Fortune 500 US Insurance Company (unnamed)

For a Fortune 500 Property & Casualty insurer, the Investor Relations team faced mounting pressure to deliver timely competitive intelligence ahead of quarterly earnings cycles. Analysts were manually reviewing competitors' financial earnings reports — dense, unstructured documents requiring deep domain interpretation — and attempting to anticipate questions executives would face on earnings calls. The process was both labor-intensive and subjective, making it resistant to automation through conventional AI or RPA tools. This bottleneck limited the team's capacity for higher-value strategic work, delayed publication of quarterly peer analysis reports, and reduced the quality of stakeholder communication preparation for executive leadership.

Cognizant collaborated with the IR team to design and deploy a generative AI-powered autonomous agent built on the ReAct (Reason and Act) framework — a methodology that allows the agent to iteratively reason through tasks and take corrective action without human intervention. The agent was deployed on AWS, with Lambda functions orchestrating agent calls to handle end-to-end financial analysis and report generation. A reflection module enabled self-correction, improving output quality over successive iterations. Critically, the system was grounded in client-provided standard operating procedures (SOPs) supplied by subject matter experts, ensuring generated outputs aligned with the insurer's existing reporting standards. This design allowed the agent to process diverse unstructured financial documents at scale while minimizing the need for manual summarization or review.

The deployment delivered measurable improvements across efficiency, output quality, and strategic readiness:

• 24% reduction in report drafting time, directly reducing the manual burden on IR analysts
• 95–98% positive feedback from client SMEs on model-generated report parameters, indicating strong alignment with internal expectations
• ~40% match rate between AI-predicted earnings call questions and questions actually posed during the insurer's earnings call — a concrete indicator of improved strategic preparation

Beyond the metrics, the IR team gained capacity to focus on higher-value initiatives, stakeholder communications improved, and the publishing cadence for quarterly peer analysis reports became more consistent.

• ReAct-framework agents outperform RPA and conventional AI for tasks involving unstructured financial documents that require iterative reasoning rather than rule-based extraction.
• Grounding outputs in client SOPs is non-negotiable — without this anchor, generative AI outputs risk misalignment with domain-specific reporting standards, requiring costly manual correction.
• Question prediction accuracy is an underused success metric — the ~40% earnings call hit rate provides a quantifiable proxy for strategic readiness that IR teams in other verticals could adapt.
• SME feedback loops should be built into the evaluation design from day one, not added post-deployment, to systematically validate output quality before full rollout.