Services · Knowledge Management
Employees leave the company. Providers change. Systems get old. Knowledge is lost. We build knowledge structures that prevent this — from the SharePoint platform to an AI knowledge assistant with your own documents.
Management · Owner
When knowledge holders leave, providers change, or in-house developments are replaced, it becomes clear whether your knowledge is structured — or only lives in people's heads. We help structure knowledge in time, before it becomes a critical gap. Investment today, risk reduction tomorrow.
Department head · Business unit
A structured SharePoint knowledge platform with information architecture, knowledge map per business unit, metadata concept, and governance. Includes a migration plan from file shares and old intranets, plus an adoption program for the teams. Ready for steering committee and HR development.
IT leadership · CIO · Solution Architect
AI knowledge assistants with EU hosting on Mistral or Aleph Alpha instead of OpenAI. Vector DB setup (pgvector, Azure AI Search, Qdrant), embedding strategy, chunking concept, re-ranking, source filtering, permission trimming on SharePoint ACLs. GDPR-compliant, audit-ready.
For Managing Directors · strategic business insurance
Baby boomers retire, providers change, in-house developments are replaced. Each of these events reveals whether knowledge is structured — or only lives in people's heads. We build your knowledge platform now, while the carriers are still here. That is cheaper than any post-departure knowledge recovery and is the strategic insurance against the most expensive of all losses: 20 to 30 years of experience that disappear irretrievably.
For department heads · SharePoint platform with knowledge map
You get a SharePoint-based knowledge platform with a documented knowledge map per business unit, clear information architecture, metadata concept, and governance model. Includes a migration plan from file shares, old intranets, and SharePoint 2013/2016 inventories, plus a structured adoption program for the teams. Ready for steering committee, with measurable KPIs for HR development and compliance requirements.
For IT leadership · RAG architecture with EU LLMs
RAG architecture on EU hosting: Mistral (Mistral Cloud, EU region) or Aleph Alpha as the LLM, vector DB setup with pgvector, Azure AI Search, or Qdrant. Embedding strategy with multilingual models, chunking concept, re-ranking pipeline, source filtering, and permission trimming on SharePoint ACLs. GDPR-compliant audit logs, prompt templates, hallucination protection. An alternative to or complement of M365 Copilot — depending on compliance requirements.
Why now
In the coming years, the baby-boomer generation leaves companies. At the same time, providers change more often, in-house developments are replaced, and Microsoft 365 Copilot is rolled out — all triggers that reveal whether your knowledge is structured or only lives in people's heads.
Knowledge holders retire. With them, 20–30 years of experience disappear, often without any structures or documentation having been created. Those who don't start now will have no chance in three years to capture the knowledge.
Anyone switching ERP, CRM, or custom-software providers notices: the knowledge was with external heads, not in the company. Knowledge Recovery before the switch costs a fraction of what a rebuild without documentation costs.
Microsoft 365 Copilot is only as good as the knowledge sources it can access. Unstructured drives, incorrect permissions, and outdated documents lead to bad answers — and to rollout stops. Knowledge management is the prerequisite for AI adoption.
Four service building blocks
The four building blocks stand on their own — and build logically on each other. You can start with SharePoint, continue with Copilot, complement with a RAG assistant, or enter acutely with Knowledge Recovery.
A well-designed SharePoint solution with 5–10 knowledge areas, clear information architecture, search strategy, and permission concept — the foundation for any later Copilot or AI initiative. Including training for the owners.
Setup of Copilot Connectors for external sources, Sensitivity Labels via Microsoft Purview, and quality tuning through test prompts and answer scoring. Linked with our Microsoft 365 Copilot adoption program.
Platform-independent document-chat solution with EU-hosted LLMs (Mistral, Aleph Alpha), vector DB setup, embedding strategy, and quality metrics. For all cases where Copilot hits license, hosting, or source limits.
Interviews with knowledge holders, reverse engineering of existing systems, and process mining from log data — the result is structured, handover-ready knowledge documentation. Trigger service in case of staff departure, provider switch, or system replacement.
Block 01 — deep dive
SharePoint has been Microsoft's knowledge platform for nearly two decades — and in 80% of mid-market tenants, it is a graveyard of team sites without concept. We have been building SharePoint since 2007 and know where it fails: at the architecture, not the technology.
A flat information architecture with 5–10 clearly delineated knowledge areas. Per area, we define owners, lifecycle rules (what stays? what is archived? what is deleted?), metadata, and a search strategy. The permission concept is set at the area level — not at the document level, because in practice that is not maintainable.
SharePoint as a technology works. What goes wrong: every department builds its own team sites, no one feels responsible for knowledge areas, permissions are a tangle, and search delivers junk. We rebuild this in 6–10 weeks — structured, documented, Copilot-ready.
Microsoft 365 Copilot accesses SharePoint, Teams, OneDrive, and Exchange — and respects their permissions. If permissions are broken, Copilot sees what it should not see. Or Copilot fails to find content that it should find. A clean SharePoint platform is the prerequisite for any serious Copilot rollout.
Practical observation: In nine out of ten tenants we look at before a Copilot rollout, there is at least one knowledge area whose permissions are set so incorrectly that Copilot would expose management or HR content to regular employees. Fixing it before the rollout is mandatory.
Block 02 — deep dive
Copilot "out of the box" sees Microsoft 365. That is enough for email summaries and Teams notes — not for real knowledge answers from your company. We integrate external sources and tune answer quality until Copilot becomes sustainable.
Over the past 18 months, Microsoft has built a connector architecture that integrates external knowledge sources into the Copilot answer — Confluence, file shares, SQL databases, industry databases, ServiceNow, Salesforce. The setup is not trivial: authentication, indexing, permission mapping, and quality checks are standalone work packages.
Confidential areas (HR, M&A, management, client data) must be explicitly excluded from Copilot. Microsoft Purview Sensitivity Labels allow this at the document level — provided they are assigned correctly. We jointly set up the labeling strategy and automatic classification.
After the setup comes what is missing in Microsoft marketing: quality tuning. Test prompts against real knowledge questions, answer scoring by domain owners, adjustment of indexing frequency and connector configuration. Only then does Copilot have a chance of being adopted in daily use.
If you want to deploy Copilot not just as a knowledge tool but strategically for AI adoption, see our page AI & Microsoft 365 Copilot Adoption — we describe the full adoption program there, with persona training, use-case identification, and ROI measurement.
Block 03 — deep dive
When Microsoft Copilot doesn't fit — because sources sit outside M365, because EU hosting is mandatory, or because user licensing doesn't work economically — we build a dedicated RAG knowledge assistant. Document chat with your content, on European infrastructure.
Retrieval-Augmented Generation (RAG) is the established architecture for AI knowledge assistants: your documents are indexed in a vector database, an embedding model creates semantic representations, and a large language model formulates the answer based on the retrieved evidence. Unlike "bare" LLMs, a RAG system answers on your data — with source citation and without hallucinating into the blue.
We rely on Mistral and Aleph Alpha as European LLM providers with EU hosting and clear GDPR commitments. Over the past 24 months, both providers have largely closed the quality gap to OpenAI — for German-language knowledge answers, they are on par.
We use Qdrant, Weaviate, or pgvector (PostgreSQL extension), depending on the requirement. Embedding strategy, chunking, re-ranking, and index partitioning per permission boundary are part of the pilot package.
A RAG assistant can integrate sources that Microsoft Copilot cannot reach: file shares, Confluence, older SQL databases, scanned PDFs (with OCR), industry databases. That makes it the choice when your knowledge is heterogeneously distributed.
Every RAG pilot at arades delivers metrics from day one: recall (which relevant documents are found?), precision (how many of the found documents are actually relevant?), and hallucination rate (how often does the LLM formulate answers without evidence?). Without these metrics, a RAG assistant remains a feeling — we make it measurable.
Block 04 — deep dive
Sometimes the question is not "how do we structure our knowledge?" but "how do we save knowledge that is currently being lost?". The Knowledge Recovery sprint is the acute trigger service before staff departure, provider switch, or system replacement.
Structured 60–90-minute interviews with people who carry critical knowledge. We work with an interview guide that does not just ask "what do you do?" but also "why do you do it that way?", "what goes wrong if someone does it differently?", and "what edge cases do you alone know?". The answers are documented in a structured way.
Where in-house developments or older standard systems run without documentation, we do reverse engineering: database schema analysis, custom-code review, workflow reconstruction. The result is technical documentation that is viable for replacement, further development, or migration.
When actual processes deviate from the documented target description — which they almost always do — process mining helps: from log data in ERP, CRM, or workflow engines, we reconstruct how the process really runs. That is the foundation for knowledge documentation that is true in daily use.
The result is not a Word document in a cabinet but a structured knowledge base — typically in the SharePoint knowledge area, optionally indexed as a RAG source. With owners, update cadence, and a search strategy.
When does knowledge management turn into a build engagement? When the recovery makes it clear that the old system is no longer viable, we hand over to our engineering team — custom software development with clear architecture, documented code, and a handover plan from sprint 1.
Cross-references
Knowledge management is rarely an isolated service. Here are the most important bridges to our other offerings.
RAG strategy, Microsoft Copilot adoption, persona training, and use-case identification. The strategic frame around knowledge integration with Copilot.
To the AI landing page →
Trigger serviceAcute trigger service for staff departure, provider switch, or system replacement. Deeper methodology, faster delivery, dedicated sprint packages.
To the recovery page →
Follow-on engagementWhen knowledge management turns into a build engagement — custom software development with architectural clarity, documented code, and a handover plan from sprint 1.
To engineering →
Frequently asked questions
Microsoft 365 Copilot is the right choice if your knowledge sources already live in Microsoft 365 (SharePoint, Teams, OneDrive, Exchange) and the audience-group licenses are predictable. A dedicated RAG knowledge assistant pays off if you want to be platform-independent, EU hosting is mandatory, hybrid sources need to be integrated (file shares, Confluence, industry databases, scanned PDFs), or Copilot license costs become unfavorable across a large user base. We recommend Copilot as a first step — and complement it with RAG where Copilot hits its limits.
Yes — that is the core of our Knowledge Recovery sprint. We combine structured interviews with knowledge holders, reverse engineering of existing systems (database schemas, custom code, workflows), and process mining from log data. The result is handover-ready knowledge documentation — suitable for staff turnover, provider transitions, or the replacement of legacy in-house developments. If a rebuild engagement emerges from the recovery, we hand over to our engineering team.
We have been building SharePoint since 2007, and we have learned: a flat information architecture with 5–10 clearly delineated knowledge areas beats any deep hierarchy. Per area we define owners, lifecycle rules, metadata, and a search strategy. Permissions are set at the area level — not at the document level. This keeps SharePoint maintainable, searchable, and Copilot-ready. The permission concept is a prerequisite for the later Copilot rollout, because Copilot respects SharePoint permissions.
Permissions are the most difficult topic in knowledge management — and the most common reason Copilot rollouts fail. We work with Sensitivity Labels (Microsoft Purview) and a clear area-permission concept: whoever reads in a SharePoint knowledge area also sees its content in Copilot — whoever doesn't, sees nothing. For confidential areas (HR, management, M&A), we set up separate, isolated knowledge areas that are explicitly excluded from Copilot. The same principle applies to the RAG knowledge assistant: index partitioning per permission boundary, no cross-tenant leakage.
To take away · two materials
Two depths for different reading needs. The factsheet is quick reference (3–5 min) and immediately downloadable. The whitepaper is market education with methodology and comparison data (15–30 min) — you receive it by email after a short request.
3–5 min reading time · direct download · no form
Compact overview: scope, key figures, pricing model, approach — ideal for forwarding to CFO, procurement, or business unit.
15–30 min reading time · by email on request
Methodology, comparison data, recommendation framework — material for internal argumentation toward stakeholders.
Knowledge strategy conversation
30-min initial conversation — we clarify whether a SharePoint build, Copilot integration, a RAG pilot, or an acute Knowledge Recovery sprint is the right entry for you. You receive a concrete recommendation, typically promptly.
