# How to add document OCR to Global Shop Solutions > Source: https://netray.co/integrations/global-shop-document-ocr > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Integrations ## Short answer Add OCR as a staging step in front of Global Shop: scan or receive documents, extract fields with confidence scores, validate against POs, parts and vendors, and have a person confirm low-confidence items before posting through a supported import path. Store the image against the ERP record. ## System Global Shop Solutions ## Capability document-ocr ## Intro Job shops and contract manufacturers handle a lot of paper: vendor invoices, packing slips from receiving, mill certs, certificates of conformance, first article reports, customer drawings. Much of it is scanned, photographed or emailed, then keyed or filed by hand. OCR plus a language model can read this paper well enough to do the first pass, but accuracy on a packing slip photographed on a loading dock is not the same as on a clean PDF. The design should assume errors and make them cheap to catch. Where an ERP edge exists in Global Shop is the matching: a PO number, a part number and a quantity can be checked against open orders before anything is posted. That check turns OCR from a guess into a verified entry. ## Use Cases ### Vendor invoice capture **Description:** Extract vendor, invoice number, date, totals and lines, then three-way match against the PO and receipt before sending to accounts payable. ### Packing slip receiving **Description:** Photograph a slip at the dock, match lines to open PO lines, and propose receipt quantities for the receiver to confirm. ### Certs of conformance and mill certs **Description:** Read heat or lot numbers, spec and material grade, and attach the cert to the receipt and traceability record. ### Customer PO scans **Description:** Capture POs received as paper or image and stage them for order entry. ### First article and inspection reports **Description:** Index scanned reports by job and part, and extract pass or fail and measured values where structured. ## System Surface - ODBC read access for matching to open POs, receipts, vendors, parts and jobs. - Supported import routes or interfaces for creating receipts, payables or attachments. Confirm availability and licence for your release. - Global Shop document handling and attachment features for storing the source image against a record; verify the method in your version. - Lot and serial traceability fields on receipts, which certificates need to link to. - Scanner and MFP output as watched-folder PDFs or TIFFs, which is the simplest capture feed on-prem. ## How It Works ### Set up capture channels **Detail:** Watched folder from scanners, a mailbox for vendor invoices, and mobile photo upload for the dock. Normalise to PDF with deskew and contrast. ### Classify the document **Detail:** Invoice, packing slip, cert, drawing, other. Different types go to different extractors and validators. ### Extract with confidence **Detail:** Use OCR for text and a layout-aware model for fields and tables. Every field carries a score and a bounding box for the review screen. ### Match to ERP data **Detail:** Look up the PO by number, the vendor by name or tax ID, and each line by part and quantity. Tolerances define what auto-passes. ### Review exceptions only **Detail:** Show the image next to the fields, with low-confidence and mismatched fields highlighted. Clean documents pass with sampling. ### Post and archive **Detail:** Post via the supported path, link the image to the ERP record, and keep the extraction log for audit. ## Example Flows ### Receiving a steel shipment with a mill cert **Trigger:** A receiver photographs the packing slip and the cert. **Steps:** - Classifier identifies both documents. - Packing slip lines are matched to the open PO. - Heat number and grade are extracted from the cert and compared to the PO spec. - Receiver confirms quantities, and the receipt and cert link are created. **Result:** Receipt and traceability are captured together, with the cert attached to the lot. ### Invoice with a price variance **Trigger:** A vendor invoice arrives by email. **Steps:** - Fields are extracted and the PO is matched. - Unit price differs from the PO beyond tolerance. - The item is routed to a buyer, not to payables. **Result:** The variance is resolved before payment instead of after. ## On Prem OCR and layout models run well on a single GPU server in the plant, and document images never need to leave your network. Use an open-weight vision-language model or a classic OCR engine with a smaller model for field extraction. The watched-folder capture, the ERP validation and the review UI can all live on the LAN. For controlled drawings or CUI documents, keep the whole pipeline inside the boundary you have documented for that data, and do not use a cloud OCR API for those files. Consider keeping an air-gapped review workstation if the site requires it. ## Security - Encrypt the staging store and purge images after posting under your retention rules. - Separate roles: who can review, who can post, who can change tolerances. - Treat PDFs as untrusted files; scan and sandbox before parsing. - Never let extracted text act as instructions to the model or the ERP. - Keep the raw image and the extracted values together as an audit pair. ## Options ### Off-the-shelf invoice capture tool **Fit:** Accounts payable invoices where a vendor connector or CSV import path to Global Shop exists. **Effort:** Days to weeks, subscription cost, cloud processing in most cases. ### In-house pipeline with open-source OCR **Fit:** Teams that can maintain a model service and a review UI. **Effort:** Several weeks per document type. ### Partner build with Netray **Fit:** Multiple document types, ERP matching and local models. **Effort:** Pilot on one document type, then expand. ## Pitfalls - Judging accuracy on clean samples instead of real dock photos and faxed scans. - Reading quantities and units wrongly, such as EA versus box, and posting a bad receipt. - No tolerance rules, so every small variance becomes a manual task or slips through. - Losing the link between the image and the ERP record. - Ignoring handwriting and stamps on certificates. ## Faq ### Does Global Shop Solutions include OCR? **Answer:** We are not aware of built-in OCR for incoming invoices and packing slips. Check your release and any add-on modules. Most shops add an external capture step that stages data and then posts through an import. ### What accuracy can we expect? **Answer:** Clean machine-generated PDFs are usually high, while photographed or scanned paper is lower and varies by vendor. Use confidence scores and ERP matching to catch errors, and start with review on every document. ### Can OCR run without the cloud? **Answer:** Yes. A local GPU server can run OCR and an open-weight extraction model. This is the right choice when documents contain controlled technical data or when policy forbids third-party processing. ### How are certs tied to traceability? **Answer:** Extract heat or lot numbers, then link the cert to the receipt line carrying that lot. The exact field depends on how your Global Shop traceability is configured, so confirm with your admin. ### Can we auto-post without review? **Answer:** Only for document types and vendors where matching passes reliably over a test period. Keep sampling for audit even then. New vendors and high-value documents should remain under review. ## Search terms global shop solutions ocr, global shop erp invoice capture, packing slip ocr receiving erp, mill cert ocr traceability, global shop solutions document scanning, ai invoice processing manufacturing erp, on-prem ocr erp documents, three way match ocr job shop --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json