Skip to content
Cloud Mechanics
AI Claims Automation iconAI Agent

AI Claims Automation

AI workflow that reads documents, extracts data, validates and assists decisions

The challenge

Claims arrive as email attachments, scans and phone photos, and a person keys every field into the core system by hand. Cycle times stretch into weeks, cost per claim stays high, and each reviewer interprets the policy wording slightly differently. Nobody can answer afterwards how long a claim really took, or why it was decided the way it was.

The outcome

Azure AI Document Intelligence reads every document into structured data, Azure AI Search grounds the assessment in the actual policy wording, and a Microsoft Foundry agent validates coverage and recommends an outcome. Straightforward claims settle the same day. The rest reach an assessor with the evidence, the extracted fields and the reasoning already assembled.

At a glance

Type
ai agent

Next step

Move from solution to engagement.

Build This Solution

01 — Architecture

End-to-end architecture

The journey starts at the front end, flows through a secure API into the application backend, is processed by the Foundry agent with Document Intelligence and AI Search, then stored in Blob and the operational database, and surfaced back to claimants, assessors and management dashboards.

USERClaimant /Assessorweb · mobile · portalFRONT ENDAzure Static WebAppsclaim portal & uploadAPPLICATION / APIAzure APIManagementgateway · auth · SLAApp Service /Container Appsclaim lifecycle APIAI & AGENTAI DocumentIntelligenceOCR · layout · fieldsAzure AI SearchRAG over policy docsMicrosoft FoundryAgenttools · actionsAzure OpenAImodelsreasoning · summariesDATA & STORAGEAzure BlobStoragedocuments & evidenceAzure SQL /Cosmos DBclaim records & stateEXPERIENCE & INSIGHTCommunicationServicesnotify the claimantHuman-in-the-loopreviewassessor decisionPower BI / Fabricmanagement dashboardsDEVOPS & DELIVERYGitHubsource control & CIDockercontainer buildContainer Registryversioned imagesBackend releasedeploy with rollback
Figure 1 — End-to-end reference architecture, entirely on the Microsoft stack.
  • User: Claimant / Assessor
  • Front end: Azure Static Web Apps
  • Application / API: Azure API Management, App Service / Container Apps
  • AI & agent: AI Document Intelligence, Azure AI Search, Microsoft Foundry Agent, Azure OpenAI models
  • Data & storage: Azure Blob Storage, Azure SQL / Cosmos DB
  • Experience & insight: Communication Services, Human-in-the-loop review, Power BI / Fabric

02 — Workflow

Process & decision workflow

How a claim is captured, extracted, classified, enriched and assessed — then branched. High-confidence, low-risk cases proceed straight through, while low-confidence or higher-risk cases route to a human reviewer before the business decision.

1CaptureClaim form, policy reference, invoices,reports and photos2StoreOriginal files secured in Azure BlobStorage3ExtractText, tables and key-value pairs viaDocument Intelligence4ClassifyClaim form, ID, invoice, medical reportor missing evidence5EnrichPolicy, coverage and claim history viaAI Search (RAG)6AssessCoverage match, missing data, risk andsettlement probabilityConfidence& risk?Path 1 · high confidenceAuto recommendationApprove or request information, notifyand closePath 2 · low confidence or riskHuman-in-the-loop reviewAssessor validates the evidence anddecidesBusiness decisionAudit trail, extracted data anddashboardsApprove or rejectRejection returns a reasoned,refileable noticeCloseclaim resolved
Figure 2 — Capture → extraction → classification → enrichment → AI assessment → confidence branch → decision.
  1. Capture: Claim form, policy reference, invoices, reports and photos
  2. Store: Original files secured in Azure Blob Storage
  3. Extract: Text, tables and key-value pairs via Document Intelligence
  4. Classify: Claim form, ID, invoice, medical report or missing evidence
  5. Enrich: Policy, coverage and claim history via AI Search (RAG)
  6. Assess: Coverage match, missing data, risk and settlement probability
  7. Path 1 · high confidence — Auto recommendation: Approve or request information, notify and close
  8. Path 2 · low confidence or risk — Human-in-the-loop review: Assessor validates the evidence and decides

03 — Components

Key Microsoft components

The solution stays inside the Microsoft ecosystem end to end — application layer, agent runtime, knowledge, data, notifications and DevOps.

  • Azure Static Web Apps iconAzure Static Web AppsClaimant and assessor experience for submission, upload and tracking.
  • Azure App Service iconAzure App ServiceAPI and application layer for the claim lifecycle and validation.
  • Azure Container Apps iconAzure Container AppsContainerised agent APIs and long-running workflow components.
  • Azure API Management iconAzure API ManagementSecure gateway with auth, throttling, versioning and partner access.
  • Azure AI Document Intelligence iconAzure AI Document IntelligenceOCR, layout, tables and key-value extraction into structured JSON.
  • Microsoft Foundry Agent Service iconMicrosoft Foundry Agent ServiceAgent instructions, tool calls, grounding and actions.
  • Azure OpenAI models iconAzure OpenAI modelsLanguage understanding, summarisation and assessment reasoning.
  • Azure AI Search iconAzure AI SearchRAG grounding over policy documents, SOPs and supporting material.
  • Azure SQL / MySQL iconAzure SQL / MySQLOperational claim records, execution state and audit metadata.
  • Azure Cosmos DB iconAzure Cosmos DBElastic, low-latency store for agent execution and document state.
  • Azure Blob Storage iconAzure Blob StorageOriginal documents, extracted artefacts and evidence packs.
  • Azure Communication Services iconAzure Communication ServicesEmail, SMS and in-app notifications for updates and outcomes.
  • Azure Container Registry iconAzure Container RegistryVersioned image hosting with rollback for every component.
  • Power BI / Microsoft Fabric iconPower BI / Microsoft FabricManagement dashboards and downstream claims insights.

04 — AI

What the agent consumes

The capabilities the agent consumes behind the scenes, and the line it does not cross.

AI capabilities embedded in the agent

  • OCR
  • Document classification
  • Key-value extraction
  • Table extraction
  • Natural language processing
  • Retrieval-augmented generation
  • Computer vision on claim photos
  • Decision assistance
  • Risk flagging
  • Human-in-the-loop routing
  • Audit summarisation
  • Workflow orchestration

AI responsibility boundaries

The agent assists decisions — it does not silently replace accountable business approval. High-confidence, low-risk cases can be recommended for straight-through processing. Exceptions, missing evidence, unusual patterns and low-confidence outputs are routed to a reviewer together with the source documents, the extracted fields, the confidence scores and the AI explanation.

05 — Personalization

Personalization & evolving process

The same methodology applies to every agent in the catalog. Tune the industry documents, business rules, workflow states, integrations and value model — the page structure stays identical.

Industry profile

Define the target industry, regulatory context, user personas, source documents, decision owners and exception scenarios. For insurance the personas are claimant, broker, claims assessor, supervisor and management dashboard user.

06 — Impact

Key outcomes & business impact

Starting targets for the value case — validate each one against the customer baseline during discovery.

  • Document digitisation80%+Faster extraction from images and PDFs once OCR is applied at intake.
  • Claims cycle timeHoursStandard claims move from manual queues to same-day AI-assisted assessment.
  • Reviewer productivity2×Reviewers start from extracted fields, an evidence summary and a recommendation.
  • Governance readiness100%Every path stores the evidence, extracted data, AI summary and reviewer action.

Illustrative improvement index

Manual baseline = 100. Illustrative targets, not a commitment — confirm against the customer baseline.

10045Handling effort10020Digitisation time10035Audit prep effortManual baselineAI-assisted target
  • Handling effort: manual baseline 100, AI-assisted target 45
  • Digitisation time: manual baseline 100, AI-assisted target 20
  • Audit prep effort: manual baseline 100, AI-assisted target 35

07 — Deployment

Deploy this agent

Turns a claim submission into a structured, evidence-linked assessment with an explicit human-review decision - in minutes, not days. This agent ships as a versioned, evaluated package: 9 Azure resources described in Bicep, 12 scoped tools, and a blocking evaluation gate that must pass before it is considered ready.

  • Package version1.0.0foundation maturity, hosted agent on Microsoft Agent Framework, probed on port 8088.
  • Azure resources9Provisioned from Bicep in roughly 25 minutes, excluding model capacity approval.
  • Evaluation cases256 evaluators run as a blocking gate on every change.
  • Review triggers149 of them stop the agent and hand the case to a person.

What gets provisioned

Every resource below is declared in the agent's Bicep templates and deployed with a user-assigned managed identity. No key or connection string is stored in the package.

  • Microsoft Foundry account and projectHosts the agent, its model deployments and its managed identity.Microsoft.CognitiveServices/accounts
  • Azure AI SearchPolicy wording index used for grounded, cited retrieval.Microsoft.Search/searchServices
  • Azure StorageClaim document intake and immutable audit output.Microsoft.Storage/storageAccounts
  • Azure Key VaultHolds integration credentials that cannot use managed identity.Microsoft.KeyVault/vaults
  • Application Insights and Log AnalyticsTraces, metrics and the human-review audit trail.Microsoft.Insights/components

Optional: Azure AI Document Intelligence.

Deployment parameters

Collected before provisioning and validated against the manifest. Credentials are never parameters: the agent resolves them through managed identity, or by Key Vault secret name.

Identity

  • Tenant identifierrequiredData isolation boundary. Every tool call is scoped to it.

Infrastructure

  • Azure regionuaenorthRegion for the Foundry project and supporting resources.
  • Environmentdev

Model

  • Document understanding model deploymentgpt-5.4-mini
  • Case reasoning model deploymentgpt-5.4

Thresholds

  • Extraction confidence floor0.7Below this score the case is routed to a human.
  • Assessment confidence floor0.7
  • Fast-track recommendation floor0.85Both confidence scores must reach this before fast track is recommended.
  • Exceptional claim amount50000Claims above this amount always reach a human, whatever the confidence.

Locale

  • Default currencyAED
  • Dates are day-firsttrueResolves ambiguous numeric dates such as 03/04/2026.
  • Policy number format[A-Z0-9]{6,20}Regular expression a policy number must match before lookup.

Integrations

  • Policy system credential secret namerequiredKey Vault secret NAME holding the policy-system credential, used only when the integration cannot authenticate with managed identity. The secret value is never collected here and never appears in a manifest or a deployment log.

Deployment lifecycle

A deployment moves through an explicit state machine. Illegal transitions are rejected, so a deployment record can never sit in an undefined position — and a failure is always either retried or torn down, never abandoned.

  1. 01 Requested
  2. 02 Validating request
  3. 03 Waiting for configuration
  4. 04 Provisioning Azure resources
  5. 05 Configuring Foundry project
  6. 06 Deploying agent
  7. 07 Connecting tools
  8. 08 Connecting knowledge
  9. 09 Running deployment validation
  10. 10 Running evaluation gate
  11. 11 Ready
  12. 12 Degraded
  13. 13 Failed
  14. 14 Updating
  15. 15 Decommissioning
  16. 16 Decommissioned

Tools and their blast radius

12 tools across 1 toolbox. 5 can write, and every write is scoped, audited and reversible.

  • get_policy
  • get_customer
  • search_policy_wording
  • get_claim_status
  • retrieve_case_history
  • validate_policy_coverage_inputs
  • calculate_claim_amount
  • create_claim · write
  • update_claim · write
  • attach_document · write
  • create_review_task · write
  • request_missing_information · write

What this agent will never do on its own

  • Approving, denying or settling a claim
  • Authorising, scheduling or releasing any payment
  • Determining that a loss is covered or excluded as a final answer
  • Communicating a claim outcome to a claimant
  • Closing or withdrawing a claim
  • Changing policy terms, limits or excesses
  • Making or recording an allegation of fraud
  • Requesting or storing data belonging to another tenant

Evidence before it ships

The evaluation gate is blocking: a regression on any evaluator stops the release rather than documenting it.

  • golden-claims13 casesRepresentative correct-path cases with expected routing.
  • adversarial-claims12 casesPrompt injection, authority-boundary and out-of-scope probes.

Evaluators

  • routing_accuracy
  • completeness_accuracy
  • finding_recall
  • citation_grounding
  • injection_containment
  • authority_boundary

Identity, isolation and network

  • Agent identityuser-assigned-managed-identity. No shared keys, no embedded credentials.
  • Public network accessconfigurable. Private endpoints available for Azure AI Search, Azure Storage, Azure Key Vault.

Role assignments

  • Foundry UserFoundry projectRun the agent and read its project configuration.
  • Search Index Data ReaderAzure AI Search serviceRetrieve policy wording for grounded citations. Read-only by design.
  • Storage Blob Data ContributorClaim intake containerRead submitted documents and write the immutable assessment record.
  • Key Vault Secrets UserKey VaultResolve integration credentials that cannot use managed identity.
  • Monitoring Metrics PublisherApplication InsightsEmit traces and metrics.

What you can see once it is running

Traced with opentelemetry and exported to Application Insights. Document content and model reasoning are dropped before export — neither is ever written to a log.

  • claims.processed
  • claims.review_required_rate
  • claims.confidence.extraction
  • claims.confidence.assessment
  • claims.recommended_action
  • claims.injection_detected
  • claims.tool_failure_rate
  • claims.processing_duration_ms

Cost drivers

  • Model inference for document understanding
  • Hosted-agent container compute
  • Azure AI Search unit hours
  • Document Intelligence pages analysed

Supported regions

  • uaenorth
  • westeurope
  • swedencentral
  • eastus2

Getting it deployed

This agent is packaged as a transactable Azure Marketplace managed application: the same Bicep infrastructure and container image below, provisioned into your own subscription through a portal wizard, with the licence fee billed against your existing Azure agreement. Plans, metered usage and an estimate of your own Azure consumption are set out under plans and pricing.

Published by Cloud Mechanics · Owner Cloud Mechanics - AI Solutions · standard support

See plans and pricing

08 — Plans and pricing

Start free. Pay only when the agent delivers value.

Cloud Mechanics AI Insurance Claim Agent is licensed as a Microsoft Marketplace subscription and deployed into your own Azure subscription as a managed application. The agent runs in your tenant, inside a managed resource group we operate on your behalf, in the region you choose. Your claim documents, policy corpus, search index and logs are created there and stay there, and no claim content ever reaches Cloud Mechanics. Model inference runs on a globally routed Azure OpenAI deployment and may be processed in another Azure region. Microsoft bills the licence fee below against your existing Azure agreement; the Azure resources it provisions are billed separately, to you, at your own rates.

These prices are not yet final

The figures below are working numbers used to build and test the offer end to end. They have not been through commercial review and should not be quoted. Talk to us for current pricing.

Free for 30 days, up to 250 claims

30 days rather than a fortnight, because an enterprise claims deployment spends its first two weeks in security review, data preparation and integration testing — a shorter trial expires before a single real claim is processed. The trial is the complete product, not a demo: the same pipeline, the same citations, the same review routing. You pay only your own Azure consumption.

  • Full claims assessment workflow
  • Document extraction
  • Retrieval over your own policy wording, with citations
  • Agent assessment with evidence links
  • Human-review routing
  • Complete audit trail
  • Basic dashboard

Two bills, not one

What we charge

The plan's monthly licence fee, plus any usage beyond the included quantity. Billed by Microsoft through the marketplace and settled on your existing Azure invoice, from which Microsoft retains a 20% marketplace service fee.

What Azure charges

The 9 resources the template provisions, consumed in your subscription at your own negotiated rates. We take no margin on it and never see it. Roughly $3,800 per month at Business volume.

  • TrialStart here

    The complete agent on your own claims, free for 30 days.

    $0/ 30 days

    Azure resources billed separately to your own subscription.

    Included in the trial
    • 250 claims (claim processed)
    Beyond that
    • Nothing. The trial stops at its limit rather than billing you, so an evaluation can never produce an invoice.
    Support

    Onboarding assistance during the trial period

    • Full assessment pipeline with citations and review triggers
    • Document extraction and policy retrieval over your own wording
    • Human-review routing and complete audit trail
    • Deploys into your subscription, in the region you choose; we never hold your data
    • You pay only your own Azure consumption
  • StarterMost deployed

    Light production for a single claims team.

    $199/ month licence fee

    Azure resources billed separately to your own subscription.

    Included each month
    • 2,000 claims (claim processed)
    Beyond that
    • $0.08 per claim
    Support

    Business hours, next-business-day response

    • Everything in Trial, without the 250-claim cap
    • 2,000 claims included each month
    • Blocking evaluation gate run on every agent release
    • One region, one environment
  • Business

    Production claims handling for a single operating entity.

    $699/ month licence fee

    Azure resources billed separately to your own subscription.

    Included each month
    • 10,000 claims (claim processed)
    Beyond that
    • $0.05 per claim
    Support

    Business hours, four-hour response, named escalation path

    • Everything in Starter
    • 10,000 claims included each month
    • Two regions, three environments (dev, staging, production)
    • Assessment audit trail retained for your configured retention period
  • Enterprise

    Multi-entity deployment with network isolation and 24x7 cover.

    $1,999/ month licence fee

    Azure resources billed separately to your own subscription.

    Included each month
    • 50,000 claims (claim processed)
    Beyond that
    • $0.03 per claim
    Support

    24x7, one-hour response for production-down

    • Everything in Business
    • 50,000 claims included each month
    • Private endpoints and public network access disabled
    • Unlimited regions, environments and integrations
    • Quarterly review of rule thresholds and review-trigger rates
  • High Volume

    Negotiated terms for 100,000+ claims a month.

    Private offer

    Target $0.02 - $0.03 per claim, agreed per deal.

    Azure resources billed separately to your own subscription.

    Beyond that
    • Agreed as part of the negotiated terms.
    Support

    24x7 with agreed service levels

    • Everything in Enterprise
    • Negotiated per-claim rate at 100,000+ claims a month
    • Committed-volume and multi-year terms available
    • Requested through the marketplace listing, agreed as a private offer

These are launch prices, set to win reference deployments rather than to maximise revenue per claim. They will rise for new customers as the product matures; anyone who starts on them keeps their rate for an agreed period rather than being repriced at renewal.

What is metered

Usage is counted by the agent itself and reported to the marketplace once an hour. What is not counted matters as much as what is.

  • Claim processedper claimOne unique claim that successfully completes the AI assessment workflow. A claim routed to human review still counts, because the agent completed its assessment and the work was done either way.

Never billed

Tokens, model calls, tool calls, retrieval queries and document pages are measured for our own capacity planning and never appear on your invoice. Billing them would tie what you owe to how we happened to build the agent, so a shorter prompt would change your bill and optimising the agent would cut our revenue. The meter above is the only thing you pay for.

Your Azure bill, estimated

Approximate list-price cost per month at the Business plan's 10,000 included claims.

  • Model inference$2,200
  • Hosted agent container compute$260
  • Azure AI Search$250
  • Document understanding$900
  • Storage, Key Vault, Log Analytics, Application Insights$190

Excludes any Azure commitment discount the customer already holds, which in practice moves these numbers materially. Model inference dominates and scales with document length, not claim count. This is the customer's own spend at their own rates - Cloud Mechanics takes no margin on any of it.

Getting it into production is a separate engagement

The licence buys the software. Integrating it with your claims system, preparing your policy corpus and calibrating the rule thresholds against your own historical claims is project work, and pricing it as though it were included would either inflate the subscription for everyone or quietly fund consulting out of a software fee. Both are quoted separately and neither is required to start.

Implementation

$5,000 – $25,000

One-off, fixed scope. Scope driven by the number of integrations, policy corpus preparation, claim document variety and calibration of the rule thresholds.

Managed service (optional)

$1,500/ month

Monitoring of assessment quality and review-trigger rates, quarterly threshold tuning, and version upgrades regression-tested before they reach you.

Getting it deployed

The offer is built and the deployment package is ready: infrastructure as Bicep, a portal wizard, metered billing wired to the marketplace, and a blocking evaluation gate on every release. The listing itself is still going through Microsoft's publishing review, so the last step today is a conversation about region, capacity and the systems this agent will read from.

Published by Cloud Mechanics · Owner Cloud Mechanics - AI Solutions · standard support

Talk to us about deploying it

Related & recommended

Derived automatically from our solution knowledge graph.

Related professional services

How we design, build and secure it.

AI Agent Development iconService

AI Agent Development

AI Agent Development delivered by Cloud Mechanics certified experts.

AI Data Engineering iconService

AI Data Engineering

AI Data Engineering delivered by Cloud Mechanics certified experts.

AI Governance iconService

AI Governance

AI Governance delivered by Cloud Mechanics certified experts.

AI Integration iconService

AI Integration

AI Integration delivered by Cloud Mechanics certified experts.

Related quick wins

Ready-made Azure AI to start fast.

OCR iconQuick Win

OCR

Automated text extraction from images and PDFs

Structured Data Extraction iconQuick Win

Structured Data Extraction

Turn layout into data

Insurance Document Processing iconQuick Win

Insurance Document Processing

Extract insurance forms

Identity Document Processing iconQuick Win

Identity Document Processing

Extract passport/ID fields

Technologies

What powers this solution.

Azure AI iconTechnology

Azure AI

Managed AI services for vision, speech, language and document.

Azure OpenAI iconTechnology

Azure OpenAI

Enterprise access to GPT models with governance.

Microsoft Copilot Studio iconTechnology

Microsoft Copilot Studio

Low-code platform to build custom copilots and agents.

Azure Functions iconTechnology

Azure Functions

Serverless compute for event-driven workloads.

Related managed services

Keep it running and optimised.

AI Managed Services iconManaged

AI Managed Services

AI Managed Services from our UAE-based 24/7 Cloud Operations Center.

Azure Managed Services iconManaged

Azure Managed Services

Azure Managed Services from our UAE-based 24/7 Cloud Operations Center.

Cloud Managed Services iconManaged

Cloud Managed Services

Cloud Managed Services from our UAE-based 24/7 Cloud Operations Center.

DevOps Managed Services iconManaged

DevOps Managed Services

DevOps Managed Services from our UAE-based 24/7 Cloud Operations Center.

Ready to move from challenge to solution?

Talk to a Cloud Mechanics expert or build your solution in minutes.

Talk to an ExpertBuild This Solution