Verisk brings insurance intelligence into AI-assisted underwriting research

Verisk combines insurance data, analytical models and software used in underwriting and claims. Its AI work includes a commercial underwriting assistant and conversational access to selected ISO intelligence through Claude. The practical value is helping insurance professionals reach the relevant evidence inside a governed workflow, while preserving the difference between an analytical indication and the decision an insurer ultimately makes.

01 / ProductSeveral insurance AI surfaces with different jobs

Verisk’s May 2026 announcement describes two MCP connectors for Claude, including Underwriting Intelligence built around ISO information. ISO is identified as a Verisk business. This blueprint retains Verisk as the company identity and focuses on insurance research rather than treating each connector or acquired product as a separate company.

The Commercial GenAI Underwriting Assistant announcement describes a cloud-based commercial-property solution for summarizing information and assisting workflow tasks, with API integration and professional oversight. That is a different surface from querying loss-cost information in Claude. A buyer should identify the specific job before assuming one offer includes another.

Augmented Underwriting describes a connected process that enriches submissions with property and peril information, scores locations and can incorporate Touchstone catastrophe modeling. These components explain Verisk’s broader AI relevance: conversational assistance sits alongside data and analytical services used in an established insurance workflow.

02 / AudienceFor professionals interpreting insurance evidence

An actuarial analyst may need to understand a loss-cost filing history. A product manager may need context for a change affecting a line of business. An underwriter may need to assemble a clearer view of a commercial-property submission. Each role has a different question, but all need to distinguish source information from a generated interpretation.

The Guidewire blueprint provides a comparison for insurance operational software and the systems that manage core workflows. The Anthropic blueprint examines the model and assistant platform behind Claude. Verisk’s role in this comparison is the specialist information and insurance tools accessible through the workflow; a general AI subscription does not supply the underlying Verisk data entitlement.

The fit is weaker for someone expecting an unrestricted public insurance chatbot or an automatic recommendation to bind a risk. The connector guide requires an entitled account, and the vendor’s assistant announcement retains professional judgment in the process. A pilot should therefore be owned by the insurance team responsible for interpreting the evidence, with technology support for access and integration.

03 / WorkflowA proposed loss-cost research note with source checks

A contained evaluation could ask an authorized analyst to prepare an internal note on a selected line, jurisdiction and historical period. Use information the organization is already entitled to access and choose a question whose source trail can be checked in the existing ISO tools. This proposed workflow is an assessment of research access and interpretation, not an instruction to set premiums or change underwriting appetite.

Begin by defining the question precisely. Name the relevant geography, business line, period and type of information sought. A broad request for the latest trend may mix filing activity with experience data or an analytical indication. Write down the intended comparison before opening the conversational interface, so that a plausible answer does not quietly redefine the task.

Connect an approved Claude Enterprise workspace to the entitled Verisk account through the documented authentication route. The connector guide lists the required ISO access and multi-factor authentication. If a dataset is unavailable to the account in the underlying service, treat that as an entitlement question rather than trying to obtain it through differently worded prompts.

Ask for the relevant evidence and identify which tool produced it. Inspect the jurisdiction, period and update basis before comparing values. The analyst should then verify a sample directly in the underlying service. A response can accurately repeat a value while still answering the wrong temporal question or confusing an observed experience measure with a proposed filing change.

Draft a short note that separates retrieved facts, analytical interpretation and remaining questions. Preserve the link or identifier needed for another professional to inspect the source. If the result is incomplete, state which records or periods are missing. Do not let a generated explanation conceal that the tool retrieved only part of the available information.

Have another qualified colleague check the note against the familiar workflow. Record incorrect scope, missing records, ambiguous measures and the time needed to recover the supporting evidence. The pilot should reveal whether conversational access reduces navigation effort without weakening the team’s understanding. Any later operational change remains subject to the insurer’s ordinary review and decision process.

04 / PricingClaude access and Verisk entitlements are separate

The connector guide identifies Claude Enterprise, an active Verisk or ISOnet account with the relevant ISO product entitlements, multi-factor authentication and possible workspace-admin approval. It is an access guide, not a published price list. Confirm the organization’s actual availability and commercial scope with Verisk and the workspace administrator.

The Augmented Underwriting page uses a book-a-demo route. The public sources reviewed do not establish universal fees for the commercial assistant or a complete modular underwriting configuration. Ask for the required products, data, implementation and support to be identified individually in the proposal. A connector should not be assumed to include all of Verisk’s underwriting data.

For a custom integration, the API getting-started guide requires credentials from a Verisk client-service representative and distinguishes testing from production environments. Those technical credentials are part of a commercial and operational onboarding process. They do not establish that a Claude connector agreement authorizes every separate API workload.

05 / DistinctionsThe assistant can stay close to insurance-specific measures

The attraction is domain-specific access. Insurance professionals often need a particular measure for a particular jurisdiction and period, not a broad web summary of the market. The Verisk connector creates a route for asking a natural-language question while drawing on a defined specialist system. Its value is greatest when the resulting answer retains the context needed to interpret the measure.

The UnderWriting API documentation describes separate services with structured JSON or XML responses. This is useful context for teams considering deeper integration: conversational retrieval and structured service calls can serve different needs. A repeatable application may require explicit fields and error handling even when a conversational interface is convenient for an analyst’s exploratory question.

The broader augmented-underwriting approach also puts several perspectives on a submission beside one another. Property information, modeled peril and the insurer’s portfolio context need not agree perfectly. A useful system should make those differences visible to the professional instead of compressing them into one apparently definitive answer. That is a practical criterion for the demonstration, not a claim of observed product performance.

06 / QuestionsCheck tool choice, data timing and operational boundaries

First, test whether the assistant selects the right information source. The connector guide discusses different tools and warns about partial results and data freshness. Have the analyst deliberately ask questions that require distinguishing filing history, experience information and indications. Record whether the interface makes the selected tool and scope understandable.

Second, establish the data-update expectations for the intended use. A quarterly experience update is not the same thing as a live market signal. The research note should state what date the underlying information represents and whether later developments are outside the answer’s scope. A conversational interface should not make a periodic dataset appear continuously refreshed.

Third, clarify the boundary between research and action. The commercial assistant announcement describes integration with underwriting systems, but the proposed connector evaluation is read-oriented research. If a future configuration can write to a workflow, require a separate demonstration of permissions, approvals and recovery from an incorrect action. Do not infer those controls from a successful information-retrieval exercise.

07 / DecisionChoose a narrow insurance question before a broad AI rollout

Verisk is a prominent AI-related company because its specialist data and modeling products are being connected to new assistance and delivery methods. The most informative first evaluation is a narrow question that the insurance team already knows how to verify. That exposes whether AI improves access while preserving the meaning of the evidence.

Start with one internal note, one entitled dataset and a clear comparison period. If reviewers can reconstruct the answer and identify its limits, the organization has a basis for considering a wider workflow. Keep the commercial assistant, connector and API scopes distinct as the project develops, so that the next step is supported by both the product and the agreement.