AI Voice Agents in insurance: 3 key takeaways

Hamza Senoussi, Senior Manager of Data & AI Transformation at Converteo, leads strategic projects that turn data into value from building BI Centers of Excellence to integrating AI agents. His approach bridges business needs with scalable technology to drive measurable impact.

Key Takeaways

• Voice gives customers complete freedom, which creates a major engineering challenge: turning an unstructured flow of speech (hesitations, tangents) into structured, actionable data.
• The agent’s true value lies in its deep IT integration (like Salesforce), allowing it to take action on cases and policies right in the middle of a conversation.
• Generative AI really shines when applied to internal operations and reps, an efficiency lever that’s often overlooked in favor of customer-facing agents.

We recently deployed an AI voice agent for an insurance carrier. This wasn’t just a basic chatbot with a speech-to-text layer slapped on top; it was an agent fully capable of having real conversations, interpreting context, and interacting with business systems in real time.

On April 14th, during the RUNTIME AI event, we broke down the architecture behind this agent: the front-end, back-end, integrations, and the technical trade-offs we navigated live. Here are the three main lessons learned from this project.

1. Voice radically transforms data capture

A web form forces its own structure through required fields and a guided user journey. Voice, on the other hand, gives the customer total freedom: hesitations, spontaneous self-corrections, and tangents. Instantly turning this unstructured flow into data that a CRM can actually use is an entirely new engineering challenge. On that front, the maturity of the solutions available on Google Cloud actually exceeded our expectations.

2. An agent’s worth lies in its IT integration

The classic pitfall of conversational AI projects is building an agent that sounds incredibly articulate but is completely disconnected from your core tech stack. We took the exact opposite approach: our agent interacts with Salesforce in real time. Whether it’s verifying policies, creating cases, or recommending next steps, it takes action while the customer is still speaking. This deep synchronization with business data is what turns a basic prototype into a fully operational solution.

3. Internal AI: the most underestimated performance driver

While the market is hyper-focused on customer-facing agents, the impact of Generative AI on internal operations often flies under the radar. The use case we built primarily targets the reps rather than the policyholders. This is arguably the most differentiating aspect of our approach, and, paradoxically, the least explored area in today’s market.