Intelligent Document Processing
From Business Documents to Draft ERPNext Entries — You Approve
Intelligent Document Processing (IDP) by Sigzen Technologies reads business documents and turns them into draft entries in ERPNext.
Documents Processed
128
this quarter
Lines Extracted
9,412
statements, bills, receipts
Drafts Proposed
9,180
97.5% of lines
Awaiting Review
7
today
Lines Auto-Matched vs Manual — by Week
WeeklyReview Queue
| Document | Type | Lines | Status |
|---|---|---|---|
| CAP-26-0091 | Bank Statement | 214 | In Review |
| CAP-26-0090 | Supplier Bill | 4 | Drafts Created |
| CAP-26-0089 | Payment Receipt | 1 | Drafts Created |
Four document types work today: bank statements, payment receipts, supplier bills and expense vouchers — scanned, photographed or digital. Every extracted document lands in a review queue; a person checks, corrects and approves before anything is created, and everything is created as a draft that a human submits. IDP also learns: every correction a reviewer makes is remembered, so the next document from the same supplier, bank or customer needs less work than the last one. Purchase Invoice, Purchase Order and GRN document types are in development.
Capabilities
Key Features
Pick a capability to explore what Intelligent Document Processing covers.
Live Today: Four Document Types
Upload the document. IDP works out what it is and prepares the right ERPNext draft.
Bank statements
One draft entry per transaction row — a statement with 60 transactions produces 60 separate draft entries with the right bank and counter legs, not one lumped entry.
Payment receipts
A receipt becomes a draft Payment Entry — receive or pay, worked out from the party on the document.
Supplier bills & expense vouchers
Bills and vouchers become draft Journal Entries with accounts, parties and amounts filled in for review.
Automatic recognition
You don't tell IDP what you uploaded — it works out the document type from the content, and says so plainly when it can't decide rather than guessing.
How It Works
Five steps between a document and a clean draft in ERPNext — with a human in the middle.
1. Upload
Attach the file — PDF or image, up to 50 MB and 50 pages. Processing starts by itself.
2. Read
The document is read page by page, whether it's a clean bank PDF, a scan, or a phone photo.
3. Extract & match
Accounting details are pulled out and matched against your real ERPNext records — accounts, parties, bank accounts.
4. Review
A guided screen asks only about what IDP couldn't work out — answer once per unique name and it applies to every matching row. Then a full preview where every field is editable and uncertain values are highlighted.
5. Approve
Only after approval are ERPNext documents created — always as drafts. A person submits the final entry, every time.
Built for Real-World Documents
Scans, screenshots and phone photos — not just clean exports.
Photos and faint scans handled
Pages are automatically rotated, straightened and cleaned up; a receipt photographed at an angle is flattened; faint thermal receipts are enhanced before reading.
Digital PDFs read perfectly
When a PDF already contains real text — a bank's own downloaded statement — IDP uses that text directly: faster, and exactly accurate.
One bad page never loses the document
If a single page can't be read, the rest still process and the reviewer sees a clear warning — the document only fails if every page fails.
Bank Statements, Done Properly
The hardest, highest-volume document gets dedicated treatment.
Any length, no silent cuts
IDP reads however many transaction rows the statement contains, across as many pages as it has — there is no row cap that quietly truncates a long statement.
Every row checked against the balance
Each transaction is verified against the running balance column. A misread figure — 5,000 read as 50,000 — is caught and flagged; an amount printed in the wrong column is moved where the balance proves it belongs.
Real bank formats understood
Thousands separators, (1,000.00) negatives, Cr and Dr suffixes, ₹ $ € £ symbols, European number grouping — and statements printed oldest-first or newest-first are detected automatically.
Missing dates filled sensibly
When a bank prints the date only on the first row of a day, IDP carries it forward — and marks those rows for the reviewer's attention.
It Learns Your Business
Answer once; IDP remembers — and the next document needs less work than the last.
One answer covers every matching row
A statement with 40 rows from the same supplier asks one question, not 40. The next statement that prints the same name asks none.
Tolerant name matching
"ACME TRADING PVT LTD" and "Acme Trading" are recognised as the same party — capitalisation, punctuation and legal suffixes are normalised away. Even a cryptic narration like TRF 201-54921 is remembered once resolved.
Nothing hidden in a black box
Every remembered mapping is a visible, editable record — and an automatic match still appears in the preview for the reviewer to see and override.
Control & Safety
Automation does the typing. Your team keeps the books.
Drafts only — a human always submits
There is no auto-approve and no confidence shortcut that skips review. Every ERPNext document IDP creates is a draft that a person checks and submits.
Three layers of duplicate protection
The same file can't be uploaded twice (even renamed), the same transaction can't be drafted twice, and skipped rows are recorded with a "duplicate of…" note — a full audit trail, with a per-row override when you know it's genuine.
All-or-nothing creation
If one row of a 60-row statement fails, the whole batch rolls back — no half-created statements to clean up, and retrying is always safe.
Roles and history
Uploader and Reviewer roles are enforced on the server, and every change to a document is tracked in its version history.
Live Today: Four Document Types
Upload the document. IDP works out what it is and prepares the right ERPNext draft.
Bank statements
One draft entry per transaction row — a statement with 60 transactions produces 60 separate draft entries with the right bank and counter legs, not one lumped entry.
Payment receipts
A receipt becomes a draft Payment Entry — receive or pay, worked out from the party on the document.
Supplier bills & expense vouchers
Bills and vouchers become draft Journal Entries with accounts, parties and amounts filled in for review.
Automatic recognition
You don't tell IDP what you uploaded — it works out the document type from the content, and says so plainly when it can't decide rather than guessing.
How It Works
Five steps between a document and a clean draft in ERPNext — with a human in the middle.
1. Upload
Attach the file — PDF or image, up to 50 MB and 50 pages. Processing starts by itself.
2. Read
The document is read page by page, whether it's a clean bank PDF, a scan, or a phone photo.
3. Extract & match
Accounting details are pulled out and matched against your real ERPNext records — accounts, parties, bank accounts.
4. Review
A guided screen asks only about what IDP couldn't work out — answer once per unique name and it applies to every matching row. Then a full preview where every field is editable and uncertain values are highlighted.
5. Approve
Only after approval are ERPNext documents created — always as drafts. A person submits the final entry, every time.
Built for Real-World Documents
Scans, screenshots and phone photos — not just clean exports.
Photos and faint scans handled
Pages are automatically rotated, straightened and cleaned up; a receipt photographed at an angle is flattened; faint thermal receipts are enhanced before reading.
Digital PDFs read perfectly
When a PDF already contains real text — a bank's own downloaded statement — IDP uses that text directly: faster, and exactly accurate.
One bad page never loses the document
If a single page can't be read, the rest still process and the reviewer sees a clear warning — the document only fails if every page fails.
Bank Statements, Done Properly
The hardest, highest-volume document gets dedicated treatment.
Any length, no silent cuts
IDP reads however many transaction rows the statement contains, across as many pages as it has — there is no row cap that quietly truncates a long statement.
Every row checked against the balance
Each transaction is verified against the running balance column. A misread figure — 5,000 read as 50,000 — is caught and flagged; an amount printed in the wrong column is moved where the balance proves it belongs.
Real bank formats understood
Thousands separators, (1,000.00) negatives, Cr and Dr suffixes, ₹ $ € £ symbols, European number grouping — and statements printed oldest-first or newest-first are detected automatically.
Missing dates filled sensibly
When a bank prints the date only on the first row of a day, IDP carries it forward — and marks those rows for the reviewer's attention.
It Learns Your Business
Answer once; IDP remembers — and the next document needs less work than the last.
One answer covers every matching row
A statement with 40 rows from the same supplier asks one question, not 40. The next statement that prints the same name asks none.
Tolerant name matching
"ACME TRADING PVT LTD" and "Acme Trading" are recognised as the same party — capitalisation, punctuation and legal suffixes are normalised away. Even a cryptic narration like TRF 201-54921 is remembered once resolved.
Nothing hidden in a black box
Every remembered mapping is a visible, editable record — and an automatic match still appears in the preview for the reviewer to see and override.
Control & Safety
Automation does the typing. Your team keeps the books.
Drafts only — a human always submits
There is no auto-approve and no confidence shortcut that skips review. Every ERPNext document IDP creates is a draft that a person checks and submits.
Three layers of duplicate protection
The same file can't be uploaded twice (even renamed), the same transaction can't be drafted twice, and skipped rows are recorded with a "duplicate of…" note — a full audit trail, with a per-row override when you know it's genuine.
All-or-nothing creation
If one row of a 60-row statement fails, the whole batch rolls back — no half-created statements to clean up, and retrying is always safe.
Roles and history
Uploader and Reviewer roles are enforced on the server, and every change to a document is tracked in its version history.
How We Compare
Why IDP inside ERPNext beats a generic OCR tool
Native drafts, not exports
Generic OCR tools give you a CSV to import. Sigzen's IDP creates draft entries directly in ERPNext, matched to your real accounts and parties, with the source document attached.
Validated against your books
Extraction is only half the job. IDP checks bank rows against the running balance, matches names to your own masters, and blocks duplicates — the half generic tools skip.
A review workflow, not a dump
Every document passes through a guided review and lands as a draft a person submits — so automation never costs you control of your books.
FAQ
Frequently Asked Questions
Which documents does Sigzen's Intelligent Document Processing handle today?
Four document types are live today: bank statements, payment receipts, supplier bills and expense vouchers — as PDFs or images (JPG, PNG, TIFF, BMP, WEBP), up to 50 MB and 50 pages. Purchase Invoice, Purchase Order and GRN document types are in development.
Does Intelligent Document Processing post entries to ERPNext without review?
No. Every document waits in a review queue, and everything IDP creates is a draft — there is no auto-approve, and no confidence threshold that lets a document skip review. A person always checks and submits the final entry.
Which banks and statement formats does Sigzen's IDP support?
IDP is not built around bank-specific templates — it reads the statement's own content, detects the ordering, and validates every row against the running balance. It handles scanned and digital statements in the supported file formats; Sigzen verifies your bank's statements with a sample during onboarding.
How is my data handled when IDP processes documents?
Uploaded documents are read using a secure, industry-standard AI extraction service, and the results are created only in your ERPNext as draft entries. Your team reviews every extraction before anything is created, access is role-controlled, and every change is tracked in the document's history.
Does IDP get better over time?
Yes. Every correction a reviewer makes is saved as a visible, editable mapping — so the next document from the same supplier, bank or customer resolves automatically. Automatic matches still appear in review, so a remembered answer never bypasses your team.
See your own statement become entries
Book a free demo — bring one bank statement and Sigzen's team will turn it into draft ERPNext entries in front of you.
Get in Touch