How to add document OCR to Epicor Kinetic
Short answer
Run scans and PDFs through an OCR and extraction layer, score each field, match results to Kinetic POs, receipts and parts, and post only high-confidence documents through the REST API. Everything else goes to a review screen, and the source file is attached to the Kinetic record.
Paper still enters Kinetic through receiving docks and accounts payable: supplier invoices, packing slips, mill certs, certificates of conformance, inspection reports and drawings. Each one is typed or scanned and filed by someone, often with no link to the transaction it supports.
This page is about document capture itself, not mailbox handling. The question is how to extract trustworthy fields, how to tie a document to the right Kinetic record, and how to prove to an auditor what was captured and who verified it.
Epicor has capture and AP automation offerings whose names and licensing vary by version and edition, and they may be sufficient for standard invoices. A custom or partner pipeline earns its place for unusual documents such as certs, multi-page packing lists and drawings with title block data.
What teams use it for
AP invoice capture with three-way match
Extract supplier, invoice number, date, totals and lines, then match to the PO and receipt before creating the AP invoice.
Packing slip to receipt
Read the supplier's packing slip, match part and quantity to open PO releases and pre-fill the receiving entry.
Certificates of conformance
Capture lot, heat number, spec and signatory from certs and link them to the received lot and inspection record.
Inspection reports
Pull measurement tables from supplier first-article or inspection PDFs into a structured attachment and flag out-of-tolerance values.
Drawing title blocks
Extract part number, revision and material from drawing title blocks to check against the part revision in Kinetic.
Customer PO scans
Digitize faxed or scanned POs where there is no native PDF text.
Epicor Kinetic integration surface
- Kinetic REST API for creating or updating AP invoices, receipts and notes, with business object names and methods that differ between versions; check your instance's API help pages.
- Document attachment capability on entities such as PO, receipt, lot and part, to store the original file with the transaction.
- BAQs to retrieve open PO lines, received-not-invoiced quantities and supplier master data for matching.
- Lot and serial tracking structures used to link cert data to specific lots.
- Epicor's own capture and AP automation products, where licensed, as a baseline to compare against.
How it works
- 1
Define document types
List types and the fields that matter for each. An invoice needs totals and PO reference; a cert needs lot, heat and spec. Do not build a generic extractor first.
- 2
Ingest and normalize
Accept scans, PDFs and images from a scan folder, scanner app or mailbox. Deskew, split multi-document files and prefer embedded text when present.
- 3
OCR and layout analysis
Use a layout-aware OCR engine, then a model or template rules to extract named fields and line tables.
- 4
Score confidence
Combine OCR confidence, format checks (dates, currency), arithmetic checks (lines sum to total) and match success into a document score.
- 5
Match to Kinetic
Resolve supplier, find the PO by reference or by supplier plus part, and compare quantity and price within tolerance.
- 6
Review
Present the image beside extracted fields with low-confidence fields highlighted. Corrections are stored for improvement.
- 7
Post and attach
Call the REST endpoint to create the transaction, attach the original, and record the extractor version and reviewer.
Example flows
Supplier invoice, three-way match
Trigger: A PDF invoice is dropped in the AP scan folder.
- OCR extracts header and lines; totals reconcile.
- The PO is found from the reference and receipts are read by BAQ.
- Quantities match, one unit price differs by a small amount within tolerance.
- The document is auto-routed for single-click approval and the AP invoice is created with the PDF attached.
Result: Clean invoices need seconds of attention, and exceptions are visible rather than hidden in a stack.
Mill cert to lot record
Trigger: A material cert arrives with a shipment and is scanned at receiving.
- The extractor reads heat number, specification, and lot references.
- The system matches the cert to the received lot by PO and part.
- Spec is compared with the part's required spec field.
- A mismatch blocks release and raises an item for quality; a match attaches the cert to the lot.
Result: Traceability is built at receiving, not rebuilt during an audit.
Keeping data on-prem
Documents in regulated environments often carry controlled technical data or customer proprietary information. Run OCR and any language model on-prem or in a sovereign private cloud, using open-source OCR and open-weight vision-language models on local GPUs, with no calls to external APIs. Keep scans on a file share inside your boundary, call Kinetic over the internal network and keep an audit log showing model version, input hash and reviewer. Store originals as Kinetic attachments or in your records system under existing retention policy.
Security checklist
- Encrypt document storage and restrict access by role; scans often include bank details and pricing.
- Never let extracted text drive code paths; validate every field with deterministic checks before a write.
- Keep an immutable record of original file, extraction and corrections for audit.
- Mask or restrict bank account changes on invoices; send them to a verification process rather than automating.
- Apply retention and deletion rules to scans and derived data.
Ways to get it done
Epicor-provided capture or AP automation, or a marketplace AP tool
Effort: Weeks, license and implementation fees vary.
Standard invoices with common formats and a team that wants a supported product.
Build with OCR libraries and your own review UI
Effort: Several weeks per document family plus ongoing tuning.
Teams with developers who can maintain templates and tests for specific suppliers.
Partner such as Netray using local vision-language models
Effort: A few weeks for a pilot on one document type; accuracy measured on your own samples.
Certs, drawings and sensitive documents that must not leave your network.
Pitfalls
- !Judging OCR on clean samples hides real performance. Test on your worst scans: faxes, stamps and handwriting.
- !Matching is where most failures happen, not text recognition; clean supplier and part references first.
- !Totals that reconcile do not prove lines are right. Check line-level tax and freight handling.
- !Revision letters on drawings and certs are easy to misread; treat them as high-risk fields.
- !Duplicate invoices arrive by email and by post; add duplicate detection on supplier plus number plus amount.
Frequently asked questions
Does Epicor Kinetic include OCR?
Kinetic itself is the system of record, and Epicor offers separate capture and AP automation products whose names and availability depend on your edition and version. Ask your Epicor representative what is licensed. For unusual documents you can add your own OCR pipeline over the REST API.
What confidence level is safe for auto-posting?
There is no universal number. Set thresholds per field and per document type using your own labeled samples, require arithmetic and match checks to pass, and start with human review for everything. Relax review only where measured error rates are consistently low.
Can OCR link certs to lots in Kinetic?
Yes, if you extract lot or heat numbers and match them to received lots through the PO and part. Attach the cert to the lot or receipt record and compare specifications with the part requirements to catch mismatches at receiving.
Can this run fully on-prem?
Yes. Open-source OCR and open-weight vision-language models run on local GPUs, and Kinetic is called internally. That is the usual design for defense, aerospace and electronics shops where documents contain controlled data.
What about handwritten delivery notes?
Accuracy drops on handwriting. Use it to propose values, require human confirmation, and consider asking suppliers for digital ASNs where volume justifies it. Do not auto-post quantities taken from handwritten text.
Want Document OCR working on your Epicor Kinetic?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment