constraints compliance exists to enforce (Hodgson, 2026).
Scope determination under ISO 42001 asks the organisation to define which AI systems and which organisational units fall within the management system. Under the product operating model, each product team operates under a distinct set of governing instructions — skill files, quality standards, and a Definition of Done — that together constitute a documentable, version-controlled scope boundary. The scope is not defined once at deployment; it is revised at each Retrospective as the team’s operating context changes.
The typical implementation gap at Clause 4 is not a failure to document scope but a failure to connect scope documentation to the operating model’s revision cycle. An organisation that defines its AI management system scope as a static document signed off at deployment is describing the governance of a system that no longer exists within two Sprint cycles. A scope that is not subject to the same revision logic as the governing instructions it describes is a record of past governance rather than current governance.
Clause 5 — Leadership
ISO/IEC 42001:2023 requires top management to demonstrate commitment to the AI management system, establish an AI policy, and assign roles, responsibilities, and authorities (International Organization for Standardization, 2023). The standard is explicit that top management accountability is evidenced through resource allocation, reporting structures, and authority design — not through policy documents alone.
The four structural commitments in Evolve are each a leadership action, not a team-level decision. Single-point accountability (Commitment 1), persistent outcome funding (Commitment 2), consumption-speed financial governance (Commitment 3), and order-maker orientation (Commitment 4) cannot be installed by a delivery manager or a product team; they require an executive to change the conditions under which teams operate (Hodgson, 2026). Each commitment produces structural evidence of leadership involvement that ISO 42001’s Clause 5.1 requires: resource allocation is changed (Commitments 2 and 3), accountability structures are redesigned (Commitment 1), and the function of the technology organisation is redefined (Commitment 4).
The Product Leadership Trinity — Product Manager, Product COO, and Chief Product Architect — provides the three-role accountability structure through which those commitments are put into practice (Hodgson, 2026, Chapter 5). Clause 5’s requirement for assigned roles and authorities is satisfied when the Trinity’s decision scope is explicit: the Product Manager holds the Persevere and Pivot investment postures, deciding in real time whether to continue or redirect the current approach; the executive at the quarterly portfolio review decides Pause and Pull (Hodgson, 2026, Chapter 10). A leadership structure in which those four postures are unassigned is not consistent with ISO 42001’s authority design requirements. The Sprint cadence makes those authorities operational: each Sprint Review produces evidence of whether the current posture is justified, and that evidence accumulates across cycles until the quarterly portfolio review applies it to Pause and Pull decisions (Schwaber & Sutherland, 2020).
The AI policy requirement at Clause 5.2 maps most directly to the team canvas and governing instructions that constitute the written governance framework for each AI team. A canvas that asserts a priority ordering of obligations — Product Goal above Sprint Goal above Definition of Done above task completion — specifies, in advance of any action, what the AI system is authorised to do and what standards its output must meet (Hodgson, 2026). That is the function of an AI policy: it establishes the operating principles before any agent acts.
The typical implementation gap at Clause 5 is a diffusion of accountability that the standard explicitly prohibits. When multiple managers share authority over what an AI team releases, contested outputs generate meetings rather than decisions. ISO 42001’s requirement for assigned roles and authorities is not satisfied by a matrix that lists multiple accountable parties. The Product Owner holds accountability at the consequence boundary — the release decision — with authority matched to that accountability. Assigning accountability at this point to a committee, working group, or shared-responsibility arrangement is not equivalent.
Clause 6 — Planning
ISO/IEC 42001:2023 requires the organisation to identify the risks and opportunities relevant to its AI systems, conduct a risk assessment, develop and implement a risk treatment plan, and establish measurable objectives for the AI management system (International Organization for Standardization, 2023).
The product operating model addresses risk assessment through the Sprint cadence rather than through a periodic review. Each Sprint Planning identifies the work the team will undertake and the governing instructions under which it will operate. Each Sprint Review surfaces what the team produced and whether any agent behaviour diverged from what the governing instructions anticipated. Each Retrospective revises those governing instructions in response to what inspection revealed (Hodgson, 2026; Schwaber & Sutherland, 2020). That cycle is a risk assessment and treatment process operating at a fortnightly cadence rather than annually.
The investment hypothesis makes the risk treatment planning explicit. The four postures — Persevere, Pivot, Pause, and Pull — are the decision options available when the Sprint cadence produces evidence that the current approach is not producing the expected outcome (Hodgson, 2026, Chapter 10). Persevere is the default when evidence supports continuing. Pivot redirects the approach within the same product when the current direction is not working. Pause suspends investment pending a governance review. Pull removes a product from the portfolio when the quarterly review determines it no longer satisfies its investment hypothesis. Each posture is a named risk treatment decision with an assigned decision-maker — the Product Manager for Persevere and Pivot; the Product Manager, Product COO, and Chief Product Architect at the quarterly portfolio review for Pause and Pull (Hodgson, 2026, Chapter 5) — which is precisely what ISO 42001’s risk treatment plan requires.
When an AI agent produces unexpected behaviour, the instinct is to add a prohibition — to expand the governing instructions with a new constraint. The more effective response is to narrow the options available to the agent, removing a decision the agent was not equipped to make rather than trusting it to exercise existing decision authority correctly under a new constraint (Hodgson, 2026). That distinction maps to ISO 42001’s requirement to select risk treatment options that remove root cause rather than add compensating controls around a condition that remains in place.
The typical implementation gap at Clause 6 is the absence of measurable AI management system objectives. An organisation can operate a Sprint cadence without stating what the cadence is designed to produce at a system level. ISO 42001 requires objectives that are measurable, monitored, and revised when circumstances change. The operating model provides the inspection and revision mechanisms; the executive must provide the objectives against which those mechanisms operate. Where those objectives are not defined explicitly, the management system has no criterion against which to evaluate whether the Sprint cadence is producing the governance outcomes the organisation requires.
Clause 7 — Support
ISO/IEC 42001:2023 requires the organisation to provide the resources, competence, awareness, communication, and documented information necessary for the AI management system to operate effectively (International Organization for Standardization, 2023).
Persistent teams (Commitment 2 of the operating model) are the primary mechanism through which competence is built and retained. A team that has operated through forty Sprint cycles has calibrated its governing instructions against forty cycles of inspection and adaptation. That calibration sits in the versioned instruction set, in the team’s Definition of Done, and in the accumulated professional judgement of the Product Owner who learned through those cycles when to release value and when to hold. Disband the team and the written artefacts survive; the calibrated judgement disperses. The replacement team inherits the documentation and starts the learning cycle from the beginning (Hodgson, 2026). ISO 42001’s competence requirement is satisfied not by documenting role descriptions but by maintaining the structural conditions under which competence develops.
Documented information under ISO 42001 includes the management system’s own documentation, records generated by the system’s operation, and the information required to demonstrate the system’s effectiveness. The skill files, governing instructions, and version-controlled Definition of Done that an AI team maintains are precisely this category of documents. Their version history constitutes an organisational record of how governance has evolved, which decisions triggered revisions, and what evidence each revision was based on — a form of documented information more auditable than a policy document updated annually.
The typical implementation gap at Clause 7 is the treatment of awareness and communication as one-time activities — an onboarding briefing, a policy distribution at deployment — rather than as ongoing components of the Sprint cadence. ISO 42001 requires awareness and communication that is proportionate to the role each person plays in the management system. Under the product operating model, the Sprint Review and Retrospective are the standing communication mechanisms. An organisation that manages its AI management system’s communication requirements outside those forums is operating two separate governance processes against the same activity, generating overhead that neither satisfies the standard nor improves team performance.
Clause 8 — Operation
ISO/IEC 42001:2023 requires the organisation to plan, implement, and control the processes needed to meet its requirements, conduct AI system impact assessments, manage the AI system lifecycle, ensure the quality of data used in AI systems, and address identified risks (International Organization for Standardization, 2023).
The irreversibility boundary — the release decision held by a human Product Owner — is the primary operational control mechanism in the product operating model. Below that boundary, agents coordinate under bounded authority, take actions within their defined scope, and produce output that is assessed at Sprint Review. Above it, a human makes the release decision and is accountable for it (Hodgson, 2026). That structure maps to ISO 42001’s operational control requirement: processes proceed only when specified conditions are met, and those conditions include human review at the points where consequences are most severe.
Persevere and Pivot sit within the Product Manager’s authority; Pause and Pull require the quarterly portfolio review, where the Product Manager, Product COO, and Chief Product Architect assess the evidence and decide together (Hodgson, 2026, Chapter 5, Chapter 10). ISO 42001’s Clause 8.1 requires planned and controlled processes — the posture structure provides named decision points, assigned authorities, and defined triggers for each level of operational response. That structure is the operating model’s mechanism for ensuring operational control is exercised at the right level rather than deferred indefinitely.
AI system impact assessment under Clause 8.4 requires the organisation to assess the potential impacts of its AI systems on individuals and affected groups, prior to deployment and at defined intervals thereafter. The Sprint cadence provides the interval structure; the release decision provides the assessment point. What the operating model does not specify is the scope of the impact assessment or the criteria against which it is conducted. Those must be defined by the organisation and encoded in the governing instructions before they can be applied consistently across a portfolio of AI products.
The AI system lifecycle requirements at Clause 8.3 align closely with how the operating model treats a product through its Introduction, Growth, Maturity, and Decline stages. At each stage, different evidence is relevant, different questions apply to the portfolio review, and the financial governance structure changes accordingly (Hodgson, 2026). That lifecycle mapping provides a ready framework for the stage-specific documentation and review that ISO 42001’s lifecycle requirements call for.
The typical implementation gap at Clause 8 is data governance. Annex A.7 addresses the quality, provenance, and documentation of data used to train or operate AI systems — requirements with no direct counterpart in the product operating model as described in Evolve. The operating model addresses output quality and decision authority; it does not specify data governance controls. An organisation mapping this clause must develop the data governance layer separately and integrate it into the Sprint cadence as an additional constraint category within the governing instructions.
Clause 9 — Performance Evaluation
ISO/IEC 42001:2023 requires the organisation to monitor, measure, analyse, and evaluate its AI management system, conduct internal audits, and undertake management reviews (International Organization for Standardization, 2023).
The Sprint Review is the operating model’s primary monitoring and evaluation mechanism. It is designed to surface working output for inspection by the Product Owner, the team, and relevant stakeholders — not a report about what was produced but the product itself, assessed against the outcome the investment was committed to produce. When the Sprint Review operates as designed, it meets ISO 42001’s monitoring requirement at a fortnightly cadence, more frequently than any periodic audit programme can match (Hodgson, 2026).
The Retrospective is the operating model’s internal audit mechanism. It inspects how the team operated rather than what it produced, and its findings result in revisions to the governing instructions before the next Sprint begins. The distinction between the Sprint Review (what was produced) and the Retrospective (how it was produced, and what that reveals about the governing instructions) maps to the distinction ISO 42001 draws between product performance assessment and management system audit. Both are required; each operates on different evidence.
The Evidence-Based Management framework used by the portfolio review provides a measurement structure relevant to Clause 9.1. Four Key Value Areas — Current Value, Unrealised Value, Time to Market, and Ability to Innovate — describe whether an AI system is producing the value it was funded to produce and whether the governance model supporting it is performing as intended. Those indicators are not identical to the performance indicators ISO 42001 requires, but they provide the evidence base from which AI management system metrics can be derived and tracked.
The investment hypothesis postures are assessed at the quarterly portfolio review. Pause and Pull decisions require the kind of evidence the Sprint cadence generates — throughput, quality indicators, outcome data — and the portfolio review — conducted by the Product Manager, Product COO, and Chief Product Architect — is where that evidence is weighed against the product’s investment hypothesis (Hodgson, 2026, Chapter 5, Chapter 10). ISO 42001’s management review requirement at Clause 9.3 is most directly satisfied by this forum, provided it evaluates whether the AI management system’s governance design is working — not only whether individual products are performing.
The typical implementation gap at Clause 9 is the management review. ISO 42001 requires top management to review the AI management system at planned intervals, with reference to audit results, performance trends, and changes in the operating context. Under the product operating model, the portfolio review assesses individual product performance. The standard requires a review of the management system itself — whether it is still fit for purpose, what resources it needs, and whether its objectives require revision. Few organisations schedule this review as distinct from the portfolio governance review. Without the distinction, the management system has no mechanism for reviewing its own design.
Clause 10 — Improvement
ISO/IEC 42001:2023 requires the organisation to address nonconformity, take corrective action, and pursue continual improvement of the AI management system (International Organization for Standardization, 2023).
The Retrospective is the operating model’s improvement mechanism. It produces revisions to governing instructions, quality standards, and bounded authority definitions — what Evolve calls double-loop correction, which revises the governing logic rather than acting within it. A third order of correction is available when the governing framework itself is reassessed: whether the structure that produces instructions still suits the problem the organisation is solving (Hodgson, 2026). ISO 42001’s continual improvement requirement is most directly satisfied by this correction cycle, which the Sprint cadence makes available at a fortnightly frequency rather than the annual or biannual intervals most management system standards assume.
The governance cycle described in Evolve — an inner loop (Sprint cadence produces output and revises governing instructions) and an outer loop (portfolio-level evidence informs whether the governance design itself is working) — maps to the two levels at which ISO 42001’s improvement requirements operate. Clause 10.2 addresses nonconformity and corrective action at the operational level; the inner loop provides the mechanism. The outer loop provides the conditions for strategic improvement of the management system itself, which the standard requires but does not specify the frequency of (International Organization for Standardization, 2023).
Pivot is a named corrective action under Clause 10.2 — the Product Manager has assessed that the current approach requires redirection and has the authority to act on that assessment (Hodgson, 2026, Chapter 10). Pull is the management system’s most consequential corrective action: the product no longer satisfies its investment hypothesis, and the portfolio review — in which the Product Manager, Product COO, and Chief Product Architect each hold a stake in the decision — terminates the investment (Hodgson, 2026, Chapter 5). ISO 42001’s continual improvement requirement is not satisfied by a revision cycle alone — the posture structure is what gives the improvement mechanism defined triggers, assigned authorities, and legible criteria.
The typical implementation gap at Clause 10 is documentation of nonconformity. ISO 42001 requires the organisation to record what nonconformity was identified, what corrective action was taken, and whether that action was effective. Under the product operating model, a governing instruction that produces unexpected behaviour is revised at the Retrospective. What the operating model does not specify is the documentation trail that must accompany that revision. The Retrospective provides the mechanism for identifying and correcting nonconformity; the organisation must ensure that findings and revisions are preserved in a form accessible to an auditor and connected to the specific nonconformity each revision addressed.
Annex A Control Alignment
ISO/IEC 42001:2023 Annex A specifies a set of controls covering policies, organisational design, AI risk assessment, system lifecycle, data management, and related requirements (International Organization for Standardization, 2023). The following maps the categories where the product operating model’s architecture most directly satisfies the controls, and identifies where gaps typically remain.
A.2 — Policies related to AI. The standard requires written policies that address the organisation’s approach to developing and using AI systems. The team canvas and governing instructions perform this function at the team level — they specify, before any agent acts, what it is authorised to do and what standards its output must meet (Hodgson, 2026). The implementation gap is the absence of an organisation-level policy that sets the framework within which team-level policies operate. The operating model does not specify this layer. The executive must establish it before team-level governance can be assessed for consistency.
A.3 — Internal organisation. The standard requires clarity on roles, responsibilities, and authorities related to AI governance. Commitment 1’s single point of accountability addresses this directly: one person is accountable at the release decision, with authority matched to that accountability. The Product Leadership Trinity — Product Manager, Product COO, and Chief Product Architect — provides the accountability chain (Hodgson, 2026, Chapter 5). The typical gap is the absence of defined escalation criteria: under what conditions does a team-level decision require Product Manager authority, and when does a product-level decision require Chief Product Architect or Product COO involvement? The operating model’s irreversibility boundary provides the principle; the organisation must translate it into specific decision rules.
A.4 — Documentation of AI systems. The standard requires documentation of the AI systems within scope, including their purpose, technical description, and governance characteristics. Skill files and governing instructions address the governance documentation requirement. A technical description of the AI systems themselves — which models are in use, how they are configured, what data they operate on — is not specified by the operating model and must be developed separately.
A.5 — AI risk assessment. The standard requires a structured risk assessment for each AI system, identifying risks to individuals and affected groups. The Sprint cadence provides the mechanism for ongoing risk identification and treatment. The implementation gap is the initial risk assessment at system deployment: the product operating model is designed for systems already in operation, and Evolve does not specify a pre-deployment risk assessment framework (Hodgson, 2026). Organisations must develop that framework separately and connect its outputs to the governing instructions before the first Sprint begins.
A.6 — AI system impacts. The standard requires the organisation to consider the impacts of its AI systems on people, communities, and the environment. The Sprint Review provides the ongoing inspection point. Voice of Customer forums provide a structured mechanism for receiving impact signals from users. The gap is external impact assessment: the impacts on individuals who are not system users but who are affected by the system’s outputs. The operating model’s customer orientation is internal to the product; assessing external impact requires a separate framework.
A.7 — Data for AI systems. The standard requires that data used to train, test, or operate AI systems is appropriate, accurate, and documented. As noted at Clause 8, the product operating model does not address data governance. This is the most significant structural gap between the operating model and ISO 42001’s control requirements. It must be developed and integrated independently before the AI management system can satisfy this control category.
A.8 — AI system lifecycle. The standard requires governance across the full lifecycle from design through decommissioning. The operating model’s lifecycle stages — Introduction, Growth, Maturity, and Decline — map directly to this requirement and specify what evidence is relevant to the portfolio review at each stage (Hodgson, 2026, Chapter 10). This is the strongest single point of alignment between the product operating model and ISO 42001’s Annex A controls.
A.9 — AI actors. The standard requires that individuals who develop, deploy, or use AI systems have the competence appropriate to their role. Persistent teams address this at the team level. The implementation gap is competence for roles above the team: what does a Product Manager need to know to govern an AI-augmented team effectively, and how is that competence assessed and maintained? The operating model describes the Product Manager’s accountability without specifying the competence baseline required to exercise it.
A.10 — AI system management. The standard requires ongoing management of AI systems in operation, including monitoring, incident management, and performance tracking. The Sprint cadence, the governing instruction version history, and the Retrospective’s corrective action cycle address this comprehensively. This category represents a strong alignment between the operating model and the standard.
This Is a Mapping Tool
ISO/IEC 42001:2023 specifies what an AI management system must do. The product operating model specifies the structural conditions under which those requirements are most directly satisfied in practice. Implementing one does not produce the other automatically.
An organisation that builds the product operating model as described in Evolve will have, by design, a functioning Sprint cadence, version-controlled governing instructions, a single point of accountability at the release decision, and a persistent team accumulating calibrated governance knowledge (Hodgson, 2026). Those are among the most significant structural commitments ISO 42001’s operative clauses require. They are not the complete set.
Data governance, pre-deployment risk assessment, external impact assessment, and an organisation-level AI policy are not specified by the operating model and require separate development. Third-party AI governance obligations and legal compliance requirements must also be mapped separately and encoded in governing instructions as explicit constraints before they can be enforced at the point of agent execution. The operating model addresses the need to restructure external engagements away from fixed-scope contracts — a structural change with direct implications for supply chain AI governance (Hodgson, 2026, Chapter 15) — but does not specify AI governance obligations for external AI system providers. The mapping above identifies where prior investment in the operating model reduces what remains to satisfy ISO 42001’s requirements, and where that investment leaves gaps that must be addressed before a certification audit. It is a diagnostic for scoping the effort — not a representation that Evolve certifies an organisation against the standard.
ZXM works with executive teams to develop the governance layers this mapping identifies as gaps: data governance, pre-deployment risk frameworks, external impact assessment, and an organisation-level AI policy that gives team-level governance its authority. To discuss where your AI management system stands against ISO 42001’s requirements, contact us.
References
- Hodgson, M. (2026). Evolve: The operating model AI demands. Zen Ex Machina.
- International Organization for Standardization. (2023). ISO/IEC 42001:2023 Information technology — Artificial intelligence — Management system. ISO/IEC. https://www.iso.org/standard/81230.html
- Schwaber, K., & Sutherland, J. (2020). The Scrum guide: The definitive guide to Scrum: The rules of the game. Scrum.org. https://scrumguides.org/scrum-guide.html