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

Certified, and not a single process has indicators

ISO 9001 clause 4.4.1 requires monitoring, measurement and performance indicators per process. Why this slips through for years and what it costs AI projects.

In short

ISO 9001 clause 4.4.1 requires monitoring, measurements and performance indicators for the processes, as far as needed for control: item c) in the 2015 edition and d) in the 2026 edition. It often slips through for years because a process map does not show it. In AI projects that is expensive: without a measured baseline, no benefit can be shown.

Auf Deutsch lesen: deutsche Fassung

A company has been certified to ISO 9001 for years. A tidy process map, clean documentation, never a major nonconformity. Then preparation for an IATF certification begins, someone goes through the processes and asks a simple question: where are the indicators, actually?

There were none. Not for a single process. Unnoticed for years, in no surveillance audit.

This is not an isolated case, nor is it a reproach to the company. It is probably the most frequently overlooked requirement of the whole standard, and it is becoming expensive right now, because without it no AI adoption can be evidenced.

What 4.4.1 really requires

I follow the ISO 9001:2015 edition here, under which existing certificates remain valid during the transition period; what shifts in the 2026 edition is stated directly below the list. In the 2015 edition the clause is titled Quality management system and its processes. It first requires the organisation to determine the processes needed and their application. Then comes a list of eight items, a) to h). This is the part that most process maps only reflect by half.

  • a) determine the inputs required and the outputs expected.
  • b) determine the sequence and interaction of the processes.
  • c) determine and apply the criteria and methods, including monitoring, measurements and related performance indicators, to ensure effective operation and control.
  • d) determine the resources needed and ensure their availability.
  • e) assign responsibilities and authorities.
  • f) address risks and opportunities in accordance with 6.1.
  • g) evaluate the processes and implement changes so that they achieve their intended results.
  • h) improve the processes and the management system.

In the new edition ISO 9001:2026, published in September 2026, the clause is titled simply Quality management system, and determining the processes needed has moved into the list as a separate item a). As a result everything shifts by one letter: c) becomes d), g) becomes h), and the list runs to i). The wording on monitoring, measurements and performance indicators is unchanged. The letters in this article follow the 2015 edition.

Item c) is the hard one. Monitoring, measurements and performance indicators are there word for word in the text of the standard, and they are not there as a recommendation. The purpose is explicitly named: effective operation and control. A process without a measure can neither be evidenced as effective nor controlled. It can only be described.

Added to this is item g), which hangs in the air without c). Evaluate the processes and implement changes so that they achieve their intended results. Evaluate against what? Without a number, all that remains is the opinion of the process owner, and that rarely turns out negative about their own process.

The clause also does not stand alone. It is the first link in a chain of three places that support each other. 4.4.1 c) determines the indicators. Clause 9.1.1 then requires four things to be determined: what needs to be monitored and measured, by which methods, when measurement takes place and when the results are to be analysed and evaluated. And 9.3.2 c), in the 2026 edition 9.3.2 d), explicitly counts process performance among the inputs to management review.

Whoever leaves out c) breaks all three. Management review is then presented, under the heading process performance, with a page of revenue figures, and top management evaluates the effectiveness of a system about whose processes it learns nothing. That is the real damage, and it is greater than the audit finding.

Incidentally, the work pays off twice. Clause 9.1 of ISO/IEC 42001 is identical in structure to 9.1.1 of ISO 9001: the same four determinations, the same duty to keep documented information as evidence, the same evaluation of performance and effectiveness. Whoever sets this up cleanly once for 9001 already has it in place for a future AI management system.

Why this slips through certification audits

There are four sober reasons for this, and none of them is negligence on the part of the certification body.

  • A certification audit works with samples across the whole cycle. Clause 4 is short, appears formal and comes at the beginning, when time still seems plentiful.
  • There are numbers, after all. Delivery reliability, complaint rate, revenue. But those are company indicators and not process indicators. They say nothing about whether the quotation process works.
  • The process map looks complete. Boxes, arrows, owners. Item c) is the only one of the eight that you cannot see in a map.
  • The question is often about the process, not about its control. Whoever answers how the workflow goes has apparently answered the question.

With IATF it then stands out immediately, because there the measurement of processes is pursued much more rigorously. That is why so many companies receive this finding precisely during IATF preparation, after years with a clean 9001 certificate.

A process map shows how things are supposed to run. An indicator shows whether they run. Most organisations have the first and take it for the second.

What the company from the opening then did is unspectacular and therefore useful. Four core processes selected, two numbers defined for each, both from data that the ERP was keeping anyway. Eight weeks later the first evaluation was available, and it brought to light something nobody had suspected: the slowest step was not production, but the internal approval of quotations above EUR 10,000. Four days on average, three of which were pure waiting time. Nobody had noticed this before, because delivery reliability towards the customer was still fine.

What this has to do with AI

Now the old formal defect becomes a current problem. Every AI adoption at some point faces the same question from the management: what did it achieve?

This question can only be answered if there is a value from before. A measured one, not an estimated one. Whoever only starts measuring after the introduction has a number and no comparison. Whoever measured nothing beforehand can claim any benefit afterwards and evidence none.

In my own company I can say that order entry has dropped from twelve minutes to 26 seconds and the process cost per transaction from EUR 5.14 to EUR 0.06. The reason is unspectacular: the twelve minutes and the EUR 5.14 had been measured beforehand. Without this baseline the whole calculation would be a story.

For an AI management system this is explicitly regulated. ISO/IEC 42001 deliberately keeps its clause 4.4 short, a single sentence on establishing, implementing, maintaining, improving and documenting. Measurement sits in its clause 9.1, and there four things are to be determined: what needs to be monitored and measured, by which methods, when measurement takes place and when the results are to be analysed and evaluated. Plus documented information as evidence of the results and the evaluation of the performance and effectiveness of the system.

So whoever has never measured their processes has to build these four determinations from scratch for 42001. Whoever took 4.4.1 c) seriously copies them across.

A final point on form. Directly after the lettered list comes 4.4.2, which requires two things to the extent necessary: maintain documented information to support the operation of the processes, and retain documented information to have confidence that the processes are being carried out as planned. The 2026 edition speaks for both only of documented information that must be available, the split into a) and b) remains. The measured values are exactly this second kind. An indicator that is determined monthly and recorded nowhere fulfils c) and fails again at 4.4.2 b).

Three indicators that are immediately available in almost every company

The most common objection is that there is no time for this. In most cases time is not what is missing, because the data has long been in the ERP and simply nobody has assigned it to a process.

  • Lead time per transaction. Timestamps from receipt to completion, per process. Available in every system that knows transactions.
  • Rework or error rate. How often does a result of this process have to be corrected before things move on. This is the most meaningful number of all and almost never collected.
  • On-time performance per process step. Not delivery reliability towards the customer at the end, but adherence to internal commitments in between. That is where the delay arises that the customer sees later.

Three numbers, monthly, per process. That fulfils c) and incidentally provides the baseline for any later automation. What matters is the assignment: the number belongs to the process and gets an owner, otherwise it is a statistic and not control.

And now the flip side, because it goes just as wrong in the other direction. The standard requires the criteria and methods that are needed. Not all conceivable ones. Whoever plasters forty processes with three indicators each produces a hundred and twenty numbers that nobody reads, and ends up measuring less than before. Start with the three to five processes on which your business really depends. For the others, a reasoned decision on why no separate measurement is needed there is entirely sufficient and easy to defend in the audit.

A thought experiment for the next audit date. The auditor says: show me that your complaint handling process is effective. Without a measure, three routes remain, and all three lead into a conversation about impressions: you show the procedure, meaning how it is intended. You show completed cases, meaning that work was done. Or you explain that it is going well. None of these routes answers the question of effectiveness. With a number, the same answer takes twenty seconds.

What you can do this week

Take your process map and write a single column behind each process: how do we see that this process works? Not how it runs. How you can see it.

If more than half of the rows remain empty, you have the same finding as the company above, only without an auditor in the building. That is the cheapest way of getting it.

Two answers to the same audit question, to make the difference visible:

  • Weak: our processes are described in the process map, owners are named, and we monitor delivery reliability.
  • Robust: for the four core processes we keep one lead time and one rework rate each, monthly, with an owner. Quotation preparation averaged eleven days with 14 per cent rework. After the change in June it is six days with 9 per cent. For the remaining processes we have given reasons why a separate measurement is currently not needed.

The second answer fulfils 4.4.1 c) and g) in one sentence. It also ends the discussion about whether the change achieved anything, because the answer consists of numbers and not of impressions. You will need exactly these numbers again as soon as AI moves into the process.

Share: LinkedIn E-Mail

Frequently asked questions

Does ISO 9001 require an indicator for every process?+

Not for every one. Clause 4.4.1 c), d) in the 2026 edition, requires the criteria and methods, including monitoring, measurements and related performance indicators, that are needed to ensure the effective operation and control of the processes. The word needed gives you the leeway. For core processes a measure is practically always necessary, for support processes often not. What matters is that the selection is reasoned and documented.

Is a company indicator such as delivery reliability enough?+

Not as the only number. Delivery reliability is the result of many processes and says nothing about which of them is stuck. The standard requires the indicators for the control of the processes, that is, at process level. In practice you need at least one quantity that can be assigned to a single process and an owner. The company indicator remains useful alongside, but it does not replace the process indicator.

Why does this often go unnoticed in the certification audit for years?+

Because a certification audit works with samples across the cycle and clause 4 appears formal. In addition, a complete process map gives the impression that the requirement is fulfilled. Item c), d) in the 2026 edition, is the only one in the list that you cannot see in a map. In preparation for IATF 16949, by contrast, it regularly stands out immediately, because process measurement is pursued much more rigorously there.

What do I need before I build AI into a process?+

A measured baseline, and before the change. Without it you cannot evidence any benefit afterwards, only claim it. Two quantities are usually enough: lead time and rework rate, collected over at least four weeks. For an AI management system to ISO/IEC 42001 you need this anyway: its clause 9.1 requires you to determine what is measured, by which methods, when measurement and when evaluation take place, plus documented information as evidence.

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