# How to add document OCR to Katana > Source: https://netray.co/integrations/katana-document-ocr > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Integrations ## Short answer Katana does not read documents itself, so OCR runs beside it. Extract fields from supplier invoices, packing slips and customer POs, match them to Katana records, and write receipts or orders through the REST API after review. ## System Katana ## Capability document-ocr ## Intro Small manufacturers using Katana still receive paper and PDFs: a supplier packing slip that has to be checked against a purchase order, a vendor invoice that goes to the accounting tool, a customer PO that has to become a sales order, a certificate that must be kept with a batch. Receiving is where the cost is highest. Someone reads a slip, finds the PO in Katana, and enters the received quantities line by line. Mistakes here distort inventory and later production. OCR with sensible matching and review turns those slips into pre-filled receipts that a person confirms. ## Use Cases ### Packing slip to PO receipt **Description:** Read the supplier slip, match lines to the open PO, and prepare the receiving quantities for confirmation. ### Customer PO to sales order draft **Description:** Extract customer, items and dates from a PDF order and draft a Katana sales order. ### Certificates attached to batches **Description:** Read material certificates, capture the batch or lot number, and store the file against the matching batch record. ### Invoice data for three-way check **Description:** Extract invoice totals and compare with the PO and receipt before handing to the accounting tool. ### Delivery note archive **Description:** Index scanned delivery notes by PO number so they can be found later without filenames. ## System Surface - Katana REST API for purchase orders and their rows, including receiving of purchase order rows; check current endpoints for the exact receive operation. - Batch and serial tracking features on the product side, where lot numbers captured by OCR can be stored (plan and settings dependent). - Sales order create endpoints for customer PO capture. - File attachment handling in the API is limited, so plan to keep originals in your own storage and reference them in notes. - QuickBooks and Xero sync for accounting, which means invoice posting usually stays outside Katana. ## How It Works ### Define document types **Detail:** Pick two or three to start, for example packing slips and customer POs, and gather a sample set of real documents. ### Choose the OCR and extraction engine **Detail:** Layout-aware OCR plus a language or vision model for field extraction; open-weight options exist if documents must stay private. ### Match to Katana records **Detail:** Use PO number, supplier and SKU lookups against the API, with fuzzy matching and a cross-reference for supplier part numbers. ### Score confidence per field **Detail:** High scores pass through; anything below a threshold is highlighted on the original image for a person. ### Review screen **Detail:** Show the document side by side with the proposed receipt or order, allow edits, and require a click to confirm. ### Write and archive **Detail:** Call the Katana API to receive or create, store the original file in your document store, and keep an audit trail. ## Example Flows ### Receiving goods against a PO **Trigger:** A photo of a packing slip is uploaded by the receiver. **Steps:** - OCR extracts the PO number and line quantities. - The open PO is fetched from Katana and lines are matched. - A variance, 95 received against 100 ordered, is highlighted. - The receiver confirms; the receipt is posted. **Result:** Inventory updates the same day with the discrepancy visible. ### Material certificate capture **Trigger:** A certificate PDF arrives with a shipment. **Steps:** - The heat or lot number and material are extracted. - The matching received batch is located. - The file is stored and linked in the batch notes. **Result:** Traceability documents are findable by lot. ## On Prem Because Katana is cloud only, only the structured outcome goes to it. Run OCR and extraction on your own hardware: document layout and text recognition models run on a single modest GPU, and an open-weight vision language model can handle messy slips locally. Keep the original scans and certificates in your file share or private object storage, and send Katana only the PO number, quantities and batch ids. This matters for customers whose drawings and certificates are under contractual confidentiality. ## Security - Store originals in your own access-controlled storage. - Do not put full document text into Katana notes. - Log who confirmed each extracted record. - Scan uploads for malware before processing. ## Options ### Off-the-shelf OCR service with Zapier or Make **Fit:** Consistent invoice formats and low volume. **Effort:** Days. ### In-house pipeline with open OCR and a vision model **Fit:** Mixed supplier formats and in-house developers. **Effort:** Four to eight weeks to tune. ### Partner build with Netray **Fit:** Private deployment, review UI, and supplier-specific matching. **Effort:** Several weeks, based on document types. ## Pitfalls - Packing slips often omit your PO number; fall back to supplier and date matching. - Units of measure differ between slip and Katana, such as each versus box. - Table extraction fails on skewed phone photos; add image correction first. - Partial deliveries must receive only what arrived, not the whole PO. ## Faq ### Does Katana include OCR? **Answer:** Not to our knowledge. Katana manages inventory, orders and production but does not read documents. You add OCR as a separate step and push the verified result into Katana through the REST API. ### Can documents be attached to Katana records? **Answer:** Attachment support in the API is limited and has changed over time, so check current documentation. A safer pattern is to keep originals in your own storage and add a link or reference number in the record notes. ### What accuracy should we expect? **Answer:** Digital PDFs from repeat suppliers can be very accurate. Phone photos, handwriting and unusual layouts are lower. Test on a few hundred of your own documents and use confidence thresholds so uncertain fields always get checked. ### Can it read supplier invoices for accounting? **Answer:** Yes technically, but Katana usually syncs to QuickBooks or Xero for accounting, so many teams send invoice data there. Use OCR in Katana for receiving and ordering, and match invoices to receipts as a check. ### Is private OCR realistic for a small shop? **Answer:** Yes. A single workstation GPU can run OCR and a modest open-weight model for a small document volume. If volume is tiny, a managed service with a data processing agreement may be cheaper and simpler. ## Search terms katana document ocr, katana ocr packing slip, katana receive purchase order automatically, ocr for katana mrp, katana invoice scanning, katana batch certificate capture, katana api receive po, katana customer po extraction --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json