How to add document OCR to Baan IV / Baan V
Short answer
Run scanned or PDF documents through an OCR and extraction pipeline outside Baan, validate the result against open POs and master data, route low-confidence fields to a reviewer and import approved data through data exchange or AFS. Store the image on a file share and reference it from the Baan record.
A Baan plant tends to have paper at both ends of the supply chain: supplier invoices and packing slips arriving, and certificates of conformance, mill test reports and inspection sheets that must be filed with lots. Accounts payable keys invoices by hand; receiving keys packing slip details; quality scans certs into a folder named after a person who left.
Baan stores no images natively in the way modern ERPs do, so any capture project has two parts: getting data out of the document, and keeping the document findable from the Baan transaction.
OCR is only useful when paired with matching and exceptions. Reading an invoice total is easy. Knowing that it matches PO 7781 line 3 within tolerance, and flagging it when it does not, is the part that saves money.
What teams use it for
Supplier invoice capture
Extract invoice number, date, totals, tax and lines, match to purchase orders and receipts, and prepare a purchase invoice entry for accounts payable to approve.
Packing slip receiving
Read item, quantity and lot from supplier packing slips to pre-fill warehouse receipts, with the document image attached for the receiving clerk.
Certificate of conformance filing
Identify the PO, lot and heat number from certs and mill test reports and file them in a structure linked to the lot or receipt.
Inspection sheet digitisation
Convert handwritten or printed inspection forms into structured measurements stored in a quality table for trending.
Drawing and revision indexing
Read title blocks of supplier or customer drawings to capture part number and revision and link them to Baan item records.
Baan IV / Baan V integration surface
- Baan purchase order, receipt and invoicing tables, read through a view layer for three-way matching of invoice against PO and receipt.
- Data exchange and EDI-style import paths for purchase invoices, receipts or item data, with the exact mapping dependent on version and local set-up.
- Application Function Server (AFS) for cases where an import is not flexible enough and the session must be driven with full validation.
- Item master and business partner tables for matching supplier names, tax identifiers and item numbers extracted from documents.
- File shares or a document management system referenced by a key stored in a text or custom field, since Baan has limited built-in image storage.
How it works
- 1
Choose the first document type
Invoices from your top twenty suppliers are the usual start because they cover most volume. Gather a sample set of at least a few hundred.
- 2
Capture channel
Take scans from a multifunction printer folder, a mailbox, or a mobile photo. Normalise to PDF or image, deskew and detect page boundaries.
- 3
OCR and extraction
Use layout-aware OCR and a vision or text model to produce header and line fields with per-field confidence. Handwriting and stamps lower confidence and should be flagged.
- 4
Match and validate
Match supplier, PO, receipt and lines using Baan data. Apply rules for price and quantity tolerances, duplicate invoice numbers and tax checks.
- 5
Human review
Show the image beside the extracted fields. Highlight fields under the confidence threshold and every rule failure. Capture corrections as training data.
- 6
Import and archive
Create the Baan entry through the import path, store the image and the extraction record, and write the file key back to the Baan transaction.
Example flows
Invoice three-way match
Trigger: AP scans a supplier invoice into a watched folder.
- Pipeline OCRs the invoice and extracts totals and lines.
- Supplier and PO are matched from Baan.
- Receipt quantities are fetched and compared.
- A price variance on line 2 exceeds tolerance and is flagged.
- AP approves after checking with purchasing, and the import posts the entry.
Result: Keying time is replaced by review, and variances are caught before payment.
Mill cert tied to a lot
Trigger: Receiving scans a mill test report along with the packing slip.
- Pipeline reads PO number, heat number and material grade.
- It finds the receipt and lot in Baan.
- Cert image is filed under the lot key and the reference is stored.
Result: During an audit or a customer complaint the cert is found in seconds, linked to the exact lot.
Keeping data on-prem
Invoices and certs expose supplier pricing, customer names and sometimes export-controlled technical data, so many Baan sites cannot send them to a public OCR API. Run OCR and extraction inside your network: open-source OCR engines plus an open-weight vision-language model served locally on a GPU or even CPU for modest volumes. Baan and its database stay unchanged; the pipeline reads through a read-only account and imports through an approved path. Document images remain on your file server. For air-gapped plants, the same pipeline runs with scanners on the isolated network and no outbound access.
Security checklist
- Restrict the pipeline's database account to read-only and keep the import account separate and audited.
- Treat uploaded files as untrusted, scanning for malware and enforcing type and size limits.
- Block extraction output from triggering any actions; extracted text is data, not instructions.
- Apply segregation of duties: the person who approves an extraction should not be the only one who can change supplier bank details.
- Retain original images, extracted values, edits and the import response together as evidence.
Ways to get it done
Commercial AP automation or IDP product
Effort: Weeks for set-up, with Baan import mapping usually the largest custom piece.
Invoice-only needs with a vendor that already has an ERP-agnostic import option.
Open-source OCR plus your own matching code
Effort: Longer to reach production accuracy, but no licence cost and fully on-prem.
Teams with Python skills and clean Baan master data.
Netray private pipeline
Effort: Phased project, starting with one document type and a review UI, then expanding.
Mixed document types (invoices, certs, drawings) and a requirement to keep everything on-prem.
Pitfalls
- !Supplier names differ between the invoice, the Baan business partner and the bank. Use identifiers such as tax numbers where possible.
- !Line-item extraction from multi-page invoices breaks easily. Test on your worst supplier first, not your best.
- !Currency and decimal separators vary by supplier locale. A misread comma can turn a hundred into a hundred thousand; add range checks.
- !Duplicate detection is essential. Scanning the same invoice twice must not create two payables.
- !Skipping the image link means auditors still need to hunt through folders. Write the document key back to Baan.
Frequently asked questions
Does Baan have built-in OCR?
No. Baan IV and V do not provide document capture. OCR and extraction are done by an external pipeline, and the results are imported into Baan through data exchange, EDI-style files or AFS. The image is stored on a file share or document system and referenced from the Baan record by a key.
How accurate is OCR on invoices?
For clean digital PDFs from known suppliers, header fields are typically read very reliably, with line items needing more care. Poor scans, stamps and handwriting reduce accuracy. A confidence threshold plus review ensures errors are caught, and measuring the correction rate per supplier tells you when to relax checks.
Can we keep documents on-prem?
Yes. OCR engines and vision-language models run locally, and documents never leave your servers. This matters when invoices or certificates include export-controlled or contractually confidential content. The cost is GPU or CPU capacity and some setup, which is usually modest for plant-level document volumes.
Which Baan tables do we match against?
The purchase order, receipt, item and business partner tables, found by package and module in your version. Exact names depend on the release and customisations, so confirm them in the data dictionary with your Baan administrator before building queries, and wrap them in views so the pipeline is insulated from schema detail.
Can this also read handwritten inspection forms?
It can, with lower accuracy than printed text. Constrain the form layout, use boxes for numbers, and send low-confidence readings to a reviewer. For safety or regulated measurements, keep a human verification step and retain the original image as the record of truth.
Want Document OCR working on your Baan IV / Baan V?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment