Ellipse Gradient for Header

Azure AI Document Intelligence Integration Guide

Azure AI Document Intelligence integrates through REST APIs that asynchronously analyze document content or URLs with prebuilt and custom models.

Azure AI Document Intelligence integration options at a glance

Azure AI Document Intelligence provides HTTPS REST APIs for submitting documents or document URLs, selecting prebuilt or custom models, retrieving model information, and polling long-running analysis operations. Analyze requests return an operation location that the client checks until processing completes; the service does not provide a general-purpose webhook mechanism for completed analysis. Authentication uses either an Azure resource key or Microsoft Entra ID bearer tokens. Martini can consume these REST endpoints, store credentials in secrets, orchestrate polling with retries and timeouts, and map extracted fields, tables, confidence values, and page data into downstream applications or databases.

Integration pointSupported by Azure AI Document Intelligence?Common use casesHow Martini supports it
REST APIsYesSubmit documents for analysis, retrieve asynchronous results, and manage document models. API capabilities depend on the selected API version.Martini can consume the Azure REST API, configure request headers and bodies, capture operation locations, and map response payloads in workflows.
Asynchronous analysisYesAnalyze operations run asynchronously and return an operation location or identifier that the caller polls until completion.Martini can persist the operation reference, poll at a controlled interval, enforce a timeout, and branch for success, failure, or expiration.
Binary document inputYesSupported analyze operations can receive document content in the HTTP request body for PDFs, images, and other supported document formats.Martini can receive or retrieve file content, send it as an HTTP request body, and correlate the analysis with the source document.
URL-based document inputLimitedSupported analyze operations can receive a document URL when Azure can access the referenced file. The service is not a general file-storage API.Martini can pass controlled document URLs, validate their availability, and use a document-bytes approach when private access cannot be provided safely.
Prebuilt and custom modelsYesPrebuilt models support scenarios such as invoices, receipts, reading, and layout; custom models address organization-specific extraction or classification needs.Martini can make the model identifier configurable, submit the selected model request, and validate expected fields in the returned result.
Model management APIsYesREST APIs support operations such as listing, retrieving, composing, copying, and deleting models, subject to API version and model capability.Martini can orchestrate model metadata synchronization or controlled model-management workflows through API calls and approval rules.
AuthenticationYesThe service supports an Azure resource key in the Ocp-Apim-Subscription-Key header and Microsoft Entra ID OAuth 2.0 bearer tokens with appropriate Azure role assignments.Martini can use environment configuration and secrets for keys or tokens, keeping credentials outside workflow definitions and applying secure request configuration.
Webhooks / outbound callbacksNot confirmedNo general-purpose Document Intelligence webhook or callback mechanism was found. The documented completion pattern is polling the operation URL.Martini should implement polling, backoff, timeout, and retry logic rather than assume Azure will call a Martini endpoint.

How Azure AI Document Intelligence exposes data and business events

Azure AI Document Intelligence REST APIs

The primary integration mechanism is the Azure AI Document Intelligence HTTPS REST API. It supports submitting analysis requests, retrieving results, and working with document models. Capabilities and response structures vary by API version and selected model.

Martini implementation pattern

Martini implementation pattern: a workflow or Martini API receives a document or document URL, builds the Azure request with the selected model, authenticates using a resource key or Microsoft Entra ID token, and maps the response into the next processing stage.

Implementation sequence

Receive a document or document URL
Select and validate the Azure model identifier
Submit the analysis request to the Azure REST endpoint
Capture the returned operation location
Map the completed analysis result to the target model

Asynchronous analysis operations

Document analysis is long-running rather than an immediate request-response operation. Azure returns an operation location or identifier, and the caller polls that resource until the analysis succeeds or fails.

Martini implementation pattern

Martini implementation pattern: the workflow stores the operation reference with the source-document correlation ID, waits according to a configured interval, polls with backoff, and branches on in-progress, successful, invalid, throttled, or failed responses.

Implementation sequence

Submit the document analysis request
Persist the operation reference and source correlation ID
Wait for the configured polling interval
Retrieve the operation status
Repeat polling until completion or timeout
Process the successful result or route the failure

Document content and URL input

Supported analyze operations can accept document content in the HTTP body or a document URL. URL-based processing requires Azure to be able to retrieve the referenced document, while binary input avoids dependency on external file accessibility.

Martini implementation pattern

Martini implementation pattern: the workflow chooses binary or URL input according to document location and security requirements, validates the source, and keeps the original source identifier alongside the analysis operation.

Implementation sequence

Identify the source document and its access method
Validate the file or document URL
Send document bytes or the supported document URL
Associate the operation with the source identifier
Preserve the input reference for audit and reprocessing

Prebuilt and custom models

Azure supplies prebuilt models for scenarios such as reading, layout, invoices, and receipts, while custom models support organization-specific document structures and extraction requirements.

Martini implementation pattern

Martini implementation pattern: model selection is externalized as configuration, and the workflow applies model-specific mappings and validation rules before writing results to a business system.

Implementation sequence

Determine the document type and required extraction model
Load the configured model identifier
Submit the document to the selected model
Validate expected fields and structures
Route unsupported or incomplete results for review

Model management APIs

Azure provides REST operations for model discovery and management, including listing, retrieving, composing, copying, and deleting models where supported by the API version and model capability.

Martini implementation pattern

Martini implementation pattern: a controlled Martini workflow can synchronize model metadata or execute approved model-management actions while recording API versions, model IDs, and deployment context.

Implementation sequence

Request the available model metadata
Compare model identifiers with integration configuration
Apply an approved model-management action
Record the resulting model state
Notify dependent workflows of configuration changes

Common Azure AI Document Intelligence integration patterns

Pattern 1: Extract invoices for accounts payable

When to use this pattern

Use this pattern when invoice PDFs or images must be converted into validated financial data. The workflow submits each invoice to the prebuilt-invoice model, checks extracted values and confidence, and sends only acceptable results to a finance application.

Integration direction
Invoice source
Martini
Azure AI Document Intelligence
NetSuite
Example Mapping
Azure AI Document Intelligence FieldCanonical FieldTarget Field
invoiceNumberinvoiceNumbertranId
invoiceDateinvoiceDatetranDate
vendorNamesupplierNamevendor
invoiceTotaltotalAmounttotal
Martini implementation pattern

Martini receives a file or document URL, submits it to prebuilt-invoice, polls the operation, normalizes dates and amounts, and validates required fields and totals. Low-confidence or inconsistent results are routed for manual review; transient Azure or target-system failures are retried with correlation and duplicate checks.

Martini capabilities used
  • REST API consumption
  • Workflows
  • Asynchronous orchestration
  • Data mapping
  • Validation and business rules
  • Error handling

Pattern 2: Process documents from Azure Blob Storage

When to use this pattern

Use this pattern when an Azure-based application places PDFs or images in Blob Storage and the organization needs extracted metadata or structured results in a database or content system.

Integration direction
Azure Blob Storage
Martini
Azure AI Document Intelligence
PostgreSQL
Example Mapping
Azure AI Document Intelligence FieldCanonical FieldTarget Field
contentrawDocumentContentdocument_content
documentTypedocumentTypedocument_type
fieldsextractedFieldsextracted_json
confidenceoverallConfidenceconfidence_score
Martini implementation pattern

A scheduled Martini workflow processes a controlled set of new document URLs, submits each URL or its retrieved bytes, and persists the raw response, normalized fields, operation state, and source identifier. A content hash prevents duplicate analysis, while throttling and polling backoff protect the Azure resource.

Martini capabilities used
  • Scheduled workflows
  • REST API consumption
  • JSON handling
  • Data mapping
  • Database integration
  • Retry and idempotency

Pattern 3: Classify contracts and route for review

When to use this pattern

Use this pattern when a portal or business application receives contracts or forms that must be classified and routed according to document type, extracted values, and confidence.

Integration direction
Document portal
Martini
Azure AI Document Intelligence
Microsoft SharePoint
Example Mapping
Azure AI Document Intelligence FieldCanonical FieldTarget Field
documentTypeclassificationcontentType
fields.effectiveDateeffectiveDateEffectiveDate
fields.partyNamecounterpartyNameCounterparty
confidenceextractionConfidenceReviewStatus
Martini implementation pattern

Martini submits the document to a custom classification or extraction model, polls to completion, and applies rules for required fields and confidence thresholds. Accepted documents are updated in SharePoint, while uncertain classifications are placed in a review path with the raw result retained for audit.

Martini capabilities used
  • API-led workflows
  • Model selection
  • Data transformation
  • Business rules
  • Human-review routing
  • Audit persistence

Pattern 4: Extract forms for case processing

When to use this pattern

Use this pattern when identity documents, applications, or regulated forms arrive through an upstream application and their structured values must populate a case-management system.

Integration direction
Intake application
Martini
Azure AI Document Intelligence
ServiceNow
Example Mapping
Azure AI Document Intelligence FieldCanonical FieldTarget Field
documentTypeformTypeu_document_type
fields.customerNamecustomerNamecaller_id
fields.addressaddressu_address
fields.confidenceconfidenceu_extraction_confidence
Martini implementation pattern

Martini exposes or consumes an intake API, submits the document to the relevant prebuilt or custom model, and maps fields, pages, and confidence values into ServiceNow. Validation rules distinguish missing or low-confidence extraction from transport failures, and retries use a deterministic source ID to avoid duplicate cases.

Martini capabilities used
  • REST API exposure
  • REST API consumption
  • Workflow orchestration
  • Field mapping
  • Validation
  • Duplicate prevention
  • Error handling

Applications commonly integrated with Azure AI Document Intelligence

Azure AI Document Intelligence is commonly placed between document repositories, intake applications, and business systems that need structured extraction. These integrations use the relevant product APIs or storage interfaces alongside the Document Intelligence REST API; Martini can coordinate the end-to-end workflow without implying a native vendor connector.

Application Scenario Direction Martini Pattern
Azure Blob Storage Store incoming PDFs and images, provide document URLs for analysis, and retain original files alongside extracted results. Azure Blob Storage → Martini → Azure AI Document Intelligence A scheduled or upstream-triggered Martini workflow identifies new blob URLs, submits each URL to the selected model, polls the operation, and writes the result and processing status back to storage or a downstream system.
Microsoft SharePoint Extract metadata and structured content from invoices, contracts, forms, and scanned files stored in SharePoint. Microsoft SharePoint → Martini → Azure AI Document Intelligence Martini retrieves eligible files through Microsoft APIs, submits document bytes or an appropriately accessible URL, maps the analysis result, and updates SharePoint metadata or routes the result to another application.
OneDrive Process user-submitted documents while preserving the source file and associating extracted metadata with the original document. OneDrive → Martini → Azure AI Document Intelligence A Martini workflow obtains the file through Microsoft APIs, submits it to a prebuilt or custom model, applies confidence and required-field rules, and stores the result with the source identifier.
Microsoft Dynamics 365 Finance Extract invoice fields and submit validated values for accounts-payable processing. Azure AI Document Intelligence → Martini → Microsoft Dynamics 365 Finance Martini submits invoices to the prebuilt-invoice model, polls for completion, normalizes invoice values, validates totals and required fields, and calls Dynamics APIs only when the result meets configured rules.
NetSuite Automate invoice and receipt capture for financial processing and reduce manual data entry. Azure AI Document Intelligence → Martini → NetSuite A Martini workflow maps invoice number, dates, vendor, amounts, and confidence values into the NetSuite integration model, routes low-confidence documents for review, and retries transient API failures.
Salesforce Attach extracted document data to Accounts, Contacts, Cases, or custom objects. Azure AI Document Intelligence → Martini → Salesforce Martini correlates the source document with a Salesforce object, maps extracted fields and document references, applies duplicate checks, and invokes Salesforce APIs after validation.
ServiceNow Extract data from forms and supporting documents and populate Cases, Requests, or other ServiceNow records. Azure AI Document Intelligence → Martini → ServiceNow Martini receives or retrieves the document, submits it for analysis, maps the structured result into ServiceNow fields, and sends low-confidence or incomplete submissions to a manual review path.
SAP S/4HANA Capture invoice or logistics-document fields and send validated data into SAP business processes. Azure AI Document Intelligence → Martini → SAP S/4HANA Martini orchestrates analysis, converts extracted values to the SAP integration model, checks mandatory fields and business rules, and calls approved SAP APIs or middleware interfaces with correlation and retry handling.

How to build a Azure AI Document Intelligence integration in Martini

Objective

Configure the Azure endpoint and authentication method without embedding credentials in workflow definitions.

Instructions in Martini

  • Choose an Azure resource key or Microsoft Entra ID authentication.
  • Store keys, client credentials, tokens, and endpoints in environment configuration or secrets.
  • Confirm Azure role assignments and regional network access where Microsoft Entra ID is used.

Objective

Select how Martini receives or discovers documents before analysis begins.

Instructions in Martini

  • Expose a Martini REST API for upstream document intake when real-time submission is required.
  • Use a scheduled workflow to process controlled sets of document URLs or pending source records.
  • Do not assume Azure AI Document Intelligence will send a completion webhook.

Objective

Send the document to the appropriate Azure model using supported binary or URL input.

Instructions in Martini

  • Select a configurable prebuilt or custom model ID.
  • Validate the document source and required metadata.
  • Submit document bytes or a supported document URL through the Azure REST API.
  • Capture the operation location and source correlation ID.

Objective

Orchestrate the long-running Azure operation until a result or terminal failure is available.

Instructions in Martini

  • Persist the operation reference with the source-document identifier.
  • Poll at an appropriate interval with backoff.
  • Enforce a maximum processing timeout.
  • Branch on in-progress, success, invalid input, throttling, authorization, and service errors.

Objective

Convert Azure analysis structures into a canonical model and apply extraction-quality rules.

Instructions in Martini

  • Map documents, fields, tables, cells, pages, spans, and confidence values.
  • Normalize dates, amounts, addresses, and document identifiers.
  • Validate required fields, totals, model expectations, and confidence thresholds.
  • Preserve raw results when audit or reprocessing is required.

Objective

Deliver validated data and document references to the target application or database.

Instructions in Martini

  • Call the target application API or write to a configured database.
  • Use deterministic source identifiers and correlation IDs to prevent duplicates.
  • Route low-confidence or incomplete results to a review process.

Common Azure AI Document Intelligence data objects used in integrations

ObjectTypical UseCommon target systemsMartini handling
Document modelsIdentify the extraction or classification model, such as prebuilt-invoice, prebuilt-receipt, prebuilt-read, prebuilt-layout, or a custom model ID.Model configuration stores, document intake applications, and workflow configuration repositoriesMartini treats the model ID as configurable workflow input, can retrieve model metadata through REST APIs, and validates that the selected model matches the expected output.
Analyze requestsSubmit document bytes or a document URL together with the selected model and analysis parameters.Azure AI Document Intelligence and upstream file or content applicationsMartini builds the request, adds authentication and correlation data, sends the content or URL, and stores the returned operation reference without duplicating submissions.
Analyze operationsRepresent long-running document analysis and expose the operation status and final result location.Martini workflow state, monitoring stores, and downstream processing queuesMartini persists the operation location or identifier, polls with backoff, applies timeout rules, and resumes or retries according to the stored processing state.
DocumentsRepresent analyzed documents and their document type, confidence, spans, pages, and extracted content.Databases, content systems, case-management applications, and business applicationsMartini maps document-level results into a canonical structure, preserves source identifiers, and optionally stores the raw response for audit and reprocessing.
FieldsCarry named extracted values such as invoice number, invoice date, vendor name, customer name, and total amount, often with normalized values and confidence.NetSuite, SAP S/4HANA, Microsoft Dynamics 365 Finance, Salesforce, and ServiceNowMartini transforms field values, checks required fields and confidence thresholds, normalizes dates and amounts, and routes exceptions for review.
Tables and cellsRepresent structured rows, columns, cells, text, bounding regions, and confidence information extracted from documents.Financial systems, databases, analytics stores, and review applicationsMartini iterates through pages, tables, rows, and cells, converts them to target line-item structures, and handles multi-page results without assuming a single table.

Authentication and security considerations

Authentication options

Azure AI Document Intelligence supports an Azure resource key sent in the Ocp-Apim-Subscription-Key header or a Microsoft Entra ID OAuth 2.0 bearer token. Entra ID access requires an appropriate role assignment on the Document Intelligence resource.

Credential protection

Martini workflows should load keys, client credentials, tokens, and endpoints from environment configuration or secrets rather than embedding them in workflow definitions. Protect document URLs, raw analysis results, and extracted personal or financial information.

Network and data protection

  • Verify the Azure resource region, endpoint, firewall rules, private networking, and managed identity configuration before deployment.
  • Use appropriately scoped and time-limited document URLs when URL-based input is selected.
  • Restrict logs and retained payloads so sensitive document content is not exposed unnecessarily.

Operational considerations for Azure AI Document Intelligence integrations

Asynchronous processing

Analyze requests return an operation location and require polling. Persist the operation reference, use a controlled interval with backoff, and enforce a timeout.

Throttling and retries

Azure quotas and throttling can affect submission and polling. Retry transient responses such as HTTP 429 with backoff, while separating authentication, invalid-input, and permanent model errors from retryable failures.

Idempotency and result size

Use a source identifier, content hash, or external correlation ID to prevent duplicate analysis. Large documents can produce substantial page, table, polygon, span, and cell data, so mappings should support multi-page results and configurable retention.

Model and schema changes

Configure the API version and model ID explicitly. Validate expected fields because model capabilities, regional availability, and response schemas can vary. Test prebuilt and custom model changes before production deployment.

Extraction quality

A technically successful analysis is not necessarily business-valid. Apply required-field, total, date, identifier, and confidence rules, and route incomplete or low-confidence results for review.

Why use Martini instead of scripts or point-to-point integrations?

Orchestration beyond a single API call

Scripts can call the Azure endpoint, but enterprise processing also requires intake, asynchronous polling, timeouts, retries, validation, target-system writes, duplicate prevention, and audit handling. Martini organizes these concerns in maintainable workflows.

Reusable integration logic

Martini can expose a controlled REST API for document intake, consume Azure REST endpoints, and reuse mappings and business rules across invoice, form, contract, and identity-document processes.

Reliable data movement

  • Centralize secrets and environment-specific configuration.
  • Map fields, tables, pages, and confidence values into different target models.
  • Separate transport failures from extraction-quality exceptions.
  • Persist operation state and raw results when resumability, auditability, or reprocessing is required.

Frequently asked questions

How can Azure AI Document Intelligence be integrated with enterprise systems?

It can be integrated through its HTTPS REST APIs. A client submits document bytes or a supported document URL to a prebuilt or custom model, receives an asynchronous operation reference, polls for completion, and maps extracted text, fields, tables, pages, and confidence values into downstream applications or databases.

Can Martini integrate with Azure AI Document Intelligence?

Yes. Martini can consume the Azure AI Document Intelligence REST API, submit document content or URLs, poll asynchronous analysis operations, and map the results into enterprise applications. No native Martini connector is documented in the supplied sources.

Do I need a connector to integrate Azure AI Document Intelligence with Martini?

No. A dedicated Azure AI Document Intelligence connector is not required. Martini can use the service's native REST API, binary or URL-based document input, resource-key or Microsoft Entra ID authentication, and asynchronous operation polling.

Is there any extra Lonti cost to integrate Azure AI Document Intelligence with Martini?

Lonti does not charge an additional per-connector or per-vendor fee to integrate Azure AI Document Intelligence. The integration is subject to the provisioned capacity of the Martini environment. Separate costs may apply from Microsoft Azure, infrastructure providers, or other third-party systems based on subscription, usage, and deployment model.

Which Azure AI Document Intelligence integration methods should be used?

The primary method is the versioned Azure REST API for analysis and model management. Use document bytes when the integration controls the file content, or a document URL when Azure can access the file securely. Prebuilt models suit standardized documents, while custom models address organization-specific structures.

Does Azure AI Document Intelligence provide webhooks or callbacks for completed analysis?

No general-purpose webhook or outbound callback mechanism was found in the supplied documentation. The documented pattern is to poll the operation location returned by the analyze request. Martini can implement polling, backoff, timeout, and failure handling in a workflow.

How does synchronization work for Azure AI Document Intelligence results?

Synchronization is generally request-driven rather than event-driven. Martini correlates the source document with an Azure operation, polls until completion, and writes the normalized result to a target system. A source ID, content hash, or external correlation ID can prevent duplicate submissions and allow processing to resume.

How does Martini handle mapping, errors, and retries?

Martini can map nested fields, tables, cells, pages, and confidence values into a canonical or target model, then apply validation and business rules. Workflows can distinguish transport errors from low-quality extraction, retry transient failures such as throttling with backoff, enforce timeouts, preserve raw responses, and route exceptions for review. Martini can also expose a REST API façade for controlled document intake.

Related Martini documentation

API Workflows

Data Processing

Security

Reliability

Build reliable Azure AI Document Intelligence integrations with Martini

Use Martini to orchestrate Azure document analysis, asynchronous polling, validation, mapping, and delivery to the systems that run your business.

EXPLORE MARTINI
BOOK A DEMO