AI in daily work
TodayThe details are in the PDF. Someone types them into the form.
AI is strong at recurring tasks involving a lot of text: reading, classifying, preparing, summarising. It is weak wherever something has to be binding. So we use it as one step inside a workflow, with a point at which a person confirms the result.
Where it usually gets stuck
- Details are retyped from PDFs and forms.
- Someone has to read long email threads to know the current status.
- Existing knowledge sits in documents nobody searches.
- Standard texts are rewritten from scratch every time.
AI is strong where a lot of text has to be read and classified – and weak where something has to be binding. The useful question is therefore not whether to use AI, but at which point in a workflow it contributes something and who checks afterwards.
An example: details can be reliably suggested from an incoming PDF. Whether they are accepted is decided by a person – and that takes two seconds instead of two minutes of retyping.
What that looks like in the workflow
01 Document
It comes in
By email, scan or upload, in whatever format it happens to arrive.
02 Suggestion
The details are recognised
Number, date, people involved, amount – as a suggestion, not as fact.
03 Check
A person confirms
Side by side: what was suggested and what the document says.
04 Acceptance
Only then does it count
The data goes to the case, the document into the filing.
What gets built
- Incoming documents evaluated and their details transferred, with a review step
- Summaries of threads and files
- A search across your own material that answers in full sentences
- Draft texts based on your templates and your wording
- A written rule on which data goes where
What changes
- Details from incoming documents are suggested and checked instead of retyped.
- Long threads can be taken in within a paragraph.
- Your own material becomes searchable, not just storable.
- At every binding point there is a human approval.
From applications we have built
What follows are excerpts from applications we have actually built. No mocked-up interfaces and no invented figures.
Questions about AI
Does our data then go to an American provider?
Only if you want it to. There are models operated in Europe and models that run in-house. We settle this beforehand and put it in writing.
What if the AI gets something wrong?
That is why there is a human approval at every point where something becomes binding. AI prepares; it does not decide.
Is it worth it for a small company?
Not always. We check it against your actual tasks. If nothing adds up, we say so.
Also searched asAI · artificial intelligence · evaluating documents · drafting text · summarising emails · searching knowledge