The real problem isn't AI, it's scattered knowledge.

Most SME leaders in Brussels and Belgium will discover the same surprise in 2026: plugging an AI assistant into their business is simple, but the answers remain disappointing. The reason is rarely the technology. It's because the information the AI needs is scattered across Word, PDFs, emails, Teams, SharePoint, network drives and a few busy heads.

A generalist AI assistant knows the public world. It does not know your contracts, internal rules, or how your A team is handling an urgent request on a Friday evening. As long as this knowledge is not written down and organised, AI improvises – and AI improvising on your procedures is a risk, not a gain.

The starting point is therefore not «which AI tool to choose?» but «is our knowledge exploitable?». This is exactly what SOPs, wikis and RAG solve.

SOP: writing down once and for all what each person does in their own way

A SOP A Standard Operating Procedure (SOP) is a document that outlines, step-by-step, how to perform a recurring task consistently. It is the answer to the most common symptom in SMEs: two employees carry out the same task differently, with no official document to decide.

A useful SOP is more than just a list of steps. It specifies:

• The objective and responsibilities Quién hace qué, y por qué. ;
• The detailed steps and checks at every key point ;
• the exceptions and specific situations, often the most valuable part; ;
• the documents, templates and tools associates ;
• The version, owner and revision date, so as to know at all times what is authoritative.

Concretely, a SOP explains how to welcome a new client, process an invoice, create an IT user, or manage an incident. Well-written, these procedures shorten onboarding, reduce reliance on a few key individuals, and become the reliable raw material your AI can consult.

Company Wiki: A Single Source of Truth

The Company wiki is the space where this knowledge lives. Rather than searching for the «correct» version of a document across multiple folders, your teams find procedures, guides, internal rules, frequent answers, and useful documents all in one place, organised in a clear tree structure.

A good company wiki brings four things:

1. a structure adapted to your services (support, HR, finance, sales, operations); ;
2. a Quick search and simple navigation; ;
3. a Version and validation management, to know who approved what and when; ;
4. a access from your usual tools, without changing working habits.

The result: a single, reliable source. No more permanent doubt about «which procedure is up-to-date.» And, above all, the wiki becomes the definitive and validated corpus on which the RAG will be based.

RAG in Brussels: the bridge between your AI assistant and your documents

The RAG (Retrieval-Augmented Generation, or Retrieval Augmented Generation) is the brick that connects AI to your knowledge. The principle: before answering, the assistant will first search the relevant passages in your documents, then generate its answer from these validated sources, rather than from generic knowledge.

An important clarification: RAG is not an AI in itself. It's a bridge. It allows the assistant to consult your internal library before formulating its response, much like an experienced colleague would check the procedure before answering you.

What RAG changes in practice for a Brussels or Belgian company:

• sourced answers The assistant cites the document and the version used (e.g. «according to SOP-014, version 3»), something no generalist model can honestly do; ;
• far fewer hallucinations the answers are rooted in your real data; ;
• GDPR compliance : your data remains under your control and access rights are integrated, a key point for Belgian SMEs subject to GDPR and NIS2 ;
• a continuous update When a procedure is updated, the assistant automatically uses the new version.

Three bricks, one single chain of knowledge

The common mistake is to treat SOP, wiki, and RAG as three separate projects. They are in reality the stages of a single strategy:

1. Collector — identify the knowledge present in your documents and among your collaborators.
2. To write — transforming real practices into simple and standardised procedures.
3. Centralise — organising content in a wiki or a structured document database.
4. Secure — define who can view, edit or use each category of information.
5. Connect — making knowledge accessible to your AI assistants and agents.
6. Automate — connect AI to your tools to prepare a response or trigger a controlled action.

Document, structure, connect, automate: each step prepares the next. You don't build a reliable RAG on contradictory documents; you don't write useful SOPs without understanding the terrain.

Where to start: a concrete use case per department

There's no need to document everything at once. The best project starts with a limited scope, with measurable results:

• Customer support and service to find a procedure, guide a colleague, propose a response, or identify the steps to resolve a request.
• Human resources answer onboarding questions, find an internal policy, explain an administrative process.
• Administration and Finance : access validation rules, templates, and billing procedures.
• Sales team : view services, conditions, tender procedures and answers to frequent objections.
• Operations & Quality apply the same methods and manage usual or exceptional situations consistently.

A single, well-handled process is often enough to demonstrate value, and then the solution expands naturally.

Security, GDPR and governance: plan from day one

A document repository contains sensitive internal information. Security is not an afterthought – it must be built in from the start, all the more so for a Belgian company subject to GDPR and, depending on its sector, to the NIS2 directive:

• from access rights by user, role or department ;
• a Separation There's a fine line between public and confidential information. ;
• The’authentication and identity checks ;
• the Version control and the revision history; ;
• the validation new procedures before publication; ;
• the revision and the deletion of obsolete content; ;
• consideration of the GDPR and your internal policies; ;
• the Accommodation choice (whose European hosting) according to your constraints.

The golden rule: an AI agent should only access information that its user is itself authorised to view.

Is there already a lot of documentation required?

No. That is the most common, and most unfounded, objection. A project can start with your existing files (Word, PDF, Excel, SharePoint, OneDrive, Teams, Notion, Confluence...) and the knowledge held by your collaborators. We start with a few essential procedures, then gradually expand. And no, it's not just for large companies: an SME can start with a single process, a simple document space, and an internal assistant limited to a specific need.

Keep Microsoft 365 and your current tools

You don't need to replace your environment. SharePoint can become your document centre, with integration into Teams, Copilot, Copilot Studio ou Power Automate. Depending on your context, the base can also connect to OneDrive, Notion, Confluence, Google Drive, your CRM/ERP, or automation workflows such as n8n. GVISION builds the solution around of your environment Microsoft 365, against him.

GVISION: your RAG and enterprise wiki partner in Brussels

GVISION is a Belgian IT partner based in Asse, on the outskirts of Brussels, which has been supporting SMEs in the Brussels-Capital Region and throughout Belgium since 2002. Your employees remain the owners of the business knowledge. Our role is to organise, write, secure and integrate it, as a single team that masters SharePoint, Microsoft 365, identities, security, automation and AI. A typical project follows six stages: documentary diagnosis, knowledge collection, writing and structuring, validation by your teams, preparation for AI, then integration and continuous improvement.

Frequently asked questions