# How to add document OCR to Fishbowl > Source: https://netray.co/integrations/fishbowl-document-ocr > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Integrations ## Short answer Capture scans and PDFs into a watched folder or mailbox, run OCR and field extraction, validate against Fishbowl parts, vendors and open POs, then create receipts, attach documents or draft orders through the Fishbowl API. Low-confidence fields go to a review screen before anything is saved. ## System Fishbowl ## Capability document-ocr ## Intro Fishbowl users handle paper at receiving docks and in the back office: vendor packing slips that must match a PO, supplier invoices, mill or conformance certificates, customer POs and signed delivery proofs. Fishbowl stores attachments against records, but it does not read them. OCR changes that only if the output is checked against what Fishbowl already knows. A packing slip is useful when its lines can be tied to open PO lines; an invoice is useful when it can be tied to a received PO. Raw text on its own just moves typing from one place to another. The right design therefore starts from the Fishbowl transaction you want to produce, works backward to the fields needed, and treats OCR confidence and matching as first-class checks. ## Use Cases ### Receiving from packing slips **Description:** Scan a vendor packing slip, match lines to the open PO, and prefill the receipt quantities for the receiver to confirm. ### Certificates attached to lots **Description:** Read lot or heat numbers from certificates and attach the file to the matching tracked part or receipt for traceability. ### Vendor invoice capture **Description:** Extract invoice number, date, totals and PO reference, compare to receipts, and pass a clean record to accounting. ### Customer PO scans **Description:** Turn faxed or scanned customer POs into draft sales orders, mapping customer part numbers. ### Proof of delivery filing **Description:** Match signed delivery documents to the shipment record and attach them automatically. ## System Surface - Fishbowl REST API on current versions for reading POs and parts and posting receipts and orders, with approved integration credentials; verify the supported endpoints for your build. - Legacy JSON API over TCP on the Fishbowl server for older environments. - Attachments on records, available in the client for documents and images, which can be a place to link the source scan, depending on version and API support. - MySQL for read-only matching: po, poitem, part, vendor and vendor part number tables, plus tracking data for lots and serial numbers. - CSV imports for POs, sales orders and inventory, usable when reviewed output is staged as files. ## How It Works ### Pick the capture path **Detail:** A scanner or MFP dropping PDFs in a network folder works at a dock. Phone photos work for field delivery. A shared mailbox covers vendor-sent files. ### Normalise the input **Detail:** Deskew, de-noise and split multi-page batches. Detect document type early because packing slips, invoices and certificates need different extraction. ### Run OCR and extraction **Detail:** Use an OCR engine plus a layout-aware model or template rules to pull header fields and line tables, each with a confidence value. ### Validate against Fishbowl **Detail:** Look up the PO, check the vendor, resolve part numbers using vendor part mappings, and compare quantities with open quantities. ### Review exceptions **Detail:** Show the page image beside the extracted fields. Highlight low-confidence and mismatched values and let the reviewer correct once, with corrections feeding back into mappings. ### Post and file **Detail:** Create the receipt or draft through the API after approval, store the document with the record or an archive link, and log the extraction for audit. ## Example Flows ### Dock receiving with a scan **Trigger:** A receiver scans a vendor packing slip at the dock. **Steps:** - The PDF lands in the watched folder and is classified as a packing slip. - The PO number and lines are extracted and matched to the open PO. - The receiving screen opens with quantities prefilled and mismatches highlighted. - The receiver confirms and the receipt is posted. **Result:** Receiving is faster and short or over shipments are noticed at the dock, not at invoice time. ### Heat number certificate **Trigger:** A supplier emails a material certificate for a received lot. **Steps:** - OCR reads heat number, spec and quantity. - The service finds the receipt and tracked lot. - The file is attached to the record with extracted values noted. **Result:** Quality can pull the certificate for a lot in seconds during an audit. ## On Prem Document images often contain pricing, customer names and sometimes controlled technical data, so many Fishbowl sites prefer to keep the whole pipeline local. Run an open-source OCR engine and a locally served vision or language model on a server in the same network, store originals on your file share, and call Fishbowl on the LAN. Nothing then leaves the building, and you can still tune extraction on your own document samples. A GPU is helpful for larger models, but modest volumes can run on CPU with simpler OCR plus rules. ## Security - Keep source documents on controlled storage with access limited to the processing service and reviewers. - Do not send scans to a public OCR service if they hold controlled technical data or confidential pricing. - Restrict the Fishbowl integration user to receiving and read rights needed for matching. - Retain original scans and the extraction output together for audit. - Require review for changes to bank details, tax identifiers or vendor master data. ## Options ### Cloud OCR or invoice capture service **Fit:** Mostly typed invoices, non-sensitive data and a team willing to use a hosted service. **Effort:** Days to connect, but matching to Fishbowl still has to be built. ### In-house open-source OCR with rules **Fit:** A handful of stable vendor layouts and in-house developers. **Effort:** Three to six weeks, with maintenance per new layout. ### Partner build with Netray **Fit:** Varied documents, on-prem inference and review UI tied to Fishbowl receiving. **Effort:** Pilot on one document type, then extend. ## Pitfalls - Trusting OCR numbers without cross-checking with PO quantities, where a misread 8 versus 3 posts a wrong receipt. - Ignoring unit-of-measure differences between vendor documents and Fishbowl. - Training on clean PDFs and then facing skewed photos from the dock. - Having no feedback loop, so the same vendor part number is corrected by hand every week. - Auto-posting receipts for tracked parts without capturing lot or serial data. ## Faq ### Does Fishbowl include OCR? **Answer:** I am not aware of built-in OCR for incoming documents. Fishbowl lets you attach files to records, and it has barcode-driven receiving, but reading the contents of a scan needs an external tool. Check your version and add-on directory for anything new. ### How accurate will OCR be? **Answer:** It depends heavily on document quality and layout. Clean typed PDFs can be very reliable; skewed photos and old fax scans are not. Plan on a review step and measure accuracy per vendor, rather than relying on a single headline figure. ### Can the system post receipts automatically? **Answer:** It can for high-confidence matches where every line agrees with an open PO line, but start with review for everything. As you see vendor-by-vendor accuracy, you can allow straight-through posting for the most consistent ones with exception alerts. ### Where should scanned files be stored? **Answer:** Store them on controlled storage and link them from Fishbowl, or attach them to the record if your version supports it and volumes are modest. Large archives of attachments can bloat the system, so many teams keep a file share and store a reference. ### Can this run without internet? **Answer:** Yes. Open-source OCR and locally served models work offline, which suits air-gapped or restricted sites. You trade some convenience and potentially some accuracy for control, and you manage updates yourself. ## Search terms fishbowl document ocr, fishbowl packing slip ocr, fishbowl invoice capture, ocr for fishbowl inventory, fishbowl receiving automation scan, fishbowl certificate of conformance ocr, automate fishbowl purchase order receiving, fishbowl ai document processing --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json