UConn, Georgia Tech Students Win ACORD Challenge With AI for Pet Insurance
ACORD named University of Connecticut graduate student Manasa Ramaka the winner of the ACORD Student Challenge 2025, awarding a $5,000 prize for her AI-driven approach to structuring pet insurance underwriting data. Two runners-up-UConn's Rutendo Mahanzu and Georgia Tech researcher Pramod Misra-each received $2,000.
This year's brief: use AI tools to compile and analyze the data fields that matter for pet insurance underwriting. With premiums exceeding $21 billion worldwide in 2024 and growing 12% annually, a canonical set of fields is the foundation for clean, comparable, and usable data across carriers.
How the winning projects worked
- Winner - Manasa Ramaka (UConn): Applied large language models (LLMs) to generate a relevant field list and validate preexisting data from five top pet insurers.
- Runner-up - Rutendo Mahanzu (UConn): Used AI to extract fields from existing ACORD forms for other lines (homeowners, auto) and adapt them for pet insurance.
- Runner-up - Pramod Misra (Georgia Tech): Built a custom AI agent using multiple LLMs to analyze templates, carrier practices, underwriting manuals, and regulatory filings, organizing outputs into key data categories.
The projects covered data architecture, prompt engineering, and process documentation. ACORD noted the results will inform future updates to ACORD Data Standards used across the industry.
Why this matters for insurers
- Cleaner underwriting inputs: Consistent field definitions reduce manual review and rework.
- Faster product rollout: Standardized inputs speed filings, rating, and distribution integration.
- Better risk signals: Structured data enables more accurate pricing and clearer appetite rules.
- Operational efficiency: AI-assisted validation catches missing, conflicting, or out-of-range values early.
- Easier partner connectivity: Shared definitions reduce mapping pains with MGAs, TPAs, and platforms.
Practical moves you can make now
- Audit your field list: Inventory current pet underwriting inputs across apps, rating, and policy admin. Flag duplicates and weak definitions.
- Map to ACORD: Align names, formats, and enumerations with ACORD where possible. Note gaps that need proposals.
- Pilot an LLM-assisted workflow: Use an LLM to propose field definitions, permissible values, and validation rules. Keep a human-in-the-loop.
- Stand up validations: Implement checks for completeness, consistency, and regulatory triggers before data hits rating.
- Version your prompts: Store prompts and outputs with timestamps and model versions for auditability.
- Close the loop: Feed underwriting and claims outcomes back into field prioritization and rules.
- Prep for filings: Document definitions and transformations so state reviewers can follow the logic.
Governance and technical notes
- Data quality first: Poor source data will swamp any AI lift. Tackle sources and ownership early.
- Model choice: Test multiple models for extraction vs. classification vs. validation-cost and accuracy vary by task.
- PHI/PII controls: Keep sensitive data masked or use approved environments with logging and access controls.
- Human oversight: Require sign-off on new or changed fields and rules before production use.
ACORD states that the prize-winning submissions will help shape future ACORD Data Standards for pet insurance. All three finalists also receive passes to ACORD Connect 2026 and one year of mentorship with an industry professional.
Want to upskill your team on AI workflows?
If you're building LLM-assisted underwriting and data standards, these resources can help:
Your membership also unlocks:
