What this is about
We do not start with the tool. We start with your workflow.
An expert report grows out of site visits, photos, dictations, files and correspondence with insurers, lawyers and courts. Most of it is manual work between two systems. That is exactly where AI pays off first: small, measurable, without anyone talking about transformation. And it is where AI is least risky, provided the data is protected first.
Four workflows worth doing first
This is what it looks like in the real system.
Every image and clip comes from our own system, with invented data. Nothing is staged. Each one was recorded by an automated test that clicks exactly this path.
-
A note on the site visit is created and then sits in the case list. · Clip, no sound, loops 1 / 4
Dictate the site visit, the note lands in the case
After the appointment you speak your findings into the phone. The text lands as a note in the case, with date, place and parties. No typing in the evening, no loose voice memos.
-
The knowledge search finds the passage in the site visit protocol and names the document. · Clip, no sound, loops 2 / 4
Ask the file, get an answer with its source
Which moisture readings were in the first protocol? You ask in plain language and get the passage from the document, not a guess. Every answer shows where it comes from.
-
A draft letter to the insurer, built from the notes of the case. 3 / 4
Prepare correspondence
A letter to the insurer, a query to the owner, an extension request to the court. The draft grows out of the case, you review and send. No text leaves the office without your eyes on it.
-
The calendar shows the site visit and the submission deadline of the report. 4 / 4
Deadlines and appointments in view
Site visits, submission deadlines, extensions. All in one calendar that belongs to the case. Whoever covers the office sees at a glance what is due today.
Interfaces
Your systems stay. We dock on.
An expert office works with Outlook, Word and Teams, with appraisal software such as Gutachten Manager, SV Works, Bauexpert or autoiXpert, with DAT or Audatex for calculation, and with its accounting. That stays as it is. What you see above is our interface. You do not have to use it: most workflows run where you already are.
-
Microsoft 365, deeply
Mailbox and calendar sync in both directions, contacts too. OneDrive and SharePoint are knowledge sources: the file stays where it is today and still becomes searchable. Sign-in with the Microsoft account, no second password.
-
Teams as the workplace
Dictation, question, photo from the site visit: all in the Teams chat, and the answer comes back there. Whoever has Teams open needs no other window. Personal data protection applies there just the same and is visible on command.
-
Appraisal software alongside
Gutachten Manager, SV Works, Bauexpert, autoiXpert, DAT, Audatex: your specialist program stays the specialist program. We attach to its export, file store or API where one exists. A lean companion, not a replacement.
-
Accounting
sevDesk directly: invoices, receipts, contacts. Other systems via export and import, without double entry.
-
Proton instead of Microsoft
Offices without a Microsoft environment work with Proton Mail, Calendar and Drive. Same workflows, same search.
-
Open interfaces
A calendar subscription for every phone, CSV import for contacts and trips, a REST interface and an MCP access through which other AI tools can use your workflows as well.
Not an expert office, but you work the same way with files, appointments and correspondence? Law firm, engineering office, planning office: the path is the same. How we work in general
Data protection
What the language model sees. And what it does not.
The one question you should ask every provider: does the model see our clients' real names, addresses and file numbers? With us the answer is no. And that is not a promise on paper, it is built.
-
Placeholders instead of names.
Names, addresses, dates, phone numbers and file numbers are replaced before the request leaves the office. In the answer we swap them back. The mapping stays with you only.
-
Detection runs on your side.
A dedicated recognition model on our own hardware finds the personal data. No cloud service sees the text before it is protected.
-
Protection is always on.
It cannot be switched off in passing. It takes two keys: the organisation's approval and the decision in the single conversation. If one model fails and another steps in, protection stays on.
-
Provable to the data protection officer.
Every anonymisation lands in a log that cannot be altered afterwards. There is a daily report. When a court or the authority asks, you have the answer.
-
DPA and sub-processors from the system.
The data processing agreement does not come from a Word template. The system knows which services are involved and checks completeness before generating it.
-
You decide where the model runs.
EU provider, your own server or a large cloud model. In all three cases the model only receives placeholders. You can switch providers without anything changing for your office.
Visible in the conversation
Every conversation carries a green mark. One click shows whether protection is active and whether detection was reachable. The team sees it before asking the first question.
The log
Every entry shows how many personal data items were detected and replaced, by which method, and whether the check came back clean. No plain text is stored, only checksums.
How we work
One workflow, one small team, four weeks.
We do not promise transformation. We promise that after four weeks one workflow is noticeably lighter and you know whether it is worth continuing.
-
1
Where you stand
We look at your workflows, from order to invoice. Where time is lost, where errors happen, what is manual work between two systems. The result is an honest list, sorted by effort and benefit.
-
2
One pilot
One workflow, the team that does it daily, four weeks. Dictation to note, for example. Then we measure: is it used, does it save time, where does it stick.
-
3
Grow or stop
If the pilot holds, the next workflow follows. If it does not, we say so, and you have invested four weeks instead of a year.
Who we are
We do not work alone.
Development, design and organisational development we do ourselves, have done for years, and we use the system above every day. For everything outside our field we bring in people from our network. Who exactly depends on your office and is part of the first conversation.
-
We do this ourselves
Development
We build and run the system you see above and work with it ourselves every day.
-
We do this ourselves
Organisational development
A workflow does not change because a tool exists. We accompany the team so the new way of working sticks.
-
We bring this in
IT law and data protection
Contracts, records and disclosure duties are checked by lawyers, not by us. We bring them in as soon as it matters.
-
We bring this in
IT security
If client data is to live on your own server, we have access, encryption and backups checked from outside.
-
We do this ourselves
Design and usability
We build the interface with the team, not for the team. A tool nobody likes to use will not be used.
Honest framing
What we do not do.
-
We do not let the AI write the report.
The findings, the assessment, the signature are your work and your liability. The AI prepares, sorts and finds. It decides nothing.
-
We do not send liability questions to a model.
Whether a defect is material or who bears the cost is not answered by a language model. For legal questions we bring in people.
-
We do not start a large project.
No platform, no annual contract, no rebuild of the whole office. One workflow, one pilot, then we see.
-
If someone else fits better, we say so.
There are tasks for which another provider is the better choice. You hear that in the first conversation, not after the quote.
