Summary of Webinar on Operating AI Agents for Financial Services

ElevenLabs summarized a webinar on AI Agents for financial services, introducing cases from Revolut, Klarna, and Better, along with real-world deployment methods and security requirements.

The financial industry has spent $2.8 trillion on digital transformation since 2011, but customer experience is still stuck with phone waits and repetitive inquiries. Repetitive calls, agent burnout, and low satisfaction were presented as core problems that multichannel AI Agents need to solve.

In a retail banking demo, the agent went through authentication questions, checked account and savings balances, immediately filed suspicious transactions with the fraud team, and handed off loan inquiries to a dedicated sub-agent. It did not answer financial advisory requests and only guided users on what was possible, maintaining compliance boundaries.

Deployment principles were also presented.

• Authentication is handled with deterministic control, while LLMs are used for intent recognition and routing.
• Separating functions into sub-agents is more stable than a single prompt.
• Layering prompts, native guardrails, and custom guardrails increases safety.
• Conversation scoring, intent classification, hallucination detection, and containment tracking should be built in from the start.
• Workflow-level analytics and authentication design must be reflected from the early stages.

On the security side, it was emphasized that SOC 2 Type II, GDPR, and PCI DSS Level 1 are met, and that it provides by default up to 75% of AIUC-1 certification level, along with over 5,000 adversarial simulations and Lloyd's insurance coverage.