Document Extraction
Convert uploaded documents into structured data.
Compliance SaaS · Onboarding Workflows
Smart Uploader
Reducing repetitive document verification in onboarding workflows
A configurable validation module that extracts uploaded documents into structured data, compares them against submitted application values, and helps reviewers focus on cases that require review.
Field Validation
Compare extracted values against submitted application data.
Configurable Rules
Support different documents and onboarding requirements.
Review Workflows
Surface mismatches and exceptions for manual review.
The Problem
Reviewers manually compare information between documents and submitted forms.
Many compliance onboarding workflows require applicants to submit both structured form data and supporting documents.
Most systems treat uploaded documents as attachments. The information inside is never extracted or compared against what the applicant submitted. That gap is closed manually, by reviewers, one document at a time.
As document types and applicant volumes grow, the review burden scales with them.
01
Open document
→
02
Locate fields
→
03
Read values
→
04
Switch to form
→
05
Find matching fields
→
06
Compare manually
→
↻ Repeat for every document
Every document requires a full read, regardless of whether it matches.
Key Participants
One workflow, three perspectives.
End-User / Applicant
Customer or applicant submitting documents across devices, typically on mobile.
Key Goals
- Upload required documents and proceed to the main service.
Pain Points
- No guidance on what makes a photo acceptable.
- Upload failures give no reason, leaving users to guess and retry.
UI/UX Solutions
- Preventive Uploader UI: file type filtering, image clarity check, and real-time guidance before submission to capture clean data at the source.
Operations Staff
Back-office reviewer, 7-8 hours daily verifying documents and overriding flagged values.
Key Goals
- Complete document verification within SLA.
- Reduce errors to zero.
Pain Points
- Eye fatigue from switching between document and form screens.
- Frustration from approving documents that are already correct.
UI/UX Solutions
- Review by Exception: surface only cases that fail automated checks.
- Side-by-Side Validation: document image alongside form fields to minimize eye movement.
Operations Manager
System administrator overseeing team performance and verification rule configuration.
Key Goals
- Maintain data quality entering the system.
- Adjust verification conditions immediately as business policy changes.
Pain Points
- Business disruption every time a rule change requires a developer to modify code.
UI/UX Solutions
- No-code Rules Configuration: a dashboard for configuring matching rules without writing code.
Account No.
Account Name
Account No.
Account Name
Account Name
Account No.
The same account name and account number appear in different positions depending on the document layout.
Selected Scenario
Verifying payment account documents across different banks.
For the payment setup step, applicants upload a bank book as proof of account ownership. The reviewer must confirm that the account name, account number, and bank name on the document match what the applicant submitted in the form.
The Challenge
Each bank arranges this information differently. There is no consistent position to look at, so every document requires a full read.
The Solution
From uploaded documents to validation outcomes.
The workflow operates consistently across document types and onboarding scenarios, reducing the amount of verification work that falls to reviewers.
Upload or Capture
Applicant submits a document or takes a photo directly through the onboarding form.
→
Extraction
Uploaded documents are converted into structured field values.
→
Validation
Extracted values are compared against submitted application data.
→
Review by Exception
Only mismatches, missing values, and exceptions reach a reviewer.
01 · Upload or Capture
Start with a clean document.
The process begins when an applicant submits a supporting document through the onboarding form, either by uploading a file from their device or taking a photo on mobile.
Before anything is extracted, the system validates the file type and checks that the image meets minimum quality requirements. Applicants receive immediate feedback if the submission needs to be corrected, keeping unreadable or wrong-type files out of the review queue from the start.
What gets accepted here flows directly into extraction accuracy and reviewer workload. A clean upload reduces downstream mismatches; a poor one creates exceptions that fall to a human.
02 · Extraction
Extract information from uploaded documents.
Before extraction begins, the system checks that the uploaded file matches the expected document type. Files that do not pass this check are flagged before any extraction is attempted.
Different formats may arrange the same information in different layouts, and the extraction layer normalises this so validation can compare across document types.
Extracted values can also be passed downstream to prefill forms, trigger decisions, or feed other systems.
The uploaded bank book is processed through OCR, converting account name, account type, bank name, and account number into structured field values.
03 · Validation
Compare information across sources.
Extracted values are compared against what the applicant submitted in the application form. The system evaluates whether values match, differ, or are absent, and records a validation outcome for each field under review.
Extraction alone does not reduce manual work. Validation is the step that turns extracted information into an operational result.
Each extracted field is mapped against its corresponding value in the submitted application form. The system evaluates them individually and records a match or mismatch outcome for each.
04 · Review by Exception
Review what matters.
Reviewers do not need to inspect every uploaded document. The system identifies cases where values do not match, information is missing, extraction confidence is below threshold, or a situation falls outside automated criteria.
When a reviewer examines a flagged case, they can record a corrected value in the OPS Value column. Once saved, the field is marked as reviewed regardless of what the OCR or submitted data contained.
Payment Setup Interactive Demo
รายละเอียดบัญชีธนาคาร
อัปโหลดหน้าสมุดบัญชีธนาคาร
Field
OCR Value
OCR Match
User Value
OPS Value
Action
ชื่อบัญชี
นาย สมชาย ใจดี
OCR Mismatched
สมชาย ใจดี
ทดลองกด save
เพื่อดูผลลัพธ์
เพื่อดูผลลัพธ์
ประเภทบัญชี
บัญชีเงินฝากออมทรัพย์
OCR Matched
บัญชีเงินฝากออมทรัพย์
ธนาคาร
ธนาคารไทยพาณิชย์
OCR Matched
ธนาคารไทยพาณิชย์
เลขบัญชี
78932468109
OCR Mismatched
78932468105
78932468105
OCR Match คือผล auto-compare ระหว่างค่า OCR กับข้อมูลใบสมัคร ไม่เปลี่ยนแปลงแม้จะมีการแก้ไข OPS Value
หาก OPS Value มีค่า ระบบถือว่าผ่านการตรวจสอบโดยเจ้าหน้าที่แล้ว ไม่ว่า OCR จะตรงหรือไม่
บางฟิลด์ใช้ dropdown — เลือก "อื่นๆ" เพื่อพิมพ์ค่าเอง
Outcome
From document review to exception review.
The onboarding workflow does not change. Documents are still uploaded. But instead of requiring a reviewer to open each one and confirm what it contains, the system processes the document, compares the result against submitted data, and surfaces only the cases that need a human decision.
Traditional Onboarding
- Upload document
- Store as attachment
- Reviewer opens file
- Compare manually
- Approve
→
↓
Attachments
Validation Sources ↓
Validation Sources ↓
Smart Uploader
- Upload document
- Extract information
- Validate against application
- Flag exceptions
- Review only when needed
Configuration
The workflow is configured, not hardcoded.
Each upload step is configured in two parts. First, teams define which keywords the document must contain to confirm it is the correct type. Then, they specify which fields to extract and which form answers to validate them against. The same engine supports any document type and onboarding scenario without custom code.
Document Validation Config Interactive Demo
Required
User will not be able to proceed without answering this question.
Question Text
Description Text
Only allow users to take a live photo
Only accept JPG, PNG, PDF file formats.
Maximum Pages
Up to 20 pages per file uploader.
Maximum File Size (MB)
You can set the maximum file size to be uploaded between 0.1 - 12 MB.
Verify Document Types
The uploaded documents need to contain
of these keywords
Separate with commas or the Enter key
Extract and Verify Document Content
Validate the extracted values in the documents against previously filled values in the form.
Fields List
Field NameExtraction RuleTypeValidate against