How We Built an AI Agent for a Legacy ERP to Automate 78% of Carrier Invoice Processing

Quick answer: Bacancy Technology built an AI agent that reads every carrier invoice, checks it against the agreed rate and the shipment data in the ERP, and prepares it for payment. Billing staff now review only the invoices the agent flags, and carriers are paid on time.

Outcomes deliveredDelivered March 2026 to August 2026 by a team of 5
78%

Of carrier invoices approved

With no manual data entry

Under 3 min

Review time per flagged invoice

Down from 12 to 15 minutes

2.4%

Of monthly carrier spend recovered

From overbilled accessorial charges

4 days

To close carrier payables at month-end

Down from 10 days of manual matching

About the Client

Qevrixa is an Australia-based third-party logistics (3PL) provider that manages full load and part load freight for shippers across Australia. Every month, its billing team matched thousands of carrier invoices by hand against rate confirmations, consignment notes, proofs of delivery (PODs), and shipment records in a 15-year-old on-premise ERP.

Bacancy Technology built an AI agent that reads every carrier invoice, checks it against the agreed rate and the shipment data in the ERP, and prepares it for payment. Because the ERP had no modern API, our AI agent legacy ERP integration works through a read-only data layer and the ERP’s own batch import, so the agent never writes to accounting tables directly. Billing staff now review only the invoices the agent flags, and carriers are paid on time.

Industry
Logistics
Region
Australia
Team
5 team members
Timeline
March 2026 to August 2026
Project size
Non-disclosable

Challenges & Solutions

Built an AI agent that matches carrier invoices to rate confirmations, consignment notes, PODs, and ERP shipment records.

Challenge 1: Manual invoice matching

Qevrixa's billing team keyed more than 9,000 carrier invoices a month into the ERP by hand. Each invoice took 12 to 15 minutes to match against its rate card, consignment note, and POD, and the month-end backlog delayed both carrier payments and customer rebilling.

Solution: Through our AI agent development services, we built an intake agent on LangGraph that collects invoices from EDI, email and carrier portals. Azure AI Document Intelligence extracts structured fields from PDFs and scans, and GPT-4o on Azure OpenAI normalizes carrier-specific line items, subcontractor invoices without standard layouts, and tax lines into one standard invoice schema.

Challenge 2: Invoices in every format

Invoices arrived as EDI files, emailed PDFs, scanned images and carrier portal exports from more than 200 carriers, from national linehaul operators to regional subcontractors. Line items, charge codes, Australian Business Numbers (ABNs) and Goods and Services Tax (GST) amounts sat in different places on every carrier's tax invoice.

Solution: Our team connected the agent to the legacy ERP without changing its core. A read-only replica synced over ODBC gives the agent current shipment, rate, and carrier data. Approved invoices go back only through the ERP’s own batch import files, which the ERP validates the same way it validates manual entries.

Challenge 3: Undetected overbilling

Accessorial charges such as waiting time, unloading fees, tail-lift and fuel surcharges often did not match the agreed rate. With staff checking under time pressure, overbilling went undetected, and disputes surfaced weeks after payment.

Solution: Drawing on our experience as a Logistics Application Development Company, we built a rule-based rate-check engine between the agent and the ERP. It compares every charge with the contracted rate, the weekly fuel surcharge table, and the accessorial terms in each carrier agreement. Any invoice with a variance above Qevrixa’s tolerance is held for review, so the model never makes the final payment decision.

Challenge 4: A legacy ERP with no API

The ERP had no REST or SOAP API, only scheduled flat-file imports and read access through ODBC. IT would not allow an AI system to write to accounts payable tables, and auditors needed a clear record of who approved every payment.

Solution: Our AI engineers worked as an extension of Qevrixa’s billing team and built a React review portal where staff see each flagged invoice next to its rate confirmation, consignment note, and POD, approve or dispute it, and send an auto-drafted dispute email to the carrier. Every decision is logged with the user, timestamp, and reason for audits.

Agent workflow

How the carrier invoice agent works

Connected the agent to a legacy ERP through a read-only data layer and the ERP's native batch import, with no direct database writes.

  1. Qevrixa's carrier invoice AI agent
    1. Intake
      Carrier invoicesEDI, email, scans, carrier portals
      Intake agentLangGraph
    2. Extract
      Field extractionAzure AI Document Intelligence
      NormalizeGPT-4o on Azure OpenAI
    3. Match
      Read-only ERP replicaSynced over ODBC
      Three-way matchRate confirmation, consignment note, POD
    4. Check
      Rate-check engineAccessorials, fuel surcharge, ABN, GST
    5. Review
      Exception review queueBilling staff approve or dispute
      Carrier dispute emailAuto-drafted
    6. Post
      ERP batch importNative import, no direct writes
      Audit logUser, timestamp and reason

0 direct writes to ERP accounting tables; every posting goes through native batch import. 100% of payment approvals logged with user, timestamp, and reason for audits.

Project Snapshot

Qevrixa carrier invoice review dashboard

Core Features

  • Multi-channel invoice intakeMulti-channel invoice intake for EDI invoice files, emailed PDFs, scans and carrier portal exports, with support for structured and unstructured invoice layouts.
  • Three-way matchingThree-way matching of every invoice against the rate confirmation, the consignment note and POD, and the shipment record in the legacy ERP.
  • Charge and fuel levy validationAdditional charge and fuel levy validation against carrier contracts, with ABN and GST checks on every tax invoice.
  • Exception review queueException review queue with side-by-side source documents, auto-drafted carrier dispute emails and a complete approval audit trail.

Technical Stack

Backend
Python | FastAPI
Frontend
React
Database
PostgreSQL
AI agent framework
LangGraph
AI / LLM
GPT-4o (Azure OpenAI)
Document processing
Azure AI Document Intelligence
ERP connectivity
ODBC read-only replica | ERP native batch import
Messaging
Azure Service Bus
Cloud and DevOps
Azure Kubernetes Service | Docker

Ready to start?

2500+ projects delivered by Bacancy Technology

Get access to an experienced team of developers and engineers from Bacancy Technology, handpicked to ace your goals. Kickstart within 48 hours, no-risk trial.

Book A 30 Min Call
14+Years of business experience
1458+Happy customers
30+Countries with happy customers
1050+Agile-enabled employees

More Case Studies

View All Case Studies