Migration step by step: from folder-based filing to AI search

Moving from folder-based filing to AI search is not a data move with a cut-over date, but a series of manageable stages: review your documents, choose a pilot area, sort out permissions, connect, test, expand. Your existing file storage keeps running throughout – what changes is not where the documents are stored, but how the team gets to its knowledge. If you follow this order, you will have a working pilot area within a few weeks rather than a major project lasting months.

In brief

• “Migration” here does not mean moving: the documents stay where they are stored, and the knowledge base connects them and adds AI search on top.
• The process follows eight clear steps – from reviewing your documents and running a pilot area to expanding step by step.
• You do not need a big clean-up before you start; you tidy up specifically where documents that are now easy to find turn out to contradict each other.
• Permissions are settled before connecting, not afterwards – this protects confidential material and keeps the team’s trust.
• The ERP and accounting systems remain the system of record for transactional records; what changes is access to knowledge: instructions, reports, project documents, correspondence.

What does “migration” actually mean for an AI knowledge base?

The word conjures up the wrong images: a switchover weekend, a data move, old system off, new system on – and heaven help you if anything is missing. An AI knowledge base works differently. Your existing file storage – file server, SharePoint, inboxes, paper archive – stays where it is and is connected: the knowledge base reads the content, makes it searchable through text recognition and automatic tagging, and points to the original document in every answer.

So what changes is not where things are stored but how people search: instead of clicking through folder trees or asking colleagues, the team simply asks its question in plain language and gets an answer backed by a source. That takes the biggest risk of traditional migrations out of the project – there is no cut-over date by which everything has to work, and no moment when the company is left without its documents.

How does the migration work step by step?

This order has proven effective:

1. Review your documents and set goals. Where does your knowledge sit today: file server, SharePoint, inboxes, paper folders, people’s heads? And which three or four questions should the search answer reliably in future? These target questions – for example “How was installation X set up?” – later become the benchmark for testing.
2. Appoint a project lead. Not a full-time job, but a clear responsibility: one person in the company who pulls decisions together, answers questions and is the provider’s point of contact for the project.
3. Choose a pilot area. An area with real everyday value and a manageable scope – typically the project folders from the last few years, or the instructions and reports. Not the whole server, and not the most sensitive material.
4. Sort out permissions. Before anything is connected, decide who may see what: confidential material such as HR and management documents is either left out or tightly restricted. Start restrictive and widen access deliberately.
5. Connect the pilot area. The technical work runs in the background: reading the content, text recognition for scans, tagging by customer, project, installation and topic – the system reads these attributes from the document and fills them in itself. The document type, on the other hand, goes to the review area, where someone in the company confirms it, because it determines who can see the document. Day-to-day work is not affected.
6. Test with real questions. For two to four weeks, a small pilot group asks the questions that actually come up day to day – including the target questions from step 1. The things to check: are the answers correct? Do the sources lead to the right document? Are any documents missing?
7. Fine-tune and tidy up. Testing brings typical issues to light: a forgotten folder, conflicting versions of a work instruction, a permission that is too generous. Now is the time for targeted tidying up – mark the valid version, archive old material, adjust rights.
8. Expand and connect ongoing sources. Add one area after another, and at the same time connect the ongoing sources – email inboxes, scanners, phone photos, voice notes – so that new knowledge flows in automatically from now on instead of disappearing into folders again.

If you start with a pilot area this year, you will have a live system within a few weeks – and can then expand based on experience rather than theory.

What do you need to tidy up beforehand – and what can you leave?

The most common project killer is the plan to “tidy everything up properly first”. It sounds sensible, but in practice it means the project never gets going – fifteen years’ worth of documents cannot be sorted in your spare time, and they do not need to be. AI search finds content even in untidy structures, recognises duplicates and does not need the folder tree.

Very little has to be done in advance: exclude private, purely technical (software archives, backups) and confidential material from the connection – a matter of hours. Everything else follows the principle of tidying up once you can see the problem: only when the search shows three conflicting instructions for the same question is it worth resolving that particular conflict. That way, the tidying-up effort goes where it is useful rather than into folders nobody looks at any more.

What is deliberately left out of the migration?

Honest planning also means listing what you will not do. Transactional records – invoices, delivery notes, accounting documents – stay in the ERP and accounting systems as their system of record; that is where the statutory retention obligations apply, and that is where they remain. The ERP knows what happened – the knowledge base knows how things are done. It only adds to the records where context is needed: the correspondence about a complaint, the report behind a service order.

The same goes for other material: not everything old deserves to be made searchable. A dead collection nobody has opened in ten years can stay in the archive; a faded, illegible fax stays illegible, and that is recorded rather than glossed over. The benchmark is always everyday value: what gets connected is what people look for – instructions, settings, project history, experience. That is the material that turns folder-based filing into a knowledge base.

What are the most common migration mistakes?

Four patterns come up again and again. First, trying to do everything at once: connecting everything in one go instead of learning from a pilot area. Second, perfectionism about tidying up, which puts the project off indefinitely before it has started. Third, leaving permissions until later – if you only decide who may see what after connecting, you risk the very breach of trust that kills such a system within the team. Fourth, forgetting the team: a search nobody uses because nobody has explained it is a technical success and a practical failure.

All four mistakes have the same root cause: the migration is treated as a technology project rather than a change in how people work. The technology is the smaller part – what matters is that the pilot group asks real questions, the answers are checked (every answer names its source, because AI can be wrong) and the team finds the new way faster than the old one. After that, people keep using it of their own accord.

How do you get your migration started?

The first step costs nothing but an hour of thought: write down the five questions in your company most often answered with “Ask your colleague” or “Have a look on the server”, and where the answers are today. That usually gives you both the target questions and the pilot area. The article Connecting documents and systems gives an overview of which sources can be connected – from network folders and inboxes to the paper archive – and in what order it makes sense to do so.

The How it works page shows step by step how junova is introduced as a supported project – with data held in the EU, role management, a source in every answer, and without your documents being used to train third-party AI models.