Workflows

Every document you send to Holofin is processed by a workflow: a configured sequence of steps that classify, segment and extract data from it. This page covers nodes, routing, and running a workflow through the API.

What are Workflows?

A workflow is a reusable document processing pipeline that you configure once and execute via a single API call. Instead of calling separate APIs for classification, segmentation, and extraction, you submit a document to a workflow and receive structured results.

A workflow can combine these steps:

Classification: identify the document type (bank statement, invoice, tax form)
Segmentation: split multi-document PDFs into individual documents
Extraction: pull structured data based on document type
Validation: optional human review gates for quality control

How Workflows Work

When you submit a document to a workflow, it flows through connected nodes. Each node performs a specific task and routes the document to the next step based on configurable conditions.

Document

Fraud Analysis

Classify

Segment

Extract

Results

A typical workflow: classify the document, segment if needed, extract structured data

Workflow Nodes

Each step in a workflow is called a node. Nodes are connected by edges that define the execution order and routing conditions.

Node Type	Purpose	Example
Fraud Analyzer	Runs forensic fraud analysis on the document	Detect tampering, metadata anomalies, suspicious edits
Classifier	Categorizes documents into types	Identify as "Bank Statement" or "Invoice"
Segmenter	Splits multi-document packages	Extract individual forms from a tax package
Extractor	Extracts structured data	Pull transactions from a bank statement
Human Review	Pauses for human approval	Verify low-confidence extractions

Conditional Routing

Workflows support branching logic to handle different scenarios. Connect nodes with conditional edges to route documents based on classification results, confidence scores, or other criteria.

Workflow branching logic example showing classifier routing to different paths

Condition Type	Description	Use Case
Classification match	Route based on detected document type	Send invoices to invoice extractor, statements to statement extractor
Confidence threshold	Route based on AI confidence score	Send low-confidence results to human review
If/Else branching	Split flow into multiple paths	Process single-page vs multi-page documents differently
Default fallback	Catch-all when no conditions match	Route unrecognized documents to manual processing

Configure routing in the workflow builder by clicking on edges between nodes and setting conditions. No code is required.

Executing a Workflow

To process a document, you create a workflow run via the API. The run tracks the execution state as your document moves through each node.

Create a workflow runPOST

curl -X POST https://app.holofin.ai/api/v1.1/workflow-runs/ \
  -H "X-API-Key: $HOLOFIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "workflow_id": "wf_abc123",
    "document_b64": "JVBERi0xLjQK...",
    "document_name": "mydocument.pdf"
  }'

See the Workflow Runs API for complete documentation on request parameters, response format, and managing workflow executions.

Workflow Status

A workflow run progresses through these states:

pendingrunningcompleted

Runs may also be paused (waiting for approval), failed, or cancelled

Getting Results

Once a workflow run completes, you can retrieve the aggregated results from all extraction nodes:

Get workflow run resultsGET

curl https://app.holofin.ai/api/v1.1/workflow-runs/wr_def456 \
  -H "X-API-Key: $HOLOFIN_API_KEY"

See the Get Workflow Run API for complete response format and result structure.

Using Webhooks

Instead of polling for status, configure a webhook to receive notifications when a workflow run completes. The webhook payload includes the full extraction results.

Next Steps

Workflow Runs APICreate, monitor, and cancel workflow runs WebhooksGet notified when processing completes SegmentationLearn about document splitting ClassificationLearn about document categorization