Skip to main content

Features Overview

This page describes the main features of TruvioSense and how they work together to automate invoice processing.

Invoice coding predictions

TruvioSense analyzes historical invoices for each vendor and learns typical coding patterns. When a new invoice arrives, it predicts:

  • Vendor number and name
  • Invoice number and currency
  • Net amount and VAT amount
  • G/L account codes per line
  • Quantities and unit costs
  • Financial dimensions (department, cost center, project, etc.)

You review the predictions and apply them with one click, then continue with your normal approval workflow.

Confidence scoring

Each prediction includes a confidence score (0–100%) indicating how certain the model is.

ScoreMeaningRecommended action
90–100%Very reliableQuick scan, apply
80–89%ReliableBrief review, apply
70–79%FairCareful review
60–69%LowVerify all details
Below 60%Very lowManual verification needed

Line items may have individual confidence scores — focus your review on low-confidence lines.

Continuous export and learning

TruvioSense runs in the background via the Job Queue:

  1. When an invoice is posted in ExFlow, it is marked for export
  2. The Job Handler processes documents in batches on schedule
  3. Document images and coding details are sent to the ML service
  4. Models are updated with new patterns

This means predictions improve over time without any manual intervention.

Document monitoring

The TruvioSense Documents page shows all documents in the system with:

  • Export status (image sent, posted doc sent)
  • Error messages for any failed exports
  • Vendor grouping and filtering
  • Document type and reference information

Use this page to verify exports are working and troubleshoot any issues.

Prediction review interface

When you click View Predictions on an invoice, you see:

  • Header section — vendor, invoice number, amounts, currency, overall confidence
  • Line items — account codes, descriptions, quantities, amounts per line with line-level confidence
  • Dimensions — predicted values for each configured dimension (up to 8)