Business Card Reader Workflow

Overview

This page describes how to configure a Business Card Scanner workflow using Document AI, including routing, extraction instructions, Data Board insertion, validation, and fallback handling.

Workflow

Purpose

  • Accept a card image from a web widget.

  • Extract contact fields using Document AI.

  • Optionally insert results into a Data Board.

  • Normalize output and respond consistently.

  • Show a retry message when extraction fails.

Routing

  • File uploaded → Document AI → Post-process → AI Assistant

  • No file → Text-only handling → AI Assistant

  • Document AI failure → Message to user

Procedure

1. Web Widget Starter

Configuration

Setting
Recommended

Webhook method

POST

URL mode

Use Test URL for development, Production URL for live

Display the Webhook URLs

ON

Allow Re-trigger After Workflow Ended

ON

Expected input

The widget event should provide:

  • Card image (preferred as a downloadable URL)

  • Optional: text input (only if you support a help/chat mode)

2. Routing (IF)

Objective

Route the workflow based on whether the incoming payload contains a valid file reference.

Recommended rule

Set the condition so it evaluates to TRUE when a card image exists.

Field
Value

Condition type

Boolean

Operation

Equal

Value 2

true

Value 1

Map to a boolean expression like “file exists / fileUrl not empty”

Route behavior

  • TRUE: Run Document AI extraction

  • FALSE: Run text-only handling (Function1 → AI Assistant)

3. Document AI (Business Card Extraction)

Required field mapping

Parameter
Requirement

Input File URL

Must be mapped to the uploaded card image URL

If Input File URL is not mapped, the connector will fail and the workflow will proceed via the Fallback path.

Output language

Set Preferred Language Output based on your primary use case (example: English).

Insert to Data Board (optional)

If you want each scan stored:

  • Enable Insert To Databoard

  • Select a dedicated board (example: OCR Scan Business Cards)

Keep business cards in a separate board from invoices/receipts to avoid schema drift.

Extraction instructions (Business Card prompt)

Use a card-specific schema prompt. Avoid multi-document prompts (invoice/receipt/credit-card) in a Business Card Scanner workflow.

Custom AI Instructions (paste-ready):

4. Document AI advanced settings

Recommended defaults (Business Cards)

Setting
Recommended

Time offset

UTC

Request Timeout

60–180 seconds (increase only if needed)

Max Concurrent Requests

2–3

Retry Failed Requests

ON

Max Retries

2

5. Fallback message (Send a Message)

When it triggers

Use this path when Document AI fails (timeout, unsupported input, inaccessible URL, OCR failure).

Recommended message

Replace generic retry text with actionable guidance:

Message (recommended):

  • “Unable to read the business card. Please retake the photo with good lighting, no glare, and the full card in frame.”

6. Text-only handling (Function1)

Purpose

Only used when no file is uploaded. Converts user text into a prompt for the assistant.

Output contract

Return:

  • prompt (string)

Recommended usage

  • Help mode (“How to scan?”, “What fields are supported?”)

  • Troubleshooting tips when the user doesn’t upload a file

7. Post-processing (Function)

Purpose

Normalize and clean the Document AI output before passing it forward.

Required behavior

  • Remove empty values: "", null, undefined, empty arrays

  • Ensure stable types:

    • emails must remain an array

    • phones must remain an array

  • Keep only business card fields (avoid filtering rules from other workflows)

Recommended output

Pass a cleaned JSON object downstream (either as:

  • data (preferred), or

  • prompt containing stringified JSON if your assistant requires it)

8. Manage AI Assistants (Normalization + schema enforcement)

Purpose

Final pass to enforce output consistency (schema, dedupe, validation flags if required).

Recommended configuration

  • Select the assistant dedicated to OCR normalization (example: OCR & Data Extraction)

  • Enable Customized Prompt

Customized Prompt (paste-ready)

9. Data Board schema (recommended)

If inserting to a Data Board, align columns to the extraction schema:

Core

  • Full Name

  • First Name

  • Last Name

  • Job Title

  • Company

Contact

  • Emails

  • Phones

  • Website

  • LinkedIn

  • Other Social

Additional

  • Address

  • Notes

Operational

  • Scan Timestamp

  • Source

  • Raw Extraction JSON (optional, recommended)

10. Validation rules (recommended)

Mark scan results as “needs review” when:

  • emails[] is empty and phones[] is empty

  • full_name is empty and company is empty

Suggested validation_flags values:

  • missing_email_and_phone

  • missing_name_and_company

  • low_confidence (if you add confidence scoring later)

11. Testing checklist

Test against:

  • Horizontal + vertical cards

  • Low-contrast designs

  • Two-sided cards (if supported)

  • Multiple emails/phones

  • Non-English text (if supported)

Verify:

  • Input File URL mapping works reliably

  • Fallback path returns user guidance

  • Output JSON matches schema every time

  • Data Board insertion creates records correctly

PreviousBusiness Card Reader AI AgentNextCustom LLM

Last updated