These days, you can hardly have a conversation without AI coming up. Artificial Intelligence—Kunstmatige Intelligentie or Artificiële Intelligentie in Dutch. The computer doesn’t “think,” but it does draw conclusions and act on them based on enormous amounts of data. Although there’s plenty of criticism and concern—especially about how far this can go—AI opens doors to a completely different way of seeing things, working, and getting support in that work.
“You can hardly have a conversation nowadays without AI coming up. Although there’s a lot of criticism and concern—especially about how far this can go—AI opens doors to a completely different way of seeing things, working, and providing support.”
The right information at the right time
When it comes to work and support, we’re immediately interested. After all, that’s the idea behind our Trevally software: ensuring that the right information reaches the right person at the right moment. That information—such as procedures, work instructions, processes, or forms—is always up to date and is only published once it’s authorized. This way, we ensure that anyone using it always sees the correct, most recent version.
AI and the Prerequisites
Looking at how Trevally works, especially considering our WebForms, Q-Learning, and Risk Management modules, there’s plenty to discuss about the application of artificial intelligence. Besides potential functionalities, there are, of course, a few conditions to consider:
- Where is my data stored?
- Will sensitive data be sent out externally?
- What does this mean for the security of my system?
These are logical and important questions, especially given the often sensitive and regulated documentation in a quality management system. Such material may be sensitive to competition or privacy concerns, so security is paramount.
Experimenting and Learning
We’re aware of these questions but are setting them aside temporarily during our research—a sort of playground where we experiment with everything AI-related in a completely separate environment. We do this consciously, because “learning by playing” is an effective way to become smarter and discover tangible possibilities. After these “lab tests,” we’ll move on to a more structured approach.
“Learning by playing is an effective way to gain wisdom and discover concrete possibilities.”
Searching with Enriched Text
For us and many of our customers, a key requirement is to be able to quickly and easily search for documents, processes, or forms within the quality management system. In the current Trevally publication, you can already conduct a “classic” search by entering search terms. However, we want to expand this by offering the ability to search via enriched text—a kind of search prompt. You ask the system a question, and based on the content of documents, processes, forms, and work instructions, you get an answer.
For now, we’ve chosen to present that answer without interpretation. Based on the query and the system’s accumulated artificial intelligence, relevant documents are found and displayed in full—without any conclusions or interpretations from the AI model itself.
“For the time being, we’ve chosen to present the answers without interpretation: the system displays the relevant documents in full, without drawing its own substantive conclusions.”
Clever Logic in Documents
Trevally documents often contain a lot of built-in logic, such as organizational units (departments, functions, groups), standard references, definitions, and relationships with other documents. This logic is included when searching, so not only is the text searched, but the underlying connections are, too. This optimizes the search results. The most relevant documents appear at the top of the list, followed by others in decreasing order of relevance.
Our Path to AI
We didn’t just start with AI yesterday. At earlier conferences, we conducted tests with speech-to-text and developed an intelligent import module. That module allows us to determine where a document fits best in the structure upon import. In this world, you can hardly call it “basic knowledge” anymore—ai, ai, ai—things move so fast! What seems ideal today might already be outdated tomorrow.
Next Steps in 2025
For us, 2025 is the year to introduce some fantastic, pragmatic functionality, keeping all data security conditions in mind. Our developers will soon wrap up their brainstorming phase on AI, and after the Spring Release, they’ll have the time and focus to convert the insights gained into real functionality—new possibilities that can then become available to our customers.
Share Your Ideas
Do you have your own experiences, questions, or ideas on this topic? Let us know; we love sharing knowledge. The beauty of sharing an idea is that both parties end up with a whole idea. But if you share a dollar, you each have only fifty cents.
“The beauty of sharing an idea is that both parties end up with a whole idea. But if you share a dollar, you each only have fifty cents left.”
PS: The AI Speaks
Naturally, we asked our artificially “thinking” computer what it thinks. ChatGPT replied as follows:
- Privacy and Security
- Sensitive information: QMS systems often contain confidential company information, such as audits and compliance documents. AI can present security risks, especially if data is processed externally.
- Regulations (GDPR, ISO, etc.): Quality management systems must meet strict regulations. AI applications must meet the same requirements, which can introduce additional complexities.
- Interpretation of Complex Documentation
- Context dependence: Quality documents often contain legal and technical terminology that AI might misinterpret.
- Different versions: Procedures change frequently, and AI might struggle to find the most current or correct version.
- Reliability and Transparency
- “Black box” effect: AI models often use complex algorithms that are difficult to decipher. That can be problematic if you have to account for how information was found.
- Error margin: AI can mistakenly flag incorrect or outdated documents as relevant, leading to errors in audits or processes.
- User-Friendliness and Adoption
- User resistance: Employees might find AI unreliable or cumbersome and prefer to search manually.
- Limited search functionality: AI must be well-trained in the specific terminology of a QMS; without proper training, results might be irrelevant.
- Costs and Implementation
- High implementation costs: Integrating AI into an existing QMS often requires a significant investment of time, money, and IT resources.
- Maintenance and updates: AI systems must be regularly updated and retrained to remain accurate, incurring additional costs.
We’re now looking for a slightly more positive “internal memory” ourselves. 😅
To be continued…
