Skip to content
All articles
Audit practice 8 min read· by Lars Zimmermann

Supplier audit programme: whom do you audit first?

Nobody audits every supplier. How to select on a risk basis, who gets their turn first, and why the order decides whether the programme is of any use.

In short

A supplier audit programme prioritises by damage, not by alphabet. Four questions are decisive: what does a failure cost, how deeply is the supplier embedded in your data and systems, what role does it play in your AI value chain, and how quickly could you replace it. ISO 19011:2026 covers managing the programme in clause 5.

Auf Deutsch lesen: deutsche Fassung

The question comes up in almost every first conversation, usually in the first quarter of an hour. We have a few hundred suppliers. Where do we start? And the honest answer is uncomfortable: not at A. You start with the one whose failure would cost you the most. Everything else is busywork. I have run supplier audits in five industries and seven countries, and the difference between a programme that achieves something and one that only generates travel costs almost always lies in the order.

An audit programme is not an audit plan

These two terms are constantly confused, and the confusion costs money. The audit plan belongs to a single audit: who is interviewed when, which areas are looked at, how long it takes. The audit programme sits one level above. It comprises all the audits you run over a period, with objectives, resources, responsibilities and the risks that arise in the process.

ISO 19011 covers managing the audit programme in clause 5, from the programme objectives through the evaluation of programme risks to monitoring and improvement in 5.6 and 5.7. The standard was published in its fourth edition on 27 May 2026 and replaces the 2018 version. On the subject of remote audits a close look is worthwhile, because this is currently being reported in abbreviated form: remote audits were already in the annex in 2018. What has changed is the perspective. Clause A.16 was called “Auditing virtual activities and locations” in 2018 and is now called “Using remote auditing methods”. The object of the audit has become the method of the audit. For supplier audits this is not a side issue: when suppliers are based in other countries, the question of what you can audit reliably from a distance and what you cannot decides a considerable part of your budget.

Anyone who only plans dates has a calendar. Anyone who has a programme can justify why one supplier gets its turn this year and another does not. And you are not on your own with this: the informative Annex A of the standard has, with A.12, a clause of its own on auditing the supply chain, in both editions.

A question that ought to be settled at the start and rarely is: who owns the programme? In many companies the responsibility sits between purchasing, quality management and IT security, and therefore nowhere. Purchasing knows the contracts, quality management knows the method, IT knows the interfaces. All three are needed, but one has to lead the programme and be allowed to say no when a date is to be brought forward out of convenience. Without this person the programme turns back into a list of dates after two quarters. The standard sees it the same way: 5.4.1 governs the roles and responsibilities of those who manage the audit programme, 5.4.2 their competence.

Why the watering can fails on the arithmetic

Work it through once, and the discussion settles itself. A proper supplier audit costs you roughly one day of preparation, one day of execution and half a day of reporting and follow-up. With three hundred suppliers that is seven hundred and fifty person-days. A working year has around two hundred. So you would need almost four full-time staff who do nothing else.

Nobody approves that, and rightly so. But it means: you are selecting anyway. The only open question is whether you select deliberately or at random. In practice, at random mostly means: the supplier that has just caused trouble gets its turn, or the one whose contact person is the friendliest. Both are poor criteria.

The most common substitute criterion is purchasing volume. You take the ABC analysis from purchasing and audit the A suppliers. That is better than nothing and still wrong, because turnover and damage are two different things. The supplier with which you spend a six-figure sum every year is often interchangeable. The small service provider that handles your remote maintenance for a few hundred euros a month and has access to the control system for that purpose does not appear in any ABC analysis. It is nevertheless the greater risk. Sort by damage, not by invoice.

The four questions I sort by

I need no software for this and no evaluation matrix with twenty fields. Four questions are enough, and you can answer them for each supplier in a few minutes.

QuestionWhat you base it onStandard reference
What does a failure cost?Production standstill, contractual penalties towards your own customers, restart time.ISO 19011:2026, 5.3 and Annex A.12
How deeply is it embedded in your data and systems?Access rights, interfaces, remote maintenance, data storage. Whoever processes your data passes its security level on to you.ISO/IEC 27001, A.5.19 to A.5.22
What role does it play in your AI value chain?Does it supply the model, training data, operation or only an interface? Anyone who renames a system or substantially modifies it can become a provider themselves.ISO/IEC 42001, A.10; Art. 25 EU AI Act
How quickly could you replace it?Data formats, export options, contractual lock-in, availability of alternatives.ISO/IEC 27001, A.5.30
Four questions that produce a ranking. Whoever ranks high on two or more questions is audited first.

The third question is the one most often overlooked in mid-sized companies. Article 25 of the EU AI Act is headed “Responsibilities along the AI value chain” and governs, among other things, when a distributor, importer or deployer is itself considered the provider of a high-risk system. That can happen without anyone in the company noticing: a purchased system is passed on under the company’s own name or used for a purpose other than the one intended. Anyone who does not know their AI suppliers does not know their own role.

The four answers become a ranking, and for that you do not need a points scale with decimal places. I work with three levels per question: high, medium, low. Whoever has two highs goes into the first group and is audited on site. Whoever has one high goes into the second group, where a document review with targeted follow-up questions is enough to begin with. Everything else is monitored and reassessed in the next round. Three groups are enough to decide, and coarse enough that nobody argues about decimal places.

A tip from practice: do the classification together with purchasing and in a meeting, not by circulating a document. The interesting sentences are said in conversation. “We can’t replace that one anyway” is an answer to question four, and it is not in any form.

The groundwork at the desk saves the travel day

Before anyone gets into the car, whatever can be checked without travelling is checked. That is not a cost-cutting programme, it is method: the document review tells you whether an on-site visit is worthwhile at all, and it tells you what to aim at on site.

  • Certificate and scope, not just the certificate. More on that in a moment.
  • The data processing agreement, if personal data is involved, together with the list of subcontractors.
  • The summary of the last external audit, at least the number and type of nonconformities.
  • The deletion concept and the commitment on how data is deleted at the end of the contract and how that is evidenced.
  • For AI services: model provenance, versioning and the question of where logging takes place.

If none of this arrives, or nothing arrives for weeks, you already have your result. A supplier that cannot find its own documents within two weeks has a folder. It does not have a management system.

If the documents do arrive, read the audit summary differently from the way most people do. The number of nonconformities says little. What is interesting is where they sit and what became of them. Zero nonconformities over three years are not a good sign, but an indication that the auditing was superficial. Recurring nonconformities in the same clause are a finding, even if each individual one was formally closed: then the correction works, but the cause has not been dealt with. And if corrective actions regularly end in training, the organisation has a pattern you should know before you entrust anything to it.

The mistake I see most often

A certificate is presented, everyone is satisfied, nobody reads the scope. That is the most expensive error in the whole of supplier management.

A certificate proves that a management system meets the requirements within a defined scope. It does not prove that the very service you are buying lies within that scope.

I have seen a certificate that covered the administrative site while the service in question was provided at another location that did not appear in the scope. On paper everything was in order. Something else had been audited. That is why, for me, the question about the scope always comes before the question about the certificate.

The second common mistake is the self-assessment questionnaire as a substitute for the audit. A completed questionnaire is a claim by the supplier about itself. It is a good filter and poor evidence. Anyone who keeps it as evidence has delegated the audit to the party being audited.

With AI suppliers a third one is added, and it is new enough that many do not yet have it on their list. The usual questions aim at the state of things: which certificates, which servers, which encryption. For an AI service that is not enough, because the behaviour can change without anyone touching a contract. A model change, a new system prompt, a different version in the background, and the output is different from the one at your acceptance. So ask about the change procedure: who decides on a model change, are you informed beforehand, and can it be established afterwards which version answered on a particular day. If a provider has no answer to that, you cannot reproduce its result. And what you cannot reproduce, you cannot evidence.

How the programme stays alive

An audit programme that is set up once and then worked through is wrong after a year. ISO 19011 devotes two clauses of their own to this: 5.6 to monitoring the programme, 5.7 to reviewing and improving it. In practice that means two things.

First: after every audit the supplier’s classification is updated. Whoever delivered nothing at the document review stage moves up in priority, regardless of size. Second: new suppliers are classified before the first order goes out, not afterwards. The moment with the greatest leverage is the one at which you can still say no.

I am often asked about the rhythm, and the honest answer is: it follows from the classification, not from the calendar. First group every one to two years on site, second group annually at the desk, third group when there is a trigger. A trigger means: an incident, a change of subcontractor, a new service, a change of ownership or an AI system that suddenly decides more than before. A fixed three-year cycle for everyone looks tidy and hits the wrong ones at the wrong time.

And record why someone was not audited. That sounds like bureaucracy, but it is your cover. If something happens, the question is not whether you audited everyone. It is whether your selection was traceable.

If you want to set up your audit programme or have a single supplier audited without building up capacity yourself: that is exactly what I do, as an independent auditor on behalf of the customer.

View supplier audits

One last thought. The order is not a formality, it is the real achievement of the programme. Ten audits at the right suppliers achieve more than a hundred at the wrong ones. And you recognise the right ones not by the turnover you do with them, but by the damage they can cause.

Share: LinkedIn E-Mail

Frequently asked questions

What is the difference between an audit programme and an audit plan?+

The audit plan belongs to a single audit and governs the sequence, interviewees and times. The audit programme comprises all audits over a period, with objectives, resources and responsibilities. ISO 19011:2026 deals with managing the programme in clause 5.

By which criteria do you select which supplier is audited first?+

By four questions: what does a failure cost, how deeply is the supplier embedded in your data and systems, what role does it play in your AI value chain, and how quickly could you replace it. Whoever ranks high on several questions gets their turn first.

Is a supplier’s ISO certificate sufficient as evidence?+

No. A certificate applies to a defined scope. Whether the service you are buying lies within that scope is another matter. That is why the question about the scope comes before the question about the certificate.

Can supplier audits be carried out remotely?+

In part. ISO 19011 has dealt with remote audits since 2018; in the 2026 edition the relevant annex clause A.16 is now called “Using remote auditing methods”, previously “Auditing virtual activities and locations”. Document review and many interviews work remotely. What you have to see, such as physical access control or the reality on the shop floor, does not.

Author & expert review: Lars Zimmermann · ISO/IEC 42001 Senior Lead Auditor & Senior Lead Implementer · ISO/IEC 27001 Lead Auditor & Lead Implementer (PECB)

Last updated: 10 November 2026. Researched and reviewed to the best of our knowledge; not a substitute for individual legal advice.

Questions about your own case?

In a free 15-minute intro call we assess where you stand on ISO 42001, ISO 27001 and the EU AI Act, honestly and without a sales pitch.

Continue reading