A provider of a system that suggests treatment urgency in emergency departments has exactly one person who can evaluate post-market monitoring and understands what the figures mean clinically. She moves to another employer. A working student takes over the evaluations, keeps the reports neatly up to date and cannot interpret the deviations. At the same time, the same small group writes the technical documentation for a second system. The processes depend on people, and those are exactly what is missing.
EN 18286 requires providers, in clause 6.1.3, to consider the availability of resources and competence when they deal with risks to the quality management system (QMS). For this, the standard refers to its clauses 7.1 (resources) and 7.2 (competence). The risk: there are not enough people with the necessary knowledge for testing, monitoring or documentation.
What is behind it?
Every process in a QMS needs someone who carries it out and understands what they are doing. A test report is only as good as the person who decides which cases are tested and what a result means. EN 18286 explicitly counts staff with their competences, application and technology knowledge, infrastructure for design, development and testing, and security of supply among the resources (clause 7.1).
What the shortage looks like
- Knowledge in one head. A single person knows the monitoring, the test procedure or the regulatory requirements. If they drop out, the process stops.
- Competing tasks. The same people build new features and run the QMS. Under deadline pressure, documentation loses.
- Gaps in expertise. Knowledge is missing about the AI Act, the field of use (such as medicine or employment law), test methods for the technology in use, or bias and its effect on fundamental rights.
- Too few for independent review. In small teams it is hard to avoid people checking their own work.
- Missing infrastructure. Computing power or test environments are not enough for a complete test run.
Which competence is needed
EN 18286 requires the provider to determine the necessary competences, ensure them through education, training or experience, close gaps and document all of this (clause 7.2). According to the standard, how demanding the requirements are depends among other things on the intended purpose and foreseeable misuse, on technology and data, on the risks to affected persons and on accessibility for different user groups.
Impact
Missing resources almost always lead to QMS-01: tests are shortened, evaluations postponed, documents stay incomplete. More dangerous is the case in which the process formally runs and the person does not recognize the signals. Then signs of bias, declining performance or new risks get lost, and a serious incident may be misjudged. The harm first hits the people the system decides about, then the provider, whose evidence does not hold up when an authority or notified body examines it.
Practical examples
Loss of the key person. The example at the start shows the most common pattern: monitoring goes on, interpretation is missing. A gradual rise in cases rated too low among older patients appears in the reports and is noticed only months later.
Small team, no independent review. A provider with six staff develops an AI system that matches applicants to apprenticeship places. Verification and release lie with the same two people. The problem is solved with an external reviewer who is bound by contract and reads every release against the evidence.
Computing budget used up. A provider of speech analysis for lending decisions has spent its computing budget on training. The full test run for the new version no longer fits, so a shortened test runs. Nobody changed the test plan for this.
Effective measures
The AI Provider Policy from the AI Compliance Starter Kit addresses this risk in these places:
- Section 8 (Life cycle processes): For each phase, the internal and external resources needed are determined: people with their expertise, computing power, tools and models, and data.
- Section 9 (Conception): Before design starts, the provider checks whether data, models, computing power and expertise are available for the system.
- Section 22 (Control of documented information): A changed document is reviewed and approved by a person with the competence for the aspect concerned.
- Section 23 (Resources and security of supply): Resources are assessed for every new provider system and at the annual review of the policy.
- Section 24 (Roles and responsibilities): The allocation sets out who is responsible and who is involved, competence included. Reviewers do not review their own work, and every change of assignment is recorded with a date.
These ISO/IEC 42001 controls contribute as well:
- A.4 Resources for AI systems: Human resources are documented per life cycle phase, with the expertise each phase needs.
- A.3 Internal organization: Roles and responsibilities are defined and assigned as needed, so that no area is left without someone in charge.
The management review under EN 18286 ends with decisions on resource needs (clause 10.1). That is where the signs of bottlenecks belong.
Link to the AI Act
Article 17(1) of Regulation (EU) 2024/1689 lists, among the minimum content of the QMS, resource management including measures related to security of supply (point (l)) and an accountability framework setting out the responsibilities of management and other staff for all aspects (point (m)). Under paragraph 2 the extent follows the size of the organization, while the strictness and protection that the conformity of the high-risk systems demands remain in place. A small provider may therefore work lean and still has to bring in the necessary competence. For the reach of the presumption of conformity when applying the standard, see EN 18286: Quality Management System for AI Providers.
Related QMS risks
- QMS-01 The QMS misses its intended outcomes: Missing people are the most common cause of failed processes.
- QMS-02 Undesired effects of the QMS: An overloaded QMS makes every bottleneck worse.
- QMS-04 AI technologies in use overstretch the processes: New technology calls for knowledge that the team often does not yet have.
Sources
- EN 18286:2026, clauses 6.1.3 d, 7.1 (resources), 7.2 (competence) and 10.1 (management review)
- Regulation (EU) 2024/1689 (AI Act), Article 17(1)(l) and (m), Article 17(2)
- AI Provider Policy, Sections 8, 9, 22, 23 and 24
- EN 18286: Quality Management System for AI Providers in the Cenedril wiki