Skip to main content

An insurance organisation

Choosing what to automate in insurance

Assessment completed and demonstrator delivered

An insurance organisation wanted to simplify the handling of policyholder requests. Team interviews led to a first transcription demonstrator, delivered on a limited corpus before any full rollout.

The project

Teams at this insurance organisation answer policyholder requests by message and telephone. Before resolving a case, they need to understand the request and recover the context of earlier exchanges. The organisation wanted to know where AI could reduce that work.

We began with team interviews and workplace observations. This established the tasks to test and the requirements for integration with existing tools.

A first transcription demonstrator was then delivered on a limited corpus. The examples show how a call becomes a transcript, a summary and an assessment against explicit criteria. Transcription and speech synthesis in these examples use entirely local models.

The assessment gives teams a concrete first use case to examine before deciding on deployment. This case ends at the delivered demonstrator, without a full rollout for this client or a measured productivity gain.

Project technologies

Language processing, Transcription, and Business integration

From a call to its summary

Examples using fictional data. No full rollout at this client.
Find the request and the response givenDemonstration data

Demonstration data

Demonstration summary: initial request, response and follow-up actions.

The timings shown come from this trial. They vary with execution conditions.

Speech-to-text (STT) and text-to-speech (TTS) run on entirely local models.

Another example: classifying an email
Classify a messageDemonstration data
Incoming emailCategory and subcategories

Message extract

“I want an explanation of this reimbursement and to know how to switch to a plan with better optical cover. If nothing changes, I will cancel.”

Identified reasons

  • Reimbursement dispute
  • Plan change
  • Risk of cancellation

The request can be routed to the relevant team.

Extract from a demonstration using fictional data. Contact details have been removed.

The software

Dactylo

Explore the transcription and minutes software. The project described here remains a demonstrator on a limited corpus.

Which requests should you automate?

Start with your case handlers’ work and the tools they use.

Discuss your requests
About this example

Assessment completed and transcription demonstrator delivered on a limited corpus, without a full rollout for this client. Voice examples use entirely local STT and TTS models. Screenshots contain fictional data.

Share with your team

Copy this case for your team.