ISO/IEC 42001 When a Customer Asks for It: What the AI Management Standard Requires in Practice
November 11, 2026 · 7 min read · Etzal Group Team
A software company in Jakarta receives a supplier questionnaire from a regional bank it hopes to win. Between the questions on information security and business continuity is a new section: "Describe your AI governance. Do you hold or plan to hold ISO/IEC 42001 certification? Provide your AI policy, your inventory of AI systems and your risk assessment method." The company uses AI in three products and has none of these documents. Questions about ISO/IEC 42001 are now arriving in procurement, and they are easier to answer if the work has started before the questionnaire does.
This article explains the standard in practical terms. It is not a substitute for the standard's text, which must be obtained from the standards body, or for advice from an auditor.
What the standard is
ISO/IEC 42001 specifies requirements for an AI management system. A management system is the organised way a company directs and controls an activity: who is responsible, what the rules are, how risks are handled, how results are checked and how things improve.
It applies to organisations that develop AI, provide AI-based products or services, or use AI in their own operations. A company can be in more than one of these roles.
It follows the common structure of ISO management standards, so a company that already runs ISO/IEC 27001 or ISO 9001 will recognise most of the framework.
What it does not do
- It does not certify a model or a product. The certificate is about the organisation's system.
- It does not guarantee outcomes. A certified company can still ship a flawed AI feature. The standard requires that it has a way to find and correct such things.
- It is not legal compliance. Regulations on AI and on personal data apply separately. The standard helps organise the work, and it does not replace legal advice.
- It does not set technical thresholds such as a required accuracy.
Understanding this prevents two mistakes: dismissing it as paperwork, and overselling it as proof that your AI is safe.
Why buyers ask
- AI risk has become a supplier risk. If your product makes a wrong decision about their customer, it is their problem too.
- They need something to check. A recognised standard is easier to assess than a narrative answer.
- Regulation is tightening in several jurisdictions, and buyers want evidence that suppliers are organised.
- They already ask for ISO/IEC 27001. This is the same habit applied to AI.
The building blocks
Seven pieces of work sit behind the standard. Each is useful whether or not you ever seek a certificate.
1. Scope and inventory
Decide which parts of the organisation and which AI systems the management system covers. Then list every AI system:
- What it does and who uses it.
- Whether you build it, buy it or embed a third party's model.
- What data it uses.
- What decisions or actions depend on it.
- Who owns it.
Most organisations are surprised by this list. It usually includes tools staff adopted without anyone deciding.
2. Policy and roles
- An AI policy approved by leadership: what the organisation uses AI for, the principles it follows, and what is not acceptable.
- Named responsibilities: who approves a new AI use, who owns each system, who assesses risk, who handles incidents.
- Leadership involvement that is real, with reviews on the calendar.
3. Risk assessment
For each AI system, a consistent method to identify and rate risks to the organisation and to the people affected. Typical areas:
- Wrong or misleading output.
- Unfair treatment of groups of people.
- Privacy and misuse of personal data.
- Security, including manipulation of the system.
- Lack of transparency to those affected.
- Over-reliance by users.
- Dependence on a supplier.
Then decide treatment: controls to apply, risks accepted, and by whom.
4. Impact assessment
Beyond risk to the organisation, the standard expects consideration of the effects on individuals and society: who could be harmed by this system, how, and how seriously. This matters most for systems that influence decisions about people, such as hiring, lending, pricing or access to services.
5. Controls across the lifecycle
The standard includes a set of reference controls. In practice they cover:
- Data: where it comes from, its quality, rights to use it, how it is prepared.
- Development and testing: requirements, evaluation before release, documented results.
- Deployment and operation: monitoring, logging, human oversight, change management.
- Information for users: what the system is, its limits, how to report problems.
- Third parties: what you require of AI suppliers and what you pass on to customers.
- Incidents: detecting, responding, learning.
You select the controls that apply and justify any you exclude, in a document usually called a statement of applicability.
6. Competence and awareness
People who build, operate and oversee AI need suitable training, and staff generally need to know the policy and their part in it.
7. Monitoring, audit and improvement
- Measure whether controls work.
- Internal audit at planned intervals.
- Management review of results.
- Corrective action when something fails.
This cycle is what makes it a management system and not a set of documents.
How certification works
In outline, as with other management-system standards:
- Build and operate the system for long enough to have evidence. Records of real assessments, reviews and audits.
- Stage 1 audit: the certification body reviews documentation and readiness.
- Stage 2 audit: it examines whether the system operates as described, with evidence and interviews.
- Certificate issued, typically valid for three years.
- Surveillance audits each year, and recertification at the end of the cycle.
Choose a certification body that is properly accredited for this standard, and check the scope written on any certificate, yours or a supplier's. A certificate covering one product line does not cover the rest.
A staged path for a mid-sized company
You do not need to begin with certification.
Stage 1: foundations, one to two months.
- Inventory of AI systems.
- A short AI policy.
- Named owners.
- Rules for staff use of AI tools.
This alone answers most of a supplier questionnaire honestly.
Stage 2: risk and controls, two to four months.
- A risk assessment method and its first application to each system.
- Impact assessment for systems affecting people.
- Evaluation before release, monitoring, incident handling.
- Supplier requirements for AI vendors.
Stage 3: run the system, three to six months.
- Operate, record, measure.
- Internal audit.
- Management review.
- Fix what the audit finds.
Stage 4: certification, if needed.
- When a customer or market requires it, engage a certification body.
Much of stages 1 and 2 can be reused from an existing information security management system: document control, risk method, internal audit, supplier management.
What it costs
- Internal time. The largest cost. Someone must own it, and each system owner contributes.
- Possibly external help to set up and to run internal audits.
- Certification fees, depending on size and scope.
- Ongoing effort. A management system is never finished.
The benefit is partly commercial, in tenders and customer trust, and partly operational: you find AI risks before a customer does.
Answering the questionnaire honestly
If you are not certified:
- Say so.
- Describe what you have: policy, inventory, risk method, oversight, incident process.
- State your plan and timeline.
- Provide evidence.
A truthful "not certified, here is our management system and our roadmap" is stronger than a vague claim of alignment. Never state certification you do not hold, and do not describe intentions as practices.
Common mistakes
- Buying templates and filing them. Auditors and customers ask for evidence of operation.
- Leaving it to one person with no authority.
- An inventory that omits the AI tools staff actually use.
- Risk assessments copied between systems.
- Ignoring third-party models embedded in your product.
- Treating the certificate as proof of product safety in marketing.
- Scope so narrow that the certificate means little to customers.
- Doing it once for a tender and letting it lapse.
We help companies build AI governance that works in daily operation and stands up to customer scrutiny, as part of AI Consulting. If a questionnaire has arrived and you are not sure how to answer it, contact us.
Summary
ISO/IEC 42001 sets requirements for an AI management system and certifies the organisation's governance, not a product. The work behind it is an inventory, a policy with named roles, risk and impact assessment, lifecycle controls, training, and a cycle of monitoring, audit and improvement. A mid-sized company can build this in stages, reuse what it has from information security, answer customer questions truthfully along the way, and seek certification when a contract requires it.
Frequently asked questions
What is ISO/IEC 42001?
It is an international standard, published in December 2023, that sets requirements for an AI management system: the policies, roles, processes and controls an organisation uses to develop, provide or use AI responsibly. It follows the same management-system structure as standards such as ISO/IEC 27001 and can be certified by an accredited body.
Does ISO/IEC 42001 certify that an AI product is safe or accurate?
No. It certifies that the organisation runs a management system for AI that meets the standard: it identifies risks, assigns responsibility, applies controls and improves over time. It does not test or approve a particular model or product, and it is not the same as compliance with any law.
Do we need certification, or is following the standard enough?
It depends on who is asking. Some customers and tenders will require a certificate. For many organisations, building the management system to the standard and being able to show the evidence satisfies customers and prepares the ground. Certification can follow when a contract or market requires it.
Related service
WhatsApp AI Chatbot
We build, train, and run a WhatsApp AI agent that answers customers 24/7, qualifies leads, and follows up.
