RAG assistant for an insurance call centre: fewer escalations, more first-line resolutions

An insurer handling a high volume of calls needed accurate answers about policies, waiting periods, coverage and internal procedures without slowing operations. Knowledge existed across corporate documents, FAQs and internal rules, leaving first-line agents dependent on supervisors.

Executive summary

About the customer

An insurer with a telephone support service focused on answering queries quickly and accurately. Its priority was to reduce customer waiting times and minimise internal transfers caused by a lack of document context at the first line of support.

The Challenge: Dispersed Critical Knowledge and Expert Dependency

Call centre agents handled questions about specific policies and terms, but had no single place to consult. Questions often meant putting the call on hold to escalate to a supervisor, affecting handling time, answer consistency and the policyholder’s experience.

The Solution: RAG assistant with automatic update from SharePoint

An internal conversational assistant for the support team was built using RAG and connected to official business documentation. The knowledge pipeline was automated with Azure Functions, semantic indexing in Azure AI Search and document version control to keep answers up to date.

Automated processes (4 key flows)

1. Daily intake from SharePoint (Timer Trigger)

The Problem: Policy content and FAQs changed frequently, and updating the assistant manually risked leaving it out of sync.

Transformation: an Azure Function with timer trigger runs daily, with the schedule configurable through a CRON expression, retrieves documents from the SPO folder and starts synchronisation.

2. Differential knowledge synchronisation

The Problem: Maintaining the RAG system manually did not ensure consistency between the document repository and the vector database.

Transformation: The process compares SharePoint with the RAG index: it adds new documents, reindexes modified ones and removes from the index any documents deleted from SPO.

3. Semantic indexing in Azure AI Search

The Problem: Traditional keyword search was insufficient for complex natural-language queries.

Transformation: each document is fragmented, embeddings are generated and indexed in Azure AI Search as a vector database, allowing relevant context to be retrieved for each agent’s question.

4. Conversation observability and operational control

The Problem: Without traceability, it was difficult to measure whether the assistant improved call centre service.

Transformation: User, conversation duration, token cost, resolution outcome and escalation to a supervisor are recorded to monitor quality, efficiency and service cost.

Operational comparison

The Impact

The project transformed call centre operations from a reactive model dependent on experts into an assisted, traceable service. Agents gained autonomy on the first call, while supervisors could focus on complex cases and higher-value decisions.

• Operational autonomy: more queries resolved without second-line intervention.
• Living knowledge: configurable daily synchronisation tracking additions, changes and removals.
• Quality and compliance: answers grounded in corporate document sources.
• Data-based management: time, cost and escalation-rate KPIs for each conversation.

Operational results

• Better customer experience: less time on hold and more consistent answers.
• Increased agent productivity: immediate access to relevant document context without manual searches.
• Knowledge governance: Current content synchronised from a controlled repository.

Conclusion

“A RAG assistant does not replace the agent: it provides real-time context to resolve more questions and escalate more effectively.”

Would you like to apply this approach to your customer service or support operations? Contact AKAEN.