Prepare Your Data for AI
Prepare your data for AI: tidy up first, then start.
The folder holds three versions of the same manual, and nobody knows which one applies. Before an AI uses your documents, it must be clear: what is current, what is correct, who may see what? We look at your storage locations and tell you what needs tidying up first and what is already good enough.
Which version applies?
Is what it says correct?
Who may see what?
What the AI starts with
What you get
You know what the AI can use and what has to go first.
Overview
Where your knowledge lives
Which folders, drives and systems matter for the planned AI and what is actually stored there.
Clean-up list
What goes, what stays
Duplicate and outdated versions, missing owners, unclear access rights. Sorted by importance.
Starter set
What the AI can start with
Which documents are already good enough and can be approved for a knowledge base.
Your effort
Someone who knows the storage
One person from the business side and someone with access to the systems. We agree the scope beforehand.
Your team does the clean-up based on the list; we can support on request. The price depends on the number of storage locations and systems. After the introductory meeting you receive a written quotation with schedule, price and what you will have in hand at the end.
What it looks like
Before: three versions. After: one that applies.
An example from a service folder, with sample data: this is how it looks once the clean-up list has been worked through. Technically this is called data preparation for RAG, meaning an AI that looks things up in your documents.
“So which maintenance manual applies now?”
Maintenance manual P200, 2025 version.Owner: head of service
Older versions go to the archive and are kept out of the knowledge base.
The two spare parts lists differ. Purchasing clarifies which one is correct.
Recorded: only sales may see the internal price list, including via the AI.
Example with sample data. Which questions to settle before the start is covered in our article on data before the first pilot.
How it works
Four steps to documents an AI can work with.
01
Review storage
Together we look at where the knowledge for the planned AI is stored.
02
Check
What is current, what is duplicated, who is responsible, who may see what?
03
Clean-up list
You get the list: what goes, what stays, who maintains what.
04
Define the start
We define which documents the knowledge base starts with.
Is this right for you?
Preparing your data pays off if …
Versions
… there are several versions
and nobody can say for sure which one applies.
Storage locations
… knowledge is spread across many places
in drives, emails, SharePoint, CRM or in individual people’s heads.
Responsibility
… it is unclear who maintains what
and who may see which documents.
Who is behind it
Prepared by the person who also builds the knowledge base.
We know from implementation which documents an AI needs. So we do not check just for the drawer.
Already in use
A real project
APIORIUM runs a multilingual website chatbot with its own knowledge base.
Personal
Martin Trixner
Has been building software and integrations since 2002 and is a certified AI-Automations-Manager.
Honest
Sometimes less is enough
If your documents are already in good shape, we tell you. Then you start straight away with the knowledge base pilot.
Frequently Asked Questions
Frequently asked questions about preparing data for AI.
Do we have to tidy up all our data?
No. Only the area the AI should start with, for example the service documents. The rest can follow later.
What is RAG?
An AI that first looks things up in your documents and then answers. How well that works depends directly on how current and unambiguous those documents are. Even with good documents an AI can be wrong, so it is tested with your questions before the start.
Who sees our documents?
We agree together beforehand which storage locations and access we need. Without explicit agreement your documents are not sent to external AI services.
What does it cost?
That depends on the number of storage locations, systems and people involved. After the introductory meeting you receive a written quotation.
Next Step
Where is the knowledge stored that an AI should use at your company?
Tell us about it in 30 minutes. We will tell you whether tidying up is needed first or whether you can start right away. In person with Martin Trixner, free and without obligation.