DEEN
Initial call

Your company’s knowledge as a digital brain.

I develop AI knowledge systems for small and mid-sized businesses. They bring together what is spread across documents, emails, wikis, CRM and ERP today, and answer questions about it with sources. For your team as well as for AI assistants, automations and AI agents. Everyone only sees what they are allowed to see.

Arrange an initial call

Reply within 24 hours. Free of charge and without obligation.

Service
Designing and building AI knowledge systems based on your own data
Sources
Documents, emails, wikis, CRM, ERP and databases – selected, not wholesale
Answers
In plain language, with the sources they are based on
Access
According to the permissions that already apply in your business
Use
By your team, AI assistants, automations and AI agents

01 Starting point

The knowledge is there. It is just everywhere.

Every business holds a great deal of knowledge. Rarely in one place, and rarely where it is needed right now.

  • Documents and filing

    Manuals, technical documentation, network drives, SharePoint

  • Emails

    Agreements with customers and solutions that only exist in one person’s inbox

  • Wikis and intranet

    Often set up with a lot of effort, rarely up to date

  • CRM

    Customers, agreements, call notes and history

  • ERP and inventory management

    Items, bills of materials, orders, deliveries

  • Databases and spreadsheets

    Product data, test values, Excel lists used as a reference

  • In people’s heads

    Experience, special cases and how a problem was solved last time

“You’ll have to ask Thomas.”

A sentence many businesses know. And its consequences: people search, ask around, interrupt colleagues or work with an outdated version. And when someone leaves the company, part of the knowledge leaves with them.

02Everyday

Questions that cost time today.

Asked in your own words. Answered with a source.

Typical questions from sales, service, engineering and administration. An AI knowledge system answers them from your own documents and systems, not from the general knowledge of a language model.

  1. Where can I find the current technical documentation?

    The system finds the document in its valid version and links to it, instead of a copy from an old folder being passed around.

    Based on: Document filing with version status

  2. How did we solve a similar problem in the past?

    It finds comparable cases, even if they were described in different words at the time, and summarises what helped.

    Based on: Service cases, tickets, project reports

  3. Which requirements were agreed with this customer?

    It gathers what is recorded in the quote, contract and correspondence, each with the place where it was found.

    Based on: CRM, contracts, email history

  4. Which products meet certain requirements?

    It compares the requirements with data sheets and product data and names suitable items with the values that matter.

    Based on: ERP, product data, data sheets

  5. What do our internal policies say?

    It states the rule that applies, for example on travel expenses or on approving purchase orders, and links to the document it comes from.

    Based on: Policies, manuals, process descriptions

These are examples. Which questions matter in your business is something we collect from your everyday work at the start. Later, they are also the benchmark for testing.

03 Definition

What an AI knowledge system is.

AI knowledge management means making company knowledge accessible so that people can ask about it in their own words.

Filing and searchGeneral AI chatAI knowledge system
What it is Folders, a wiki or a system’s search function. A language model that answers from its general training knowledge. A system that answers questions based on your own, selected content.
Result A list of files. Reading and piecing things together is up to you. A fluent answer that does not know your business. An answer in full sentences, with references to where it was found.
Finds Whatever contains the exact search term. Only what you paste in yourself. What matches the meaning of the question, even if worded differently.
Sources The document itself. Usually none that can be checked. Every answer states what it is based on.
Access rights Those of the system in question. No connection to your permissions. Adopts the permissions of the connected systems.
Limit Finds nothing if something was worded differently. Can produce mistakes that sound plausible. Only as good as its sources. No basis, no answer.

A digital brain that only knows what is in its sources. That is exactly what makes it reliable.

04 How it works

How an answer comes about.

First the permissions are checked, then the relevant passages are found. Only then does the AI write, and only from what it has found.

Example / A question to an AI knowledge system

RAG in three steps

The technique behind it is called retrieval-augmented generation, or RAG for short: an AI answer is generated with knowledge retrieved from your sources.

Retrieve
The passages in your sources that match the question are found, by meaning, not just by exact wording. The search only covers what the person asking is allowed to see.
Augment
Only these passages are passed to the language model together with the question. The rest of your knowledge stays where it is.
Generate
The model writes the answer from the passages found and names them. If nothing fits, it says so, instead of filling the gap with general knowledge.

05 Data sources

Documents are searched. Systems are asked.

A knowledge system connects different sources. Which ones belong in it is selected, not everything wholesale.

Documents and texts

are made searchable

  • Technical documentation, manuals, data sheets
  • Policies, process and quality management documents
  • Quotes, contracts, project documents
  • Wikis, intranet and selected email or ticket threads

The content is split into passages and indexed so that it can be found by meaning. A question about “unit not heating” also finds the passage “Fault: no heat output”. When a document changes, it is read in again.

Data from systems

is queried when needed

  • CRM: customers, agreements, history
  • ERP and inventory management: items, orders, stock
  • Databases and product data
  • Industry software with an interface

Structured data is usually not copied, but queried via interfaces (APIs) at the moment the question is asked. That way the data is current, and the system’s permissions continue to apply.

More on system integration

And the knowledge in people’s heads?

It cannot be read out, only written down: as a short answer to a frequent question, as a checklist, as a note on a solved case. A knowledge system shows where such knowledge is missing, because that is where questions go unanswered.

06 Control

Trust comes from being able to check.

Four questions that are settled before the system gives its first answer.

  • Sources and traceability

    Principle

    Every answer names the documents or records it is based on, with a link to where it was found.

    In practice

    Without a matching source there is no answer, just a note that nothing was found. That way every statement can be checked with one click.

  • Access rights

    Principle

    Anyone who is not allowed to see something in the filing system or the CRM will not see it in the knowledge system either.

    In practice

    Permissions are adopted from the connected systems and checked with every question, before searching. Confidential areas stay out entirely.

  • Being up to date

    Principle

    Answers are based on the current version, not on a copy from last year.

    In practice

    Changed documents are read in again, deleted ones removed, system data queried directly when needed. Every source has a person responsible for it.

  • Data protection and control

    Principle

    Which content is processed, by which AI service and where, is defined beforehand.

    In practice

    This includes the provider, where the data is stored and a data processing agreement. You decide what is logged and who manages the system.

Your company knowledge is not simply loaded into an AI.

With every question, the language model only sees the few passages that match the question and that the person asking is allowed to see. Which sources are connected in the first place is your decision.

07 Use

One body of knowledge for people and software.

Once built, the same knowledge is available, with the same permissions, to everyone who needs it.

  • Employees

    Ask in their own words, where they already work: on the intranet, in the messenger or in a dedicated interface.

    For example: A new colleague in service finds the answer without interrupting the experienced colleague.

  • AI assistants

    An assistant that drafts or summarises texts relies on your content instead of general knowledge.

    For example: A quote text with the current product data and your own wording.

  • Automations

    A fixed process specifically queries the knowledge it needs for a step.

    For example: An incoming enquiry is assigned to the right contact person based on the product documentation.

    More on process automation

  • AI agents

    An agent looks things up in the knowledge system before it prepares a reply or a case, with a source and within its permissions.

    For example: An agent answers a customer’s question about a delivery date and backs it up with the order in the ERP.

    More on AI agents

Without reliable knowledge, even the best AI agent is just a good guess.

08 Fit

When a knowledge system pays off.

Not every knowledge problem needs AI. What matters is what makes access to knowledge reliable in the simplest way.

An AI knowledge system makes sense

when several things come together

  • The same questions are asked again and again, often to the same people.
  • The knowledge sits in many documents and systems, not in one list.
  • Answers have to be backed up: technical details, agreements, policies.
  • New employees take a long time to learn where things are.

Something simpler is enough

and is then the better choice

  • It is about a few documents that can easily be organised.
  • The answers are in a system whose search works well.
  • The knowledge has hardly been written down. Then writing it down comes first.
  • Nobody is available to maintain the content. Then a knowledge system goes out of date too.

Not the biggest solution. The most sensible one.

Sometimes a tidy filing system, a maintained list of frequent questions or the search in an existing system is enough. A knowledge system does not tidy up a mess, it makes contradictions and gaps visible. If something simpler will do, I will tell you.

When automation pays off

09 Approach

Questions first. Then the technology.

A knowledge system grows one area at a time, not everything at once. With a defined result and an end.

  1. 01

    Initial call

    30 minutes by phone or video, free of charge. We talk about the questions that cost your business time today and work out whether a knowledge system is the right approach.

  2. 02

    Define the questions and users

    Who should be able to ask what? We collect real questions from your everyday work. If it is still open whether a change is worth it, a Process Check will clarify that. You can record how much time searching and asking around takes today with the “Measure First” tracking sheet, for example.

  3. 03

    Clarify sources and permissions

    Which documents and systems belong in it, and which explicitly not? Who is responsible for which source, and which permissions apply? This includes data protection and the choice of AI service.

  4. 04

    Build and test

    I build the system in stages, starting with one area, and test it with the questions we collected: is the answer right, is the source right, are permissions respected? It only goes live once that is the case.

  5. 05

    Introduce

    Your team uses the system where it already works. Feedback on wrong or missing answers goes straight into maintaining the sources.

  6. 06

    Hand over and extend

    Responsibilities for sources and maintenance are defined, the hand-over is documented. Further areas, automations or AI agents are only added once the first area is working well.

Arrange an initial call

10 Questions

Frequently asked questions.

On definitions, RAG, access rights, keeping knowledge up to date, costs and data protection.

What is an AI knowledge system?

An AI knowledge system makes internal knowledge from documents, emails, wikis and business systems accessible through questions asked in your own words. It searches the connected sources for the relevant passages, writes an answer from them and states what it is based on. The access rights already defined in the business continue to apply.

People also speak of an AI knowledge base or AI knowledge management. The idea is the same: making company knowledge easy to find and use with AI.

How does an AI knowledge system differ from a search or an AI chat?

A classic search finds documents that contain a search term and leaves the reading to you. A general AI chat answers from its training knowledge and does not know your business. An AI knowledge system combines the two: it works with your own data, understands questions in your own words and answers with sources.

What does RAG mean?

RAG stands for retrieval-augmented generation. For a question, the relevant passages are first retrieved from your own sources. Only these passages are passed to the language model together with the question. The model then generates the answer from them and names where it found them.

This makes it possible to use AI with your own data without training your own model and without handing all of your company knowledge to a language model.

Is all company data loaded into an AI for this?

No. Which sources are connected is selected beforehand; confidential areas can be left out entirely. With every question, the language model only sees the few passages that match the question and that the person asking is allowed to see.

How do you make sure everyone only sees what they are allowed to see?

Access rights are adopted from the connected systems, such as the document filing system or the CRM. Before searching, the system checks which content the person asking is allowed to see. Anything they cannot open in the source system does not go into their answer either.

How is the knowledge kept up to date?

Changed documents are read in again and deleted ones removed. Data from systems such as the CRM or ERP is usually queried via interfaces at the moment the question is asked and is therefore current. Every source has a person responsible for it, because outdated documents stay outdated, even when an AI reads them.

Can an AI knowledge system give wrong answers?

Yes, that can happen, for example when sources contradict each other or are outdated, or when a question is ambiguous. That is why every answer names its sources and can be checked. If the system finds no suitable basis, it says so instead of making up an answer. Before going live, it is tested with real questions from your everyday work.

How are AI knowledge systems and AI agents connected?

An AI agent completes tasks, such as answering a customer enquiry or preparing a case. For its results to be right, it needs reliable knowledge. The knowledge system provides it, with sources and within the permissions defined for the agent. Your team, AI assistants and automations use the same knowledge.

What does an AI knowledge system cost?

That depends on the number and type of sources, the interfaces needed and the requirements for permissions and data protection, which is why there is no price list. Before implementation, we define the scope, the goal and a fixed end to the project, and you receive a quote for it.

Running costs include use of the AI service, depending on the number of questions, and operation. Both belong in the calculation from the start.

How do you handle data protection?

Which content the system processes, where it is stored and which AI service is involved is something we clarify before implementation. This includes the provider, where the data is stored and a data processing agreement. Areas with personal or confidential data are only connected if there is a reason and a legal basis for it.

Get in touch

Which questions cost your business the most time today?

Describe them to me in the initial call. Then we will know whether a knowledge system is the right way or whether there is a simpler one.

Arrange an initial call

Reply within 24 hours. Free of charge and without obligation.