General details
EDIHs involved
Challenges
The company Wilhelm-Klein, located in the Südwestfalen region of Germany, is a retailer and distributor of hygiene and healthcare products, among others. The company experienced significant changes during and after the COVID pandemic, with a sudden increase in demand for its products. This growth led to the hiring of additional personnel, the expansion of storage, planning, commissioning, and delivery capacities, as well as the continuous addition of new distribution and retail products to the portfolio.
This sudden and ever-increasing increase in business, together with many new employees, resulted in somewhat (in)organic and ad-hoc styled growth of work processes (often on a needs basis) as well as rapidly changing roles and work practices across various stages of the workflow, from buying, ordering, and purchasing to selling, sorting, storing, commissioning, planning, and delivering.
In addition to these internal changes, external shifts such as the emergence of new customer segments, for example, not only retail stores or companies but also private customers, schools, and nursing homes, introduced further complexity.
The changes across all three “P”s — people, processes, and products — led to multifaceted and multilayered challenges. These included duplicated work done by different employees; the existing ERP system no longer fitting the growing number of users, work processes, or product range; rigid workflows that conflicted with the need for agility; communication gaps within and between teams; and misalignment across ERP sales, order, and purchase modules. Altogether, these issues resulted in numerous workarounds, as the business still had to function within these constraints.
The most noticeable impact was the increasing number of support tickets for the support team. The issue was not only because of "no time for onboarding" but also the absence of accessible knowledge, caused by the ad hoc nature of the company’s growth.
Solutions
Following the context-driven socio-technical tradition, we began by observing and analyzing the existing work practices within their real operational environment. This analysis revealed the need for a growing knowledge infrastructure that integrates a document management system enhanced with artificial intelligence, not only for locating relevant information but also with interactive features that enable knowledge creation, curation, and retrieval, while addressing value-sensitive aspects such as trust and explainability.
To achieve this, the company’s previous file-based document management system was migrated to a cloud-based Microsoft SharePoint intranet solution. Within this new infrastructure, roles and categories were defined to ensure structured organization and accessibility.
To make the knowledge within the shared document system accessible, a demonstrator AI solution was developed featuring two competing models presented in parallel chat interfaces: a transformer-based DeBERTa model and a generative GPT-based model. Both were tested within the company for over two months for functional and interactive features, including multi-document search, contextual and non-contextual questioning, confidence scoring, answer traceability within the source documents for trust and relevant visual materials, and user interaction features such as contacting the document author for improvements.
The use was continuously analyzed with employees to refine the final solution (generative model with all the interactive features). The comparison of different AI models allowed users to understand the scope, strengths, and limitations of each system and to identify the most suitable configuration for their workflows. After iterative evaluation, the final AI-knowledge infrastructure was implemented and is now being deployed within the company’s internal environment using its own hosting resources and licensed model subscription (a setup defined collaboratively through consultation).
Results and Benefits
The analysis of the work practices resulted in an in-depth and context-rich understanding of the multi-layered and multifaceted challenges that emerged with the company’s rapid growth. It became clear that the numerous workarounds employees had developed were not merely temporary fixes to stay operational but represented systematic, experience-driven adaptations and ways to make processes simpler and more effective in practice. This was an interesting finding, as processes are typically defined at the managerial level and implemented in a top-down manner on the shop floor. However, such formal procedures often lack the flexibility and creative freedom that workers need for daily problem-solving. The analysis revealed that critical thinking and creative liberty are essential for adapting to dynamic work conditions, which were often constrained by rigid standard operating procedures (SOPs).
Another important finding concerned how employees used the knowledge embedded in documents. Apart from many documents being incomplete, inconsistent, or outdated due to constantly changing practices, workers could only read and apply the information as best as they could, without the possibility for interaction or clarification with the authors of these documents, who were, in many cases, the creators of procedural knowledge. This gap in knowledge exchange and feedback led to recurring problems: repeated questions, duplicated efforts, growing numbers of support tickets, and an increasing reliance on informal workarounds.
These findings provided crucial design implications for creating a knowledge infrastructure that not only provides access to organizational knowledge but also enables exchange, interaction, and continuous evolution. The goal was to transform the document management system into a living, growing organism where users can both consume and contribute to knowledge. To achieve this, several core features were defined:
- They can interact with the whole body of knowledge — multi-document search.
- When the answer is found, they are able to see if the answer is appropriate and valid for their problem — the confidence score, and retracing the answer in documents, seeing which document contributed to the answer, and to what extent.
- If they find the answer to be valid, they are able to apply it and do follow-up questions on the same answer within the whole knowledge base — chat with AI models, with and without context.
- If the answer is not helpful, they can give feedback, suggest new knowledge to be added to the documents for others, instead of creating workarounds — feedback loop to the author.
Together, these features led to the development of a knowledge infrastructure that is more than just a document management or AI-based system. It became an organizational ecosystem for continuous improvement, knowledge creation, and collective learning. The infrastructure embodies the fundamental socio-technical principle in knowledge sharing that knowledge is not static but co-created, maintained, and refined through active human–AI collaboration.
When this solution was tested, the evaluation indicated an expected decrease in the number of IT support tickets by over 60% and an expected increase in the engagement with knowledge and document management by 50%. It allowed employees to resolve most operational issues independently through the chatbot’s interactions. They could access verified solutions, understand the context behind them, and even propose optimized alternatives directly within the system to be embedded in the documents.
In essence, this initiative transformed knowledge management from a passive repository into an active socio-technical infrastructure that not only engages and empowers employees across the organization but also formalizes knowledge and processes.
Perceived social/economic impact
As the company grows and plans to expand further through multiple innovative endeavors, this project establishes a formalized process for knowledge creation, dissemination, and curation as an infrastructure, not only for onboarding but also for day-to-day problem-solving and mindful work practices. Economically, it allows constrained personnel and time resources to be better managed through AI interactions with human-in-the-loop and collaborative feedback mechanisms. Socially, it provides employees at all levels with a sense of engagement and confidence to act with creative liberty at work, while enabling the optimization of processes that were previously seen as workarounds to be formalized and incentivized.
DMA score and results - Stage 0
At DMA0, the company had 39% digital readiness but was developing a digital business strategy, which was shown as 83%. They were focused on human-centered digitalization (assessed as 68%), but were comparatively low on data governance and automation, and AI (70% and 40% respectively) at that time.
DMA score and results – Stage 1
During and through the initial workshops, train-the-trainer events, and the AI-enabled knowledge infrastructure project, at DMA1, their digital readiness increased to 48% and human-centered digitalization initiatives to 87%. They reduced increasing their digital business strategy and the decrease was shown to be at 60% because they wanted to focus on completing the initiatives they had already started and focus more on making them human and practice-centered. On a positive note, their data governance and automation and AI increased substantially through the projects (92% and 80% respectively). This shift is a direct indicator of their involvement with EDIH-SWF and the focus on knowledge infrastructure during that time.
Lessons learned
Understanding the work context and actual work practices is immensely important in order not to develop something from ideation that might be fancy and trendy but not aligned with what users actually need for their tasks and assistance.
Generating literacy about the technology implicated through contextual studies is crucial. This can be effectively done during engagement with companies by providing tangible solutions, allowing them to compare and contrast different technologies.
Testing with managers alone is not fruitful if the application is intended for use across the entire organization. Infrastructure and infrastructuring concepts become extremely important to define the reach, scope, conventions of practice, and agency of the different users within the technological ecosystem.
When prototyping and later deploying, overcommitting should be avoided at all costs. It should be made clear what we do as an innovation hub and what is expected from them. For example, after testing the demo, we let them select features and then consulted the company on what kind of hosting is needed and what type of AI model subscription should be purchased to fit the technologies into their existing infrastructure.
When conducting contextual research, even topics that seem non-sensitive may reveal insights about employees’ work patterns that could potentially harm them, create distrust, or provoke negative emotions in the workplace. This must be handled with utmost sensitivity and ethical consideration. Ensuring employees understand that we aim to give them a voice and engage them in the process helps build trust. Additionally, formalities such as consent declarations, maintaining anonymity, and offering the option to remove anything they said or showed from analysis are critical.
The implications of technology must not be directed at individuals but should be presented as ways to improve processes. This approach minimizes the risk of stigmatization and victimization in the workplace.
Need support?
Consult our catalogue to locate the Eupopean Digital Innovation Hub nearest to you and accelerate your company's digital transformation.