How to add document OCR to ECI E2 Shop System
Short answer
Scan or receive documents into a watched folder, run OCR and field extraction, match against open POs and receipts in E2, and send low-confidence items to a reviewer. Store the original files against the E2 record and keep processing on your own servers.
In an E2 shop the paperwork that arrives with material is the real bottleneck: packing slips at receiving, vendor invoices at payables, mill certs and certificates of conformance for traceability, and outside-processing paperwork from platers and heat treaters. Most of it is still scanned, filed in a cabinet, or keyed by hand.
OCR on its own is not the solution, because turning pixels into text is the easy part. The value comes from matching the text to E2 data: which PO line, which received quantity, which heat or lot number belongs to which part, and whether the certificate satisfies the customer's requirement.
This page covers the pipeline from scanner to E2, with attention to confidence handling, traceability data and the difference between invoices, packing slips and quality certificates.
What teams use it for
Packing slip receiving
Read the vendor packing slip, match it to the open PO, propose received quantities by line, and let the receiver confirm before posting the receipt in E2.
Vendor invoice capture
Extract invoice number, date, totals and lines, match to PO and receipts, and stage for payables with variances highlighted.
Mill and material certs
Pull heat number, material specification and chemistry or mechanical values from mill test reports and attach them to the received lot.
Certificates of conformance
Capture CoCs from outside processors and check that the process, specification and part number match the OSP purchase order.
Customer drawings and POs
Read title blocks for part number, revision and material, and compare them with the E2 part master to catch revision mismatches.
Inspection and traveler scans
Digitise handwritten inspection sheets or travelers, extracting job number and measurements for quality review, with clear handwriting limits.
ECI E2 Shop System integration surface
- SQL Server data for POs, PO lines, receipts, vendors and parts supplies the matching targets. Table names differ by release, so map them from your own schema.
- Receiving and invoice transactions in E2 are posted through the application, so you should create them by a supported import or let a user post them from a prepared screen.
- Document linking: E2 can link external files to several record types. The method varies by version, so decide whether to store files in the E2 attachment area or on a file share with a URL in a note.
- Lot and heat tracking: if your E2 configuration tracks lot or heat numbers on receipts, OCR output can populate that data, but confirm which fields are available.
- Folders and scanners: most multifunction scanners can scan to folder or email, which gives you a simple ingestion point without custom hardware.
How it works
- 1
Set up ingestion
Configure scanners and a mailbox to drop files in watched folders by type, such as receiving, payables and quality. Name files with a timestamp and source.
- 2
Pre-process the images
Deskew, denoise and split multi-page batches. Detect page type so a mill cert and a packing slip take different paths.
- 3
Run OCR and layout extraction
Use an open OCR engine or a document-understanding model hosted locally. Output text with positions and per-field confidence.
- 4
Extract fields by document type
For packing slips: PO number, part numbers, quantities. For invoices: totals and tax. For certs: heat, spec, lot and sign-off.
- 5
Match to E2
Find the PO by number and vendor, match lines by part or vendor part number, and check quantity tolerances against open balance.
- 6
Review and post
Route fields below the confidence threshold to a reviewer with the image beside the data. Approved data is posted via a supported path.
- 7
Archive with the record
Store the original and extracted JSON, and link both to the receipt, invoice or lot for audit and customer requests.
Example flows
Receiving dock packing slip
Trigger: The receiver scans a packing slip at the dock.
- OCR finds the PO number and three line items.
- Lines match the open PO and show remaining balance.
- One quantity reads low confidence and is highlighted.
- The receiver corrects it and confirms, and the receipt is prepared for posting.
Result: Receiving takes a minute instead of manual lookup, and the slip is stored with the receipt.
Mill cert traceability
Trigger: A mill test report PDF arrives with a bar stock delivery.
- The document is classified as a material certification.
- Heat number, spec and grade are extracted.
- The heat is compared with the PO requirement, such as the specified alloy.
- A mismatch raises a hold flag for quality review before material is issued.
Result: Non-conforming material is caught at receiving rather than at final inspection or at the customer.
Keeping data on-prem
Document capture is one of the easiest AI workloads to run fully on-prem. Scanners write to a file share, an OCR service and a small vision-language model run on a local GPU or even a modest CPU box for lower volumes, and the extraction results go to a database on your network. Open-weight document models served with vLLM or Ollama can read tables and stamps, and classic OCR engines handle clean scans. Matching reads SQL Server through a read-only login. The review application is an internal web page protected by your directory login. Nothing leaves the building, which matters for ITAR or export-controlled drawings and for customer NDAs. Keep a retention policy for scanned originals to match your quality system and customer flowdowns.
Security checklist
- Segregate document folders by sensitivity so export-controlled drawings are only reachable by cleared staff.
- Use a read-only SQL login for matching; post transactions only through a controlled path.
- Keep extracted values and originals together with immutable timestamps for audit and traceability.
- Mark OCR output as untrusted input, validate numeric fields and reject impossible values.
- Apply retention and access controls to certificates in line with your customer and regulatory obligations.
Ways to get it done
Scan to folder and manual attach
Effort: A day or two, minimal benefit beyond searchability.
Shops wanting digital filing only, with no extraction.
Commercial AP or intake product
Effort: Weeks, with subscription fees and limited handling of mill certs.
Teams focused on invoices who accept cloud processing and can build an E2 hand-off.
Partner build with Netray
Effort: A scoped project of several weeks, delivered with an E2 domain partner.
Shops with certs, CoCs, drawings and receiving paperwork who need on-prem extraction tied to E2 data.
Pitfalls
- !Poor scan quality is the top cause of errors; fix scanner settings before blaming the model.
- !Packing slips list vendor part numbers, not yours; you need a cross-reference to match lines.
- !Quantity unit differences such as feet versus pieces cause silent receiving errors; extract and compare units.
- !Handwritten notes on certs are unreliable; require review for any hand-entered value.
- !Posting receipts automatically at moderate confidence corrupts inventory and cost; set conservative thresholds.
Frequently asked questions
Does E2 include OCR?
Not as a built-in capability that reads and matches documents to POs. E2 can store linked documents, but extraction and matching come from an external pipeline. Check your version and any add-on modules for features that may have been added.
Can OCR read mill test reports?
Yes, with limits. Heat number, grade and values can be extracted from typed reports, but formats vary widely. Plan a review step, and a per-mill template or model tuning if you receive from a few repeat sources.
What confidence threshold should we use?
Start high and send most items for review, then lower the threshold per field as you measure accuracy on your own documents. Financial totals and quantities warrant stricter thresholds than descriptions or dates.
Can we keep all of this on-prem?
Yes. OCR engines and document models run locally, matching uses a read-only SQL login, and storage stays on your servers. Volume determines whether you need a GPU, and many small shops can start without one.
How is this different from email automation?
Email automation focuses on intake of orders and RFQs from mailboxes. Document OCR focuses on paperwork around receiving, payables and quality. They share the extraction engine but have different matching logic and review rules.
Want Document OCR working on your ECI E2 Shop System?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment