Insurance · Finance & admin

Every letter reaches the right team on day one.

Inbox and broker mail are matched by the AI to customer and policy, and the task lands in the right queue. Unclear mail and complaints go to people.

Free intro callDiscuss this case
Inboxsorted
  • Ms Yilmaz08:14Water damage kitchen, photosto claims, high priorityClaim
  • Maklerbüro Kern08:02Application buildings, Roth familyto new business, 3 documents foundApplication
  • Mr Albers08:10Cancellation as of 31 Decto policy admin, deadline checkedPolicy
MatchingLetter, scanned
  1. Document typeCancellationrecognised
  2. CustomerThomas Albersfound
  3. PolicyHR-3391-0072, contentsmatches

Where time goes today.

  1. 70m

    Documents a year at one large insurer

    That is how many documents AXA Germany says it receives every year across all lines. Each has to be read and assigned before anyone can work on it.

    procontra, how insurers use AI (German)

  2. 45 sec

    Search per letter, often more than once

    That is how long the search for policy number and client in the CRM alone took at the broker MRH Trowe, before a task could be created.

    Bitkom, Beyond the Pilot, 2026 (German)

  3. Urgent items sit between routine

    A burst pipe, a complaint and a change of address arrive in the same inbox. Work follows the order of arrival, not urgency.

One case, step by step.

Shared inbox214 unread
  • Maklerbüro Kern08:02Application buildings, Roth familyBiPRO, 3 attachments
  • Scanning08:10Letter Mr Albers1 page, signed by hand
  • Ms Yilmaz08:14Water damage kitchen, photosurgent

Everything arrives in one inbox.

Emails, scanned letters, faxes and submissions through the broker portal or BiPRO. The AI reads every item, including attachments and multi-page scans.

Your team decides

Which inboxes and channels are connected.

Letter Mr AlbersScan, 1 page

Cancellation of my contents insurance

Policy number HR-3391-0072Policy

I hereby cancel the policy with due notice as of 31 December 2026.Cancellation

The reason is my move to my daughter in Bremen.

Please confirm receipt in writing.Confirmation

The AI recognises request and document type.

It tells an application from a claim, a cancellation, a change of address or a complaint and reads policy number, deadlines and wishes from the text.

Your team decides

Which document types exist and what counts as a complaint.

MatchingLetter Mr Albers
  1. Document typeCancellationrecognised
  2. CustomerThomas Albers, KD-448120found
  3. PolicyHR-3391-0072, contentsmatches
  4. Move mentionedContents cover can move alongnote

Suggestion: to policy admin, note that a move does not require cancelling

Customer and policy are found.

The AI looks up customer and policy in policy admin, even when the number is missing or wrong. Anything it cannot match with confidence goes to a review queue.

Your team decides

Above which certainty items are routed without review.

Task AUF-26-10233Queue policy admin
CustomerThomas Albers
PolicyHR-3391-0072
RequestCancellation as of 31 Dec 2026
Notice periodmet
NoteMove mentioned, offer to continue cover at the new address
DocumentLetter of 6 Oct, filed in the DMS
Into the queue

The task is in the right work queue.

With customer, policy, request and the document in the DMS. Urgent items are on top. The team works instead of searching.

Your team decides

How the request is handled.

What it delivers and what it connects to.

Works with the systems you already use

Who does what.

What the AI takes on

  • Reads emails, letters and broker submissions
  • Recognises request and document type
  • Finds customer and policy
  • Creates the task in the right queue

What your team decides

  • Handles the requests
  • Resolves what the AI cannot match with confidence
  • Takes over complaints
  • Sets rules and priorities

Who it fits.

A good fit if

  • A lot of mail comes in every day through several channels
  • Letters are scanned or already arrive digitally
  • Customers and contracts live in your policy or property management system
  • Several teams work from their own queues

Less of a fit if

  • So little mail comes in that one person sorts it on the side
  • Letters are not scanned and travel through the building on paper

What it costs

Fixed prices, clear steps.

You know up front what each step costs and what it delivers.

  • Recommended start

    First call

    Free

    30 minutes by video

    • We look at where you stand
    • You find out whether an AI strategy is worth it
    • No obligation, with a concrete next step
    Free intro call

  • AI strategy

    €1,990plus VAT

    Fixed price

    • One workshop day on real cases
    • A calculation with your figures for the best use cases
    • A paper for management and shareholders

    Money-back guarantee: three use cases that pay back within 12 months on paper, with your figures, or your money back.

    Request the AI strategy

  • Pilot with one use case

    from €5,000plus VAT

    Fixed price

    • On your data, in your process
    • Acceptance criteria agreed in writing
    Request a pilot

All prices net. Operation afterwards is from 20% of the project price a year, cancellable monthly. The code and the data are yours.

Built fromData & documentsWorkflow automationAI integration

Further reading

All posts
MCP explained: how AI assistants reach your systems and what that means for your ITAI toolsTiemo Timtschenko
Feeding ChatGPT your own data? What a knowledge system does differentlyKnowledge systemsTiemo Timtschenko
Why the AI cannot find your part number: pgvector, Azure AI Search and others compared in plain termsKnowledge systemsTiemo Timtschenko

Frequent questions

Common questions about it.

The questions that almost always come up about this use case.

What happens to items the AI cannot match with confidence?

They go to a review queue with the best suggestion and the reason for the doubt. A person assigns them, and the decision feeds into later matching.

Does it work with scanned mail?

Yes. The AI reads scans, faxes and multi-page PDFs. Where handwriting is hard to read, it flags it instead of guessing.

Does this fit property managers too?

Yes. Instead of applications and claims the requests are repair reports, service charge questions or contractor invoices, and items are matched to property, unit and tenant. The process is the same.

More use cases

  • Customer service

    Structured claims intake

    AI turns a claim notification by email, form or photo into a complete claim file and asks for missing details. Claims handlers decide on settlement.

    • Claims system
    • DMS
    • Email

  • Finance & admin

    Prepare policy changes

    AI recognises changes of address, new bank details and cancellations in email, letters and the portal, checks deadlines and prepares the booking.

    • Policy admin
    • Email
    • DMS

  • Customer service

    Sort the tenant inbox, draft replies

    AI sorts requests from tenants and owners from mail, portal and letters by property and deadline and drafts replies. Management checks and sends.

    • Email
    • Property ERP

The next step

Let us find out where AI pays off for you.

30 minutes, free and without obligation.

Free intro call