Euna invoice OCR needs field-level confidence decisions
Euna Procurement says its invoice workflow uses optical character recognition to extract fields, shows extracted data beside the original document, and flags low-accuracy fields through confidence scoring before approval. A buyer should retain confidence and review at field level, then keep extraction, matching, approval, and payment states separate so a corrected field never becomes evidence that the invoice was authorized or payable.
Editorial figure by Procurement Technology Current. Source context: Euna Procurement.
An extracted invoice field is a candidate value
The direct buyer answer is to preserve every extracted value as a candidate tied to the original evidence. Euna Procurement's page says its invoice workflow uses optical character recognition for emailed PDF invoices, presents extracted data beside the original, and uses confidence scores to flag low-accuracy fields. For each field, retain the source file and durable identifier or hash, page and region, raw extracted value, normalized value, field type, extraction service and version where available, processing time, confidence output, threshold and rule version, and review state.
Do not keep only the final corrected invoice row. Supplier identity, invoice number, invoice and due dates, currency, purchase-order reference, line description, quantity, unit, price, tax, total, remit-to account, and payment terms can have different error costs and evidence needs. Missing, unreadable, conflicting, inferred, and manually entered values should remain distinguishable. A later reprocessing run should create a new extraction result rather than silently replacing the values presented to the original reviewer.
Confidence thresholds need field-specific policy
A single confidence cutoff is unlikely to express the buyer's control needs. Define which fields can pass automatically, which always require review, which require comparison with supplier master data or a purchase order, and which changes trigger a second approver or independent supplier verification. Preserve the threshold, exception rule, reviewer, displayed source, correction, reason, and time. A high model score is still a provider-generated signal; it does not establish that the invoice is genuine or commercially correct.
Test whether confidence is calibrated for the buyer's document population and languages, layouts, scans, handwritten marks, credits, taxes, and line-item density. Measure expected fields, extracted fields, accepted without change, corrected, rejected, missing, and incorrectly accepted by field class. Aggregate accuracy can hide a weak bank-account or total field behind many correctly read low-consequence labels. The public page does not provide a customer-specific score definition or performance method.
Keep extraction, match, approval, and payment distinct
Euna's page separately describes invoice digitization, two- and three-way matching, approval, okay-to-pay status, and ERP connection. Preserve those as distinct transitions. Extraction asks what the document appears to say. Matching compares selected fields with purchase-order and receipt evidence under defined tolerances. Approval records accountable acceptance or exception authority. Payment and ERP export create additional states owned by finance, treasury, accounting, tax, or another authorized function.
A reviewer correction should not be treated as supplier confirmation, receipt evidence, budget approval, tax determination, duplicate clearance, fraud clearance, or payment authority. Each transition needs its inputs, rule version, result, exception, actor or system, time, and downstream receipt. When a corrected field changes a match or approval result, retain the former result and show why the later decision changed.
Test low-confidence and confidently wrong invoices
A representative demonstration should include a clean digital PDF, low-resolution scan, rotated page, multi-page invoice, credit memo, two currencies, several purchase orders, a split receipt, changed tax, duplicate invoice number, altered remit-to account, and a total that is legible but inconsistent with the lines. Reviewers should reproduce each extraction, confidence value, threshold decision, correction, match, exception route, approval, okay-to-pay state, ERP handoff, and audit record without exposing live banking information.
This analysis reviewed Euna Procurement's official procurement page on September 23, 2026. The page supports the attributed invoice-intake, OCR, side-by-side review, confidence-scoring, matching, approval, audit-trail, okay-to-pay, and ERP-integration positioning. It does not establish a configured field model, score meaning, extraction accuracy, document authenticity, match correctness, approval authority, payment result, fraud prevention, accounting treatment, compliance state, or customer outcome.
Enterprise buyer test
Translate this change into the exact population, record type, workflow stage, decision owner, effective date, and evidence that could be affected. Ask current or prospective providers to demonstrate the named workflow with representative data and an exception—not a polished feature tour. Record what official documentation establishes, what a provider states, what the team observes, and what remains unresolved.
A defensible review also identifies the dependency outside the product. Authority interpretation, policy configuration, data quality, integrations, human judgment, approval rights, release governance, training, and retained evidence may remain customer or service responsibilities. The evaluation should preserve those boundaries instead of treating a technology claim as the complete operating model.
What we will watch next
Procurement Technology Current will watch the named source and affected market records for later evidence that changes status, scope, availability, implementation timing, workflow consequence, or the limits of the initial report. A later announcement does not silently overwrite this dated account; the change ledger preserves the sequence.