Agentic AI for P&C Insurance

Contextual answers

Answers questions about the policy, claim, or task in progress, with links to original source documents and notes.

Recommended actions

Recommends next-best actions on policies and claims and executes administrative tasks when prompted.

Claim Summarization¹

Generates claim summaries and ongoing updates in seconds.

Policy Change²

Executes multi-step policy updates, validates requested changes against rules and product models, and provides AI-drafted change jobs for review and binding.

Application availability

ClaimCenter, PolicyCenter, BillingCenter³, InsuranceNow

¹ Restricted Availability (RA) in ProNavigator for ClaimCenter; ² Early Access (EA) in ProNavigator for PolicyCenter; ³ Coming soon

Insurance knowledge base

Centralizes unstructured documents, forms, endorsements, business guidelines, rate manuals, and standard operating procedures into a retrievable knowledge base for ProNavigator answers.

Contextual retrieval

Produces insurer-specific responses by retrieving policies, business guidelines, and core system transactional data rather than searching the open web.

Agentic Framework¹

Build your own custom AI agents for non-standard workflows and proprietary business processes.

Developer Assistants

Developer Assistant for Gosu; Developer Assistant for Integrations

Model Context Protocol (MCP) server access

Claims MCP Access with ProNavigator for ClaimCenter; Policy MCP Access with ProNavigator for PolicyCenter; Billing MCP Access with ProNavigator for BillingCenter²; InsuranceNow MCP Access with ProNavigator for InsuranceNow

¹ Included for all ProNavigator customers on Guidewire Cloud Platform; ² Coming soon

Developer Assistant for Gosu

Writing, refactoring, and optimizing Gosu code

Developer Assistant for Integrations

Building, changing, and maintaining Integration Gateway flows

Developer Assistant for Jutro

Building digital experiences on the Jutro Digital Platform

AI coding tools supported

Claude Code; OpenAI Codex; Cursor; Gemini CLI; GitHub Copilot CLI; GitHub Copilot

IDEs supported

IntelliJ; Guidewire Studio

Accessible by MCP

Guidewire Cloud APIs; proprietary programming languages; data models; curated libraries; design patterns

Bring your own key

Developer Assistants are MCP servers that are connected in the insurer’s existing AI coding agents using your own service-provider credentials; Guidewire does not charge for token consumption using Developer Assistants. Tokens are consumed and charged by your third-party AI service provider.

Code privacy

Data retrieved from Guidewire Cloud Platform by Developer Assistants is protected by our AI security protocols. Users are still responsible for adhering to their own company data policies and security protocols when using third-party AI service providers.

Agent Studio

Low-code visual interface to build, test, deploy, and manage custom AI agents.

Agentic SDK

TypeScript/JavaScript toolkit to build agents via code and integrate them with existing engineering practices and CI/CD pipelines.

Task composition

Decomposes high-level goals into subtasks, routes dynamically from intermediate outputs, and validates data against strict JSON schemas before passing it to the core system.

Memory management

Preserves active session state across interruptions and indexes long-running workflow history so agents can resume later using metadata-filtered context such as claim ID, jurisdiction, or line of business.

Agent and subagent orchestration

Delegates discrete workflow steps from a coordinating agent to specialized subagents and passes summarized results between steps.

Insurance knowledge base

Retrieval-augmented generation (RAG) with metadata filters for state, policy type, effective date, and line of business, returning retrieved information with citations.

MCP servers & tools

Exposes Guidewire Cloud APIs, external databases, and partner systems as MCP tools; agent actions run through Guidewire core validation pipelines and business rules.

Foundation models

Anthropic Claude; OpenAI GPT; AWS Nova; Meta Llama

Gateway controls

Centralized model routing, token limits, usage metering, and automated failover through AI Connect.

Offline evaluations

Evaluations against insurance-specific custom datasets and ground truth using Recall@K, Precision, F1, and nDCG.

Production evaluations

Automated checks of responses and planned actions for safety, groundedness, and relevance; failures are blocked and routed to a human.

Audit tracing

Unified trace of every model call, retrieval step, and tool call, scoped to the permissions of the human user that invokes the AI agent.

Observability

Datadog evaluation scores and step-by-step audit trace trees for agent reasoning and production performance.

Access controls

AI agents run in the insurer's isolated cloud tenant and inherit the acting user's single sign-on (SSO) and role-based access control (RBAC) permissions.

PII redaction

Personally identifiable information is redacted from agent activity before it reaches audit logs.

Model training restriction

Data sent to a model is blocked from being used for model training.

Predict for Profitability

Embedded analytic insights in PolicyCenter for rating and risk selection

Predict for Claims

Embedded analytic insights in ClaimCenter for severity segmentation, litigation potential, claim routing, and subrogation likelihood

Predictive models

Create, manage, and evaluate custom models from imported datasets; import third-party models; integrate Guidewire Industry Intel models

Supported algorithms

Blazing Text; CatBoost; Decision Trees; Explainable Boosting Machine; Forward Stepwise GLM; Hugging Face; Individual GAM-H2O; Individual GLM; Random Forest; Text Mining; TF-IDF; XGBoost

Model development tooling

Auto-bucketing; AutoML hyperparameter tuning; chaining models; exploratory data analysis; model comparison; pipeline construction; post-bucketing analytics; real-time scoring; variable management

Model evaluation metrics

Deviance; Gini coefficient; model estimate and output scores; RLLS point estimates; signal, noise, and selection scores; consistency; spread

Performance monitoring

Transaction and model performance metrics, score distributions, and input-value deviation monitoring

Model governance framework

Machine learning and statistical model governance for reliability, ethics, and regulatory compliance, based on fairness, accountability, transparency, and proactiveness

Pooled industry data

P&C data from more than 115 insurers representing $326B in direct written premium, used to build prebuilt, market-validated predictive models and scores

Federated machine learning

Analytics code operates in the insurer’s system locally and returns only mathematical insights; raw sensitive data does not move or get extracted to another system.

Model governance framework

Reliability, ethics, and regulatory compliance based on fairness, accountability, transparency, and proactiveness

U.S. Auto Collision Subrogation Intel

Subrogation recovery likelihood for collision claims

U.S. Auto Bodily Injury Intel

Attorney likelihood; litigation likelihood; payment likelihood; estimated severity; estimated loss

U.S. Workers' Compensation Intel

Legal likelihood; medical payment likelihood; medical severity tier; indemnity payment likelihood; indemnity severity tier; overall loss tier

Application availability

Claims Intel models are embedded directly in ClaimCenter and InsuranceNow through Predict with Analytics Manager.