How We’re Building AI-Based Enterprise Knowledge Management that is Secure, Structured, and GDPR-Compliant

First and foremost, a heartfelt thank you to the organisers of InsurNXT for orchestrating such an insightful and inspiring event. The quality of conversations, the energy of innovation, and the spirit of collaboration were genuinely remarkable. I left feeling deeply appreciative and thoroughly energised by the experience.

1. Insights on AI – A Mirror to Our Potential

Throughout the event, it was fascinating to observe how pervasive discussions about AI were but also how wide-spread real-word implementation at least in the insurance sector were far and wide between For me as decision-maker and knowledge worker, AI is more than an advanced search engine or a way to automate repetitive processes (though it excels at those too!). I see it as a kind of mirror, reflecting and clarifying our own implicit knowledge and ideas, helping to give them structure and context, and making them more easily digestible by others. Perhaps it’s like a translator, helping to bring complex concepts that might be swirling in one's mind into a form that is more easily understood and relevant to others. The true power of AI for users like me, I believe, lies in its potential to transform how we work—especially when applied to AI-based enterprise knowledge management—by making implicit insights explicit, structured, and actionable.

Having spent many years speaking with experts, decision-makers, and implementers, listening, learning, and refining my own understanding, I recognise the common challenge we all face in clearly expressing our implicit knowledge. AI, particularly through voice-to-text technologies and iterative engagement, has become invaluable in helping turn unstructured thoughts into tangible, actionable content. 

At KASKO, we’re putting this into practice across a growing set of internal use cases:

• Converting unstructured voice messages into internal memos 

• Generating sales outreach sequences

• Parsing unstructured inputs like emails

• Responding to RFPs

• Experimenting with LLMs to speed up application and product delivery

• Selectively using LLM-based coding tools

Moreover, we also roll out AI across the insurance value chain for users of our insurtech platform, such as: 

• Reviewing uploaded customer documents

• Supporting claims condition assessment

• Comparing policy data

Some of these are experimental, but many are already helping us improve daily work. This iterative process helps individuals, and organisations, clarify and enhance collective understanding, bridging critical knowledge gaps or simply automating repetitive processes.

2. How We’re Building AI-Based Enterprise Knowledge Management That Works

I've taken this insight and have started to embed this approach into our knowledge sharing and generation processes. As anyone requesting change, I feel it’s important to champion these tools myself, and I’ve personally found AI to be an incredibly powerful daily partner in documenting, structuring and communicating information.

Whenever information needs to be shared, we aim to codify it (both the content and the process of codification) and expand the corpus of knowledge we are generating. Our guiding principle is simple: we want to answer questions thoroughly once, so that knowledge is preserved and easily accessible.

This is currently in an MVP stage and it’s yielding positive results, though it requires dedicated effort. 

3. From MVP to Product: Embedding AI into a GDPR-Compliant Platform

What enables this is not just the AI, but our modular, GDPR-compliant platform. It provides the structured environment to automate emails, generate documents, store data with version control, and handle inputs securely, making these AI use cases actually workable at scale. We’ve also built in features like role-based access control, audit trails, and permission management to ensure governance is handled by design, not as an afterthought.

What makes them truly effective isn’t the AI layer alone, but the foundation beneath it: email triggers, document generation, versioning, and secure data storage. Because our platform is modular, clients can adopt only the components they need—like document generation, chatbot logic, vector search, or storage—without being locked into a full-suite solution. These features may not be exciting on their own, but they’re what make AI usable in real workflows.

The next stage is to productise this process. 

Here’s what we are currently building:

• We already have an internal chatbot linked to multiple LLMs that we host in our secure GDPR compliant infrastructure as part of our efforts to build a scalable, GDPR-compliant system for AI-based enterprise knowledge management. Because we use hosted infrastructure and modular API connections, we’re able to upgrade underlying AI models quickly andwithout waiting for internal IT to catch up or revalidate (or train) every system change.

• We are looking to integrate this into Slack (it can also work with Teams).

• The goal is for any expert, not just a few individuals, to add content they don’t want to repeatedly explain by simply adding files, texts, and emails.

• This will then be automatically added to a human-readable database for quick review of base data and vectorised for easier retrieval by the LLMs.

• Now, whenever another team member asks a question, the corpus will be queried. If we have the answer, it will be provided. If not, it will either a) suggest an AI-generated answer for further use (which is then forwarded to an expert to review, refine, and add to the Corpus) or b) request an answer from the expert (who will receive a suggestion and can finalise with AI and add to the Corpus).

While this system shows a lot of potential, we’re realistic about the challenges involved in making it truly work at scale. We’ve also learned that managing different vendors for each AI use case (especially when personal or customer data is involved) can become a compliance and operational nightmare. That’s why we’ve centralised AI execution on our own platform: to scale securely, reduce governance complexity, and avoid fragmented, high-risk setups.That said, building AI agents that perform reliably in production has proven more complex and time-consuming than expected. So while the concept is solid, implementation is still evolving.One open question we’re still exploring is how to reduce model hallucination, even with high-quality internal data. The current system may still require expert validation in many cases—something we are actively testing during this MVP phase.

Of course, there are many other challenges to address: from overconfidence to nuance, time-sensitivity, governance, cost, and bulk processing. We have roadmaps for these, but that’s a conversation for another article.

This structured yet flexible approach aims to ensure that our most valuable resource, our collective knowledge, is not lost but rather continuously enriched and easily leveraged. This isn’t about testing AI features in isolation. We’re intentionally building AI-based enterprise knowledge management into the core of our platform architecture and way of working together.

4. Making AI Useful for Our Clients – From Outreach to Execution

While the excitement around generative AI is real, we’ve found that it only delivers lasting value when paired with reliable infrastructure and being willing and curious to change the way we work together. That’s why our focus has always been on combining AI with deterministic platform capabilities: workflows, triggers, templates, document logic, and compliance rails plus the clear expectation of trial and error and it being ok to go slow now to go faster later. We don’t expect any miracles but appreciate it’s hard work and smart to leverage what already works

Beyond internal processes, we’re using AI (embedded within our platform for AI-based enterprise knowledge management) to refine how we communicate and support clients  (Tech, InsurTech Sparring, EU Market Access, Capacity Search & Recruitment). For example, AI can help triage incoming service requests, route claims to the right team, and surface relevant product information to customer-facing staff—all without human intervention. Understanding that every stakeholder, from heads of motor insurance to marketing leaders, commercial brokers, or specialists, has unique needs, we've worked on developing targeted AI-driven outreach messages, then custom use-cases, ROI calculations, and board memos customised for the likely KPIs of each relevant board member (or other critical person) for an initiative.

If you've recently received outreach from us, it was likely crafted through this process. Each message aimed to provide two or three relevant insights based on our extensive understanding. While we acknowledge these communications may not always be perfectly tailored, the intention was sincere, to provoke thought, spark dialogue, and add immediate value. If the messages we shared with you felt broad or missed their mark, please let us know! We are genuinely committed to improving, to listening, and to better aligning our conversations with your specific interests and needs.

5. Doing Cool Things That Matter – Our Way Forward

During the event, an exchange with another startup founder (someone I greatly respect) stood out. When asked, "What do you really want to do?" my instinctive reply was simply, "I want to do cool things." This didn't quite resonate with him at that moment, as he suggested I needed sharper focus on a platform vision he recalled from years ago.However, I’ve always believed in going broad, and that in today’s world, especially with AI, breadth isn’t a weakness. It’s a different kind of speciality. That actually aligns with how I think about our work: connecting dots across domains, people, and ideas to build things that wouldn't emerge from a narrow focus alone, and that’s also how Marc Andreessen a well known American entrepreneur, software engineer, and venture capitalist, who played a pivotal role in the development of the internet, sees things.

This conversation was clarifying. It reinforced for me that while established paths have merit, exploring approaches that genuinely resonate with our team’s strengths and vision,even if unconventional,is vital for fostering genuine innovation. Years of focusing on trying to really understand what sits in the way of my (potential) partners and letting us help them improve and grow have become central to our approach. I've never felt clearer about our direction: to strive to be genuinely insightful, to deeply understand your business, and to implement use-cases as if they were our own —or to do cool, meaningful work together that you might not otherwise pursue.

6. Let’s Talk – Building AI-Based Knowledge Systems Together

This is my invitation to you,whether we've spoken before, whether our outreach has resonated, or whether it's the first time you're hearing from us:

You might already have a platform in place, or perhaps you're just starting to explore AI applications—either way, our experience shows that you don’t need to start from scratch. Using what you already have (and embedding AI where it makes real operational sense) can be a practical and scalable path forward. Let’s talk openly about use-cases, especially around AI-based enterprise knowledge management, knowledge sharing, and building systems that actually work. . Together, we can explore opportunities that not only aim to generate positive ROI within the next year but also prove interesting and valuable in the longer term. I am confident that a conversation with me and my team will be a worthwhile investment of your time.

Our aspiration is to become one of the most intriguing and insightful technology partners you've worked with. Let's collaborate openly, build trust through genuine usefulness, and harness the transformative power of AI and common sense to unlock extraordinary potential together.

Reach out anytime, we are looking forward to the conversation.