01
Discuss your workflow
30 minutes, no obligation. We look at which questions keep coming back and which documents hold the answers.
for HR, admin offices and teams with lots of procedures
Your team asks a question, the assistant searches your own procedures and policies and answers with a link to the source document. If the evidence is too thin, it refers the question to a colleague. You start with a pilot of 4 to 6 weeks.
No obligation · 30 minutes · online, with someone who builds, tests and fine-tunes it
Ethic is the AI studio of Zenjoy from Leuven. Since 2011 that team has built and maintained software, platforms and integrations for more than 400 organisations.
At a glance
Example, fictional
Using the HR policy from the AI fact sheet below. The first question is covered by the sources. The second is not, and it concerns one person.
Do I need to request working-from-home days in advance?
Yes. Under the working-from-home policy you agree your home-working days in advance with your manager.
Source: Working-from-home policy, point 3
Can I get extra leave to care for my father?
I can’t find anything about that in the agreed documents, and it concerns your personal situation. I’m forwarding your question to HR.
No source · forwarded to HR
Sound familiar?
Searching
A new colleague asks where the expenses procedure is. Someone spends ten minutes looking through folders.
The answer appears in Teams, with a link to the procedure.
Which version applies?
Three versions of the working-from-home policy are going round. Nobody knows which one applies.
The assistant only reads the folder with current documents. The source shows you which version it is.
Always the same colleague
HR or the office answers the same questions about leave and procedures every week.
The assistant handles simple questions. If it isn’t written down anywhere, the question goes to the right colleague.
Free chat tools
Staff paste internal documents into a free chat tool to get an answer faster.
They ask their question inside your own environment. The assistant sees no more than they are allowed to see.
How it works
This is how a knowledge assistant works in practice. If it finds too little evidence, it gives no answer and refers the question on.
A staff member asks a question in Teams or on the intranet, as they would ask a colleague. The assistant says straight away that it is an AI assistant.
The assistant only searches the folders listed in the AI fact sheet, and only what this staff member is allowed to open.
The answer links to the source document, so you can read it yourself. A source does not make an answer error-free.
If the agreed sources don’t cover the question, the assistant gives no answer and passes the question to the right colleague. We test that beforehand with real questions.
For each question the logs show which sources the assistant used, what it answered and where it referred the question. That way the owner sees which documents are missing.
01
30 minutes, no obligation. We look at which questions keep coming back and which documents hold the answers.
02
In your own environment, with one folder or one topic. We measure time, correct sources and corrections.
03
Roll out, adjust or stop. Based on figures, not gut feeling.
After a first conversation you get a proposal with a clear scope and a fixed price.
AI fact sheet
What does the assistant do, which sources does it read and who keeps an eye on it? That goes on one sheet that your team, your IT and your DPO can read together, and that you can pass around internally.

Example
HR policy knowledge assistant
Version 1.2 · March 2026
What it does
Answers staff questions about leave, working from home and expenses, with a link to the source document. Refers the question on if the sources don’t cover it.
Which data
Only reads the HR policy folder on SharePoint. Personnel files and pay data stay out of reach.
Model and location
A small language model that runs in the EU. Your data is not used to train models; that is in the contract with the provider.
Who checks
If the sources don’t cover the question, the assistant refers it to HR. HR checks a sample every month.
What it doesn’t do
What we log
Every question and every answer, kept for 90 days and visible only to HR.
Owner and stopping
Owner: the HR coordinator. You can stop at any time: we withdraw the access and hand the logs over to you.
Which systems
Your documents don’t have to move. The assistant reads them where they are, with as few permissions as needed.
SharePoint, Teams and OneDrive
Procedures and articles in Odoo’s Knowledge app (Enterprise edition)
We look at what is possible for each source
Frequently asked questions
Question not listed? Just ask us.
A data processing agreement and security overview are available on request.
Not always. If you work entirely in Microsoft 365 and the permissions on SharePoint are in order, first check whether Microsoft Copilot or an agent in Copilot Studio is enough. Both only show what a staff member is already allowed to open. We then help by checking permissions, making agreements and training your team, with no build. Go to Microsoft 365
If your procedures are out of date or contradict each other, tidy up first, at least for the topic you start with. An assistant doesn’t make messy documents better, it just passes them on faster.
Your own knowledge assistant pays off when your documents are scattered or sensitive, or when you want a small model in Europe.
The introductory call takes 30 minutes. For the pilot we ask for one colleague who knows the documents: 2 hours to define the folder or topic together, then checking now and then which questions the assistant couldn’t answer. Your IT lead and DPO approve the access beforehand. If Copilot or a tidied-up folder is enough, we’ll tell you.
Only the folders and sources we agree together in the AI fact sheet, and only what the person asking is allowed to open. As standard, we choose a small model that runs in the EU, from a provider that doesn’t use your data to train models. If you work with Microsoft’s Copilot, Microsoft’s terms apply. We go through those with your IT for each use case.
Yes, no model is always right. That is why the assistant only answers with a source, so you can check it yourself. If it finds no source, it says so. The owner in your team reviews a sample of questions and answers every month. We agree in advance how many mistakes are acceptable and what happens then.
That happens regularly. The integrations have checks and error alerts, so we get an alert when something goes wrong. We fix it under the maintenance agreement. Zenjoy has been maintaining platforms and integrations for years after launch; your assistant gets the same care.
Article 50 has applied since 2 August 2026: anyone talking to an AI system must know it is AI, unless that is already obvious. That is why the assistant introduces itself as AI at the first question. Who has which role, provider or deployer, we record for each assistant in the AI fact sheet. The rules on AI literacy (Article 4) have since been amended by the Digital Omnibus. Explaining to your team how the assistant works and where it falls short remains sensible either way. This is not legal advice: have it checked by your DPO or lawyer.
No, not everything. In the pilot we start with one folder or one topic that comes up often, such as leave or expenses. You choose which version applies, and the assistant reads only that one. The logs then show which questions found no source: that tells you which documents to update first.
Related
Integration
Microsoft 365Sector
EducationWorkflow
Shared mailboxBack
All agents and workflowsTell us. In a first conversation we look at which documents hold the answers, and whether Copilot or a tidied-up folder is already enough.
No obligation · 30 minutes · online, with someone who builds, tests and fine-tunes it