How to add document OCR to Microsoft Dynamics GP
Short answer
Run scans and PDFs through OCR plus a field extraction model, score each field's confidence, send low-confidence items to a reviewer, and post approved results into GP through eConnect. Keep the original image attached to the GP record using Document Attachment.
GP customers who handle paper, scanned packing slips from the dock, supplier certificates, faxed POs, still depend on people for data entry. The core GP product does not read documents. Its attachment feature stores files, but nothing is extracted from them.
OCR has changed in the last few years. Older template-based tools needed one template per vendor. Layout-aware models can read a vendor's invoice they have never seen, but they make confident-looking mistakes, so the design must assume errors and catch them.
This page focuses on the capture pipeline: scan quality, field confidence, matching to GP data, and the specific controls that make the output safe to post.
What teams use it for
Packing slip to receipt
A photographed or scanned packing slip is matched to an open PO, quantities are extracted per line, and a draft receipt is staged for the receiving clerk.
Invoice capture with three-way match
Invoice lines are compared with PO and receipt quantities. Matches within tolerance go to a quick-approve queue; mismatches go to a buyer.
Certificates of conformance filing
Mill certs and CoCs are read for heat or lot number, item and supplier, then attached to the receipt or lot record and indexed for retrieval.
Freight bill matching
Carrier bills are read for PRO number and charges and matched to shipments, flagging overcharges before they reach payables.
Legacy paper archive indexing
Historical vendor files are scanned and indexed by vendor, document type and date so AP can find them from the GP record.
Microsoft Dynamics GP integration surface
- eConnect inserts the resulting payables, receivings or inventory documents with GP's own validation. This is the supported write path for extracted data.
- Document Attachment Management links files to GP cards and transactions. Files are stored in the GP database by default, so heavy scanning can grow it quickly; plan for external storage or retention rules.
- Reference data for matching comes from SQL tables such as PM00200 (vendors), POP10100 and POP10110 (PO header and lines), POP10300 and POP10310 (receipt header and lines), and IV00101 (items). Verify names on your version.
- Optional lot and serial tracking tables apply to received items; if you capture lot numbers from certs, the lot fields must exist for the item class first.
- Smartlist or SQL views can power a review dashboard showing which PO lines remain open for a scanned slip.
How it works
- 1
Standardize intake
Define one path for each document type: scan-to-folder, mailbox or a mobile capture page. Specify minimum 300 dpi for scans and reject blurry photos at upload.
- 2
Preprocess images
Deskew, denoise and split multi-page scans into documents before OCR. Poor preprocessing causes more errors than the model.
- 3
Run OCR and extraction
Use an OCR engine plus a layout-aware extraction model that outputs a fixed JSON schema, including per-field confidence.
- 4
Match to GP records
Resolve vendor, PO and items against GP tables. Use vendor item cross references and exact PO number first, fuzzy matching second.
- 5
Apply thresholds
Auto-accept only fields above a tuned threshold and passing rules such as line totals equalling header total. Everything else goes to a reviewer.
- 6
Review and post
The reviewer sees the image beside extracted values. On approval, eConnect posts the document and the image is attached to it.
Example flows
Dock scans a packing slip
Trigger: Receiving scans a packing slip into a watched folder
- OCR reads the PO number and line quantities.
- The service finds the open PO and compares remaining quantities.
- One line exceeds the open quantity, so it is flagged as over-receipt.
- Clerk confirms the others and holds the over-receipt for the buyer.
- eConnect posts the receipt and the image is attached.
Result: Receiving is posted the same day with a visible audit trail and over-receipts are caught immediately.
Invoice with a total that does not add up
Trigger: Scanned vendor invoice arrives for OCR
- The extraction returns lines and a header total.
- A rule adds the lines and tax and compares with the total.
- The difference exceeds tolerance, so the confidence is overridden to low.
- Reviewer inspects the image, corrects one misread digit, and approves.
Result: A misread digit is caught by arithmetic before it becomes a wrong payable.
Keeping data on-prem
Documents like invoices, drawings and certificates often carry contract terms, part numbers and controlled technical data that should not go to a public API. Run OCR and the extraction model locally: modern open-source OCR and open-weight vision-language models can run on a single GPU server or even CPU for modest volumes. The review interface and the staging database stay on the same network as GP, and the only GP interaction is local eConnect and SQL. Plan storage: scans as files on a file share with GP holding references is often better than stuffing everything into the GP attachment tables.
Security checklist
- Treat scanned documents as sensitive; encrypt storage and restrict the review UI by role.
- Segregate duties: the person who approves an extraction should not also release the payment.
- Log the original file hash, extracted JSON, reviewer, thresholds and resulting GP document number.
- Scan uploads for malware and sandbox PDF processing.
- Flag changes to vendor remit-to details found in invoices for separate verification.
Ways to get it done
Commercial AP or capture product that supports GP
Effort: Weeks plus subscription
Best for high-volume invoices on standard layouts. Review GP version support, pricing per document and where the data is processed.
Build with open OCR, a local model and eConnect
Effort: Two to four months for the first document type
Suits controlled data and varied document types; needs development and tuning time.
Netray with a GP partner
Effort: Pilot in weeks, depends on document variety
For on-prem capture across invoices, slips and certs with confidence-based review and GP-aware matching.
Pitfalls
- !Judging accuracy on clean samples; test on your worst scans, faxes and phone photos.
- !Auto-accepting header totals without line arithmetic checks.
- !Letting attachments bloat the GP database; set size limits and move storage out.
- !Skipping duplicate detection for resubmitted invoices.
- !Extracting lot numbers for items that are not lot tracked in GP, which then fail to post.
Frequently asked questions
Does Dynamics GP include OCR?
Not in the core product. GP can attach and store documents, but it does not read their contents. OCR and extraction come from a third-party product or a custom pipeline that then posts through eConnect.
What accuracy should we expect?
It varies by document quality, language and layout variety. Clean digital invoices can be extracted reliably; poor scans and handwritten notes need review. Measure on your own samples and set thresholds from observed error rates.
Where should the scanned files live?
Either in GP's attachment storage or on a file share referenced from GP. For large volumes, an external store with backups and retention is easier to manage and keeps the company database smaller.
Can OCR run fully offline?
Yes. Open-source OCR engines and open-weight vision models can run on local hardware. Quality depends on the model and GPU, so test with your documents before committing.
Is three-way match possible with GP data?
Yes. PO lines, receipts and payables are all in GP, so a service can compare invoice lines with received quantities and PO prices. Tolerances and exception routing are policy decisions you configure.
Want Document OCR working on your Microsoft Dynamics GP?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment