1. Model Details

• Organization: Coverage, LLC ( a Red Ventures company)
• Model Developer: OpenAI
• Model Date: 2024-11-20
• Version: gpt-4o-2024-11-20 (base model, no fine-tuning)
• Type/Architecture: Multimodal large language model, transformer architecture
• Training Algorithms: Gradient-based optimization with reinforcement learning from human feedback (RLHF)
• Parameters: Proprietary to OpenAI
• Fairness Constraints: Guardrails prevent financial advice, carrier or coverage recommendations, and off-topic responses. Governance processes include regular review of samples, bias monitoring, and escalation protocols.
• Contact Information: Coverage Compliance — compliance@coverage.com
• Citation: “gpt-4o-2024-11-20, OpenAI; used by Coverage.com during a lead collection experience with system instruction constraints.”

2. Intended Usage and Limitations

Benefit and Intended Usage
The Coverage Chatbot provides insurance education for property and casualty domains, focusing on auto and home. The chatbot is embedded with the Coverage quote experience as an expandable modal. It accepts free-text input from the user and provides corresponding answers/support to explain core insurance concepts in plain language for the purpose of improving understanding and promoting confident decision-making during the lead collection  process.

Functional Range
The model currently supports general insurance education only. It does not access user-provided data or Coverage.com’s quoting systems. It cannot retrieve quotes, execute transactions, or provide individualized coverage recommendations.

Known Limitations

• No access to carrier-specific data, including rating criteria, underwriting rules, or state-level requirements.
• No integration with user data or lead collection flow context.
• Educational only; cannot take actions or generate live quote results.
• May return incomplete or generic responses when asked about carrier- or state-specific topics.

Implementation Environment
All inference and model execution occur through OpenAI’s API. Coverage does not host or modify the model and performs no fine-tuning.

Sustainability
Coverage relies on OpenAI’s sustainability and energy efficiency commitments for model training and inference. Coverage does not maintain compute infrastructure for model operation.

3. Ethics and Safety

Governance Model
Coverage maintains an internal compliance and product governance framework that oversees chatbot deployment. Interactions are monitored through an automated scoring system that evaluates alignment with defined safety and compliance criteria. Potential violations trigger alerts for manual review. Escalations are handled by the Coverage compliance function, which determines remediation and corrective action.

Safety Evaluations
Adversarial prompt and red-team testing were conducted to assess the chatbot’s ability to resist attempts to elicit financial or coverage advice. The model consistently maintained scope adherence and did not override system instructions during these evaluations. Ongoing and regular evaluation is conducted to ensure compliance with Coverage’s standards.

Bias and Fairness Testing
Preliminary evaluations tested the model’s consistency across demographic variables, including gender, age, and race. No differential treatment or bias violations were observed. Broader fairness evaluations will continue as usage expands.

Mitigations
Coverage applies layered moderation and content safety controls. The system restricts financial or legal advice, suppresses off-scope outputs, and redirects users to human support where appropriate. Users are informed they are interacting with an AI system. These measures align with Coverage’s responsible AI governance standards.

Escalation and Oversight
Responses flagged through automated or manual review undergo evaluation by the compliance team. Confirmed issues are documented, corrected, and integrated into future guidance and governance updates. System-level performance and risk management are reviewed periodically.

Ethical and Societal Considerations
The chatbot supports informed consumer decision-making but does not replace licensed agents or offer binding advice. It advances transparency and fairness in insurance education. Users are clearly notified when engaging with the AI system and directed to licensed professionals for individualized support.