Guidewire ClaimCenter AI Agents: API and Automation Guide
Explore ClaimCenter native AI and eight verified Cloud API operations, then design governed signature tracking, intake triage, and adjuster briefs on Major.

Guidewire ClaimCenter AI agents can triage claim activities, assemble adjuster briefs, and explain document exceptions while ClaimCenter remains the claims system of record. This guide covers Guidewire's built-in AI, verified Cloud API operations, and how Major adds governed cross-system work through deterministic apps, agents, and workflows.
What Guidewire ClaimCenter does for claims teams
ClaimCenter supports property and casualty claims work from first notice of loss through adjudication and financial handling. Its Cloud API consumer guide separates claim resources from shared resources such as activities and notes, which matters when designing an integration.
An intake specialist needs enough claim context to arrange first contact. An adjuster needs the outstanding activities and evidence before taking the next action. A supervisor needs visibility into ownership, overdue work, and exceptions. Integration teams need to preserve those responsibilities rather than give one service unrestricted authority.
Major adds a durable operational layer around that work. Its agents build deterministic apps for repeatable operations, including claim-to-document mappings, review queues, and action histories. The agent retains the judgment step: interpreting incomplete context, prioritizing an exception, or drafting an explanation.
Start with one bounded process, such as signature follow-up, rather than autonomous claim settlement. The insurance automation overview provides the broader operating context. A useful first outcome is a traceable recommendation that an adjuster can accept or reject without reconstructing the whole file.
Guidewire ClaimCenter's built-in AI
Guidewire already offers AI inside ClaimCenter. The official Palisades release highlights describe ProNavigator for ClaimCenter, an embedded insurance knowledge capability. Adjusters can ask coverage or process questions and receive cited answers from approved guidelines, procedures, regulations, and other sources.
That is valuable at the point of decision. An adjuster checking a jurisdiction-specific acknowledgment timeline should not have to hunt through unrelated documents before understanding which approved guidance applies.
The same release notes describe Claim Summarization for ClaimCenter, part of ProNavigator, as restricted availability. It generates focused claim summaries inside the ClaimCenter interface, with globalization support noted in Palisades. Confirm availability for your deployment rather than assuming every tenant includes it.
Major adds value when the work extends beyond that interface. A signature exception may combine ClaimCenter activities, DocuSign evidence, and a supervisor's approval. A Major app keeps the cross-system state; an agent judges the mismatch; a workflow routes the proposal and executes an approved update. Avoid rebuilding an in-product summary merely to put the same text in another screen. Build where persistent coordination, external evidence, and accountable approvals are missing.
How ClaimCenter connects to Major
Guidewire ClaimCenter has no native or Pipedream connector in the catalog: it connects as a custom API connector, with base URL and authentication configured once by IT. Major orchestrates ClaimCenter; it never replaces it.
IT supplies the tenant's HTTPS Cloud API base URL ending in /rest, not a shared public Guidewire API host. The endpoint conventions place the Claim and Common APIs beneath that base. The relative paths below follow the documentation and therefore do not repeat /rest.
For unattended work, Guidewire documents an OAuth2 client-credentials flow for standalone services. Store the client credentials centrally in Major's connector vault, never in application code or prompts. Scope connector access separately for reading evidence and executing approved writes.
SSO, RBAC, and audit logs cover the Major app, agent, and workflow. These controls supplement ClaimCenter authorization; they do not override it. Before enabling writes, have IT verify the actual service role and accessible resources against the tenant's configuration. A workflow that only prepares adjuster briefs should not receive activity completion permission.
ClaimCenter API guide for useful builds
Authentication and operating constraints
The standalone-service example distinguishes the cc.service resource-access strategy from API-role scopes. Its example role scope, scp.cc.acme_fnolreporter, belongs to that example application, not your integration. IT must register your own role and deployment scopes. The documented base service configuration makes all resources available, so inspect resource-access configuration rather than treating a narrow role name as proof of narrow data access.
Send the JWT in the Authorization bearer header, as documented in authenticated service calls. Unauthorized individual resources can appear as 404 responses; inaccessible collection members are omitted. An empty result is not proof that no work exists.
Pagination uses pageSize, pageOffset, and includeTotal. Follow returned next and previous links. Default page sizes are 25 with a maximum of 100 unless overridden for the resource. Totals stop counting at 1,000, so do not use that number as an exact backlog above the threshold.
Universal Cloud API request limits and quota headers are not documented publicly in the reviewed reference. Obtain tenant limits from your Guidewire administrator. Do not invent a requests-per-minute allowance or assume another vendor's headers apply.
App Events supports outbound business events, including claim and incident changes, with at-least-once delivery and safe ordering by primary object. Webhook signature headers, verification algorithms, and retry intervals are not specified in that public overview. Confirm the configured subscription's security and delivery contract before accepting callbacks; otherwise begin with scheduled API reads.
For testing, Guidewire documents a Test Util API for development only. It requires explicit enablement and a mock policy-search configuration. Never enable it in production or describe it as a public self-service sandbox.
Read a claim: GET /claim/v1/claims/{claimId}
Returns a claim only when resource access permits it.
- App: Display the authoritative claim reference beside an exception.
- Agent: Read permitted context before judging an intake escalation.
- Workflow: Refresh the claim before presenting a proposed update for approval.
Discover activities: GET /common/v1/activities
Returns accessible activities, with behavior affected by ACLs and configuration, not necessarily only the caller's assignments.
- App: Populate an operational queue with activity IDs and owners.
- Agent: Prioritize ambiguous work within the permitted queue.
- Workflow: Poll through every response page and reconcile previously seen IDs.
Inspect one activity: GET /common/v1/activities/{activityId}
Returns details such as status, assignment, priority, and due date.
- App: Show the source activity beside its signature evidence.
- Agent: Explain whether the current task matches the received document.
- Workflow: Recheck status immediately before an approved action executes.
Read a claim's activities: GET /claim/v1/claims/{claimId}/activities
Collects activities associated with one claim.
- App: Maintain a claim-level outstanding-work view.
- Agent: Distinguish a real blocker from a routine reminder.
- Workflow: Refresh the activity collection before generating a daily brief.
Read supporting notes: GET /claim/v1/claims/{claimId}/notes
Returns claim notes; collection responses omit note bodies by default, so request body fields only when needed.
- App: Retain permitted note references as evidence for a brief.
- Agent: Summarize relevant narrative without inventing missing history.
- Workflow: Retrieve the minimum necessary fields under the retention policy.
Create follow-up work: POST /claim/v1/claims/{claimId}/activities
Creates an activity from an activity pattern, with optional overrides such as subject and priority.
- App: Present an editable proposed follow-up before submission.
- Agent: Recommend the appropriate approved pattern and explain urgency.
- Workflow: Create the activity only after the required approval and validation.
Route an activity: POST /common/v1/activities/{activityId}/assign
Reassigns an activity using documented group, user, queue, claim-owner, or assignment-rule options.
- App: Display the proposed destination and current owner together.
- Agent: Recommend a permitted destination for an unusual exception.
- Workflow: Apply the approved assignment without broadening the destination allowlist.
Complete general work: POST /common/v1/activities/{activityId}/complete
Completes an open general activity assigned to a user. Approval activities follow a different lifecycle.
- App: Hold the evidence checklist and exact proposed completion.
- Agent: Identify contradictory evidence that should block completion.
- Workflow: Complete only after approval, then verify the resulting state. This does not authorize payment.
ClaimCenter automation use cases on Major
Signature exceptions with ClaimCenter and DocuSign
Use the existing signing experience and add a cross-system exception ledger. Activity reads identify outstanding work; the approved completion operation closes only the corresponding general activity.
- App: Store claim, activity, envelope, and required-signer mappings.
- Agent: Judge ambiguous matches and draft an evidence-backed explanation.
- Workflow: Reconcile signature status, route exceptions, and require adjuster approval before completion.
Intake triage with ClaimCenter and Slack
Extend the queued FNOL triage concept with a restricted team notification. Use claim and activity reads, followed by approved activity creation and assignment. Never turn a triage classification into an autonomous coverage or payment decision.
- App: Keep the original intake reference, recommendation, and review status.
- Agent: Assess urgency and propose the next permitted task.
- Workflow: Apply approved routing and send Slack a link to the review record, not sensitive claim narratives.
Adjuster handoff briefs with ClaimCenter and Google Drive
Extend the queued activity-summary concept with an approved handoff document. Claim, activity, and note reads supply the evidence. A generated brief is a working aid, not a replacement for the claim record.
- App: Keep source references, retrieval times, and accepted brief versions.
- Agent: Summarize unresolved work and identify conflicting or incomplete information.
- Workflow: Refresh sources, request review when needed, and save an approved brief in a restricted Drive location.
These are proposed build patterns, not claims that the catalog installs a finished insurance process. For each, define the responsible adjuster, allowed outputs, and exception owner before configuring execution.
Build the architecture on Major

ClaimCenter stays authoritative while Major coordinates evidence and approved work.
Build the deterministic app first. Its managed tables should distinguish source records, incoming evidence, agent recommendations, approvals, and execution attempts. A recommendation is not an executed change, and an approved change is not necessarily a successful API write.
- Connect: IT configures the ClaimCenter custom API connector and scopes access to supporting systems.
- Persist: The app records stable source IDs and the last successful reconciliation.
- Orchestrate: A workflow fetches permitted context and retries transient reads.
- Judge: The agent classifies the exception or drafts the brief, citing source references.
- Review: The app displays the proposal, evidence, and authorized approver.
- Execute: The workflow validates fresh state, performs the approved operation, and records the outcome.
As the process becomes clear, repeatable matching and validation belong in app code. Keep the agent for cases that still need interpretation. That is Major's predictable-agent approach: reason once, run forever for the repeatable parts, with state surviving between runs rather than living only in a conversation.
A worked example with numbers
For example, suppose a claims team reviews 200 signature-related activities each week. Assume manual checking takes six minutes per activity: 200 × 6 = 1,200 minutes, or 20 hours. These are illustrative planning assumptions, not a customer result.
A proposed Major process assembles an evidence packet for each activity. Assume 160 clear cases require two minutes of adjuster review and 40 exceptions require eight minutes. That is 160 × 2 + 40 × 8 = 640 minutes, or 10 hours and 40 minutes. The modeled difference is 560 minutes, or nine hours and 20 minutes per week.
The workflow first reconciles the activity and envelope mapping. The agent explains a mismatched signer or contradictory status. The adjuster reviews the evidence and authorizes any ClaimCenter write. The app records the decision and subsequent API result separately.
Measure these assumptions in a read-only pilot before claiming savings. Track review time, exception frequency, stale-source incidents, false matches, and missed signature requirements. Include integration maintenance and supervisor review in the business case. Faster review is useful only if evidence completeness and approval quality remain acceptable.
Governance and failure handling

Read-only status can refresh directly; consequential updates wait for an authorized person.
The agent must never independently approve settlement, deny coverage, release payment, waive a required signer, or treat a similarity match as confirmed identity. A signed document establishes signature evidence, not payment authority.
Require an approver to review the exact target activity, proposed operation, source evidence, and any uncertainty. Before execution, refresh the source. If ownership, status, or relevant evidence changed, invalidate the old proposal rather than carrying its approval forward.
For duplicates, retain an event ledger and an execution ledger. Guidewire's duplicate-transaction header documentation describes GW-DBTransaction-ID: a globally unique transaction identifier for qualifying commits. A duplicate is rejected with a 400 response, not replayed as the original success. Its limitations mean it complements, rather than replaces, reconciliation.
Retry transient failures with bounded exponential backoff and jitter. Honor server-provided retry instructions when present, pause on throttling, and stop blind retries on authorization or validation failures. After an ambiguous write timeout, reconcile before retrying.
Audit records should connect the source event, agent rationale, approver identity, approved payload, execution attempt, and result. Limit sensitive narrative retention and restrict review access by role. Assign a named operational owner to unresolved failures so an exception cannot disappear into an unattended queue.
Guidewire ClaimCenter AI Apps, Agents and Workflow Automations
- Track claim-document signatures and adjuster approvals across ClaimCenter and DocuSign: This guide focuses on correlating envelopes with claim activities, retaining signer progress, and handling unmatched events. It separates signature completion from payment readiness and puts consequential updates behind adjuster review.
This is the current core implementation guide for this hub. It addresses a bounded process rather than general autonomous claims handling, making it a useful starting point for teams that need a durable exception ledger. Read it when your main problem is knowing which document is signed, which requirement remains open, and who authorized the next step.
The activity-summary and intake-triage patterns above are additional design options, not published walkthroughs listed as finished guides. Keep that distinction when planning delivery. A hub should help choose a process without implying that a queued article is an available implementation.
For surrounding systems, see the DocuSign automation hub and Slack agent hub. Use the insurance overview to compare process ownership and governance needs before expanding into another claims workflow.
Frequently asked questions
What are Guidewire ClaimCenter AI agents?
They are agents that use authorized ClaimCenter context to perform judgment tasks such as triage, summarization, and exception analysis. On Major, deterministic apps hold state and review queues, agents make contextual recommendations, and workflows coordinate retrieval, approvals, and permitted execution.
Does ClaimCenter already have built-in AI?
Yes. Guidewire documents ProNavigator for cited insurance guidance and Claim Summarization for focused in-product summaries. The Palisades release notes mark Claim Summarization as restricted availability. Verify your tenant's release and entitlements before planning around either capability.
Does Major have a native ClaimCenter connector?
No. The catalog specifies a custom API connector for ClaimCenter, with the tenant base URL and authentication configured once by IT. Credentials stay in Major's vault and access is scoped to the apps, agents, and workflows that need it.
Can ClaimCenter send webhooks instead of requiring polling?
Guidewire App Events supports outbound webhook subscriptions and at-least-once delivery. Confirm tenant availability, subscribed business events, verification requirements, and retry behavior before implementation. The public overview reviewed here does not specify signature headers or retry intervals. Scheduled Cloud API reads are an alternative starting point.
Can an agent complete an activity or release payment automatically?
This design requires human approval before a consequential activity update. Completing a general activity is distinct from approving an approval activity or releasing payment. The agent proposes and explains; an authorized person approves, and the workflow executes only the permitted operation after rechecking state.
How should teams test a ClaimCenter integration?
Use a development tenant with synthetic or approved test data. Guidewire documents a development-only Test Util API that requires explicit configuration. Test pagination, missing permissions, duplicate events, uncertain writes, stale approvals, and incomplete evidence before permitting production writes.
Sources
These official pages support the API behavior and product descriptions above. Business-flow examples and authentication documentation come from different documented release paths, so validate the deployed tenant's schemas and configuration before implementation. Treat tenant authorization and enabled product capabilities as implementation checks, not assumptions inherited from an example.
- Cloud API endpoints and path conventions
- Querying for activities and access-control behavior
- Creating activities from approved patterns
- Assigning activities to users, groups, and queues
- Closing general activities and distinct approval activities
- Querying for a claim by claim ID
- Querying for notes and retrieving narrative fields
- Pagination parameters, response links, and total-count behavior
- OAuth2 client-credentials flow for standalone services
- Standalone-service example, scopes, and resource access
- Sending authenticated calls and interpreting inaccessible resources
- App Events delivery mechanisms and ordering guarantees
- Development-only Test Util API enablement
- Preventing duplicate database transactions and documented limitations
- Palisades ClaimCenter release highlights, ProNavigator, and restricted Claim Summarization
Start building on Major
Build a ClaimCenter exception-review app that keeps evidence, agent recommendations, and adjuster approvals together. Start with read-only retrieval, then add narrowly scoped writes after validating the review and failure paths.
About the author
Ritvij Gautam is Head of Growth at Major.
Last reviewed November 20, 2026 by Ritvij Gautam.
Frequently asked questions
- What are Guidewire ClaimCenter AI agents?
- They are agents that use authorized ClaimCenter context to perform judgment tasks such as triage, summarization, and exception analysis. On Major, deterministic apps hold state and review queues, agents make contextual recommendations, and workflows coordinate retrieval, approvals, and permitted execution.
- Does ClaimCenter already have built-in AI?
- Yes. Guidewire documents ProNavigator for cited insurance guidance and Claim Summarization for focused in-product summaries. The Palisades release notes mark Claim Summarization as restricted availability. Verify your tenant's release and entitlements before planning around either capability.
- Does Major have a native ClaimCenter connector?
- No. The catalog specifies a custom API connector for ClaimCenter, with the tenant base URL and authentication configured once by IT. Credentials stay in Major's vault and access is scoped to the apps, agents, and workflows that need it.
- Can ClaimCenter send webhooks instead of requiring polling?
- Guidewire App Events supports outbound webhook subscriptions and at-least-once delivery. Confirm tenant availability, subscribed business events, verification requirements, and retry behavior before implementation. The public overview reviewed here does not specify signature headers or retry intervals. Scheduled Cloud API reads are an alternative starting point.
- Can an agent complete an activity or release payment automatically?
- This design requires human approval before a consequential activity update. Completing a general activity is distinct from approving an approval activity or releasing payment. The agent proposes and explains; an authorized person approves, and the workflow executes only the permitted operation after rechecking state.
- How should teams test a ClaimCenter integration?
- Use a development tenant with synthetic or approved test data. Guidewire documents a development-only Test Util API that requires explicit configuration. Test pagination, missing permissions, duplicate events, uncertain writes, stale approvals, and incomplete evidence before permitting production writes.