In short:
Interest in AI for quality work is high, and for good reason. But the organisations that get real value from AI aren't the ones with the most advanced technology. They're the ones with the right foundations in place. The most important of these is a management system that brings all quality work together in one shared structure, with AI capabilities built directly into the system. In this article, we explain why, and which questions to ask when you evaluate systems.
New to the topic? Read our article on what AI in quality management actually means, or our step-by-step guide to introducing AI in your quality work.
AI works with the information it has access to. If deviations, documents, contracts and risks are spread across separate systems, inboxes and Excel sheets, or are recorded inconsistently, AI has nothing meaningful to work from. "Garbage in, garbage out" applies here in the most literal sense.
A management system that brings all quality work together in one shared structure gives AI the context and structured data it needs to deliver real value. That means more than a single case or a single document: it means the connections between them. A deviation that can be linked to a risk register, a contract and a history of similar incidents gives a completely different basis than the same deviation sitting alone in an email thread. This isn't a technical detail. It's the basic condition for AI to be useful at all.
Employees already turn to ChatGPT and similar tools with questions about their daily work, whether or not the organisation has formally decided to allow it. This pattern is often called shadow AI. An analysis by the data security company Cyberhaven, covering data movements across millions of employees, found that a significant share of what employees paste into ChatGPT is sensitive company information. So the problem isn't that employees use AI. The problem is twofold: the answers are based on general information from across the internet rather than your own governing documents, and sensitive information risks leaving your controlled environment.
The first problem is about reliability. A general AI model doesn't know how your organisation classifies deviations, what your procedures say about a specific process, or which requirements apply to your operations. At best, the answer is irrelevant. At worst, it's wrong but stated so confidently that it's hard to question. This is a well-known behaviour of language models, often called hallucination. The way around it is to have the model answer from the organisation's own documentation instead of its general training data. This technique is called retrieval-augmented generation (RAG), and it has been shown to give more specific, fact-based answers than a model that relies only on what it was trained on.
The second problem is about confidentiality. For a general AI to give a relevant answer, it needs information about the specific case: details about what happened, the context, and the parties involved. Organisations that handle sensitive information about patients, customers or employees can't enter that information into an external tool without risking it leaving their control. The risk is real, not theoretical: several large companies have banned general AI tools internally after confidential information leaked that way. The result is a tool you can give neither the right question nor the right context, so it can't help in the very cases where you need help most.
AI built directly into the management system solves this at the root. Answers are based on your own information, and the AI works within the system's existing permission model: it can only draw on the objects an employee is already allowed to see. Research notes that access following user permissions in this way is an often overlooked requirement that becomes critical in enterprise environments. Put simply, you control which knowledge provides the answers, and for whom.
The third reason is about risk and consistency, and it's the argument most often overlooked.
When an employee writes their own instructions to a general AI, the quality of those instructions determines the quality of the answer. Language models are also highly sensitive to how a question is phrased. Research shows that even small, seemingly insignificant differences in how a prompt is written can lead to large differences in the result. This means results vary from person to person. An experienced case handler who asks a precise question gets a useful answer. A colleague who phrases it more vaguely may get a misleading or incomplete answer, without knowing it.
In quality work, this is a fundamental problem. Consistency isn't a nice-to-have. It's a requirement (ISO 9001:2026).
When AI support is instead built in as predefined functions within your existing workflows, the prompt has been designed by someone with deep knowledge of both the domain and the system, once and carefully. Every employee who then uses the function gets the same well-designed basis, regardless of experience or how well they phrase things. This minimises the risk of errors and ensures that the core quality requirement of consistency is actually met. Having a person review and decide afterwards isn't a compromise. It's good practice, especially for high-stakes decisions.
If you're considering a management system or evaluating alternatives to your current one, start with the foundations, not the feature list. Five questions will take you a long way:
The best way to get answers to these questions is to ask for a demo using your own types of cases. Not a general product presentation, but a workflow that resembles yours. That's when you'll see whether the system actually delivers what it promises.
Centuri is a modern platform that brings all quality work together in one place: risk, documents, deviations, competencies, contracts and registers in one shared structure, with workflows that connect them. This gives you the data quality and context AI needs to be truly useful.
Centuri builds AI support directly into the management system, where the work already happens, and packages it as ready-made functions within your workflows, available at the click of a button. You can summarise cases and documents with one click, get draft responses and reports generated from the information you already have, and let the system highlight what's easy to miss. Answers are based on your own information, and the AI works within the platform's permission model, so each employee gets answers based on the objects they're allowed to see. The AI suggests, people review and decide, and every action takes place within the platform's traceable environment, with data stored in Sweden and no customer data used to train models.
This is also what makes Centuri a safe choice for the future. The qualities that make AI trustworthy are the same ones that have always made a management system strong: structured, governed and traceable data, with human review where it matters. With ISO/IEC 42001, an emerging standard for AI governance, this is moving from a bonus to an expectation. Quality work is moving towards a management system where governance, risk and compliance are connected, with AI as a natural and responsible support throughout.
Want to see how it works in practice, or learn how Centuri can create real value for your organisation? Contact us for a demo.