Invoice processing went from three days to twenty minutes.

Overview

A regional logistics company with operations across three countries was processing every supplier invoice by hand. Hundreds of invoices arrived each week by email, in every format imaginable: PDFs from large carriers, scanned paper from small subcontractors, spreadsheets from fuel suppliers and the occasional photo taken on a phone.

HASNA TECHNOLOGY mapped the full accounts-payable process, then built an AI-driven pipeline that reads each invoice, checks it against purchase orders and goods receipts in the ERP, posts the ones that match and sends only the real exceptions to a person, with all the context attached.

Ten weeks after kick-off, 92% of invoices go through without anyone touching them, and the average invoice is processed in about twenty minutes instead of three days.

The problem

Too many invoices, too many formats

The finance team of six received between 900 and 1,400 supplier invoices a month. There was no standard format. Even invoices from the same supplier changed layout when the supplier updated its billing system, which made template-based OCR tools unreliable. The team had tried one before and given up after a few months because it needed constant re-training.

Every invoice was typed in twice

The process looked like this:

1. An accounts assistant opened the shared mailbox, downloaded the attachment and saved it to a folder.
2. They typed the supplier, invoice number, dates, amounts and tax into the ERP.
3. They searched for the matching purchase order and goods receipt, often by emailing operations to ask whether a delivery had actually happened.
4. They forwarded the invoice to the budget owner for approval and waited.
5. Once approved, they posted it and filed the PDF.

Each invoice took between 8 and 15 minutes of hands-on work, but the elapsed time was about three days, mostly spent waiting for information and approvals.

Month-end kept slipping

Because of the backlog, invoices were often posted in the wrong period. Month-end close regularly ran four to five days late, and management reports were based on incomplete costs. Suppliers chased payments, and the company missed early-payment discounts it was entitled to.

The real cost

When we added it up during discovery, the team was spending roughly 1,200 hours a year on data entry and chasing, time that should have gone into supplier management, cash-flow planning and analysis.

Our approach

We didn't start with technology. We spent the first two weeks with the finance and operations teams, mapping the process as it really happened rather than as it was written down.

• We sampled 400 real invoices to understand the formats, the error rates and the most common reasons for mismatches.
• We measured each step, so we could agree on a baseline and prove the result later.
• We separated the rules from the judgement. About 85% of the decisions the team made followed clear rules (amounts match, PO exists, goods received). The rest needed a person. That split defined what to automate and what to keep human.
• We agreed what "done" looks like: at least 80% touchless processing, no loss of control for approvers and a full audit trail for the auditors.

The solution

The finished system has four parts, each one replacing a manual step.

1. AI invoice reading

Every email that reaches the accounts-payable mailbox is picked up automatically. An AI document model reads the attachment, whatever the layout, and extracts the supplier, invoice number, dates, currency, line items, tax and totals. Each field comes with a confidence score. Because the model understands documents rather than relying on fixed templates, a new supplier format works from day one.

2. Automatic three-way matching

The extracted data is checked against the ERP:

• Does the supplier exist, and do the bank details match the supplier master?
• Is there an open purchase order, and do the quantities and prices agree within the agreed tolerance?
• Has the goods receipt been posted?
• Is this invoice a duplicate of one already received?

If everything matches, the invoice is posted straight into the ERP with the PDF attached. Nobody has to open it.

3. Smart exceptions

When something doesn't match, the invoice is routed to the right person instead of a shared inbox. A price difference goes to purchasing; a missing goods receipt goes to the warehouse; a missing PO goes to the budget owner. Each person sees a short summary of what's wrong, the invoice image and the PO side by side, and can resolve it in one click from email or Microsoft Teams.

4. Live dashboard

The finance manager has a dashboard showing invoices received, posted automatically, waiting for action and overdue, broken down by supplier and by reason. For the first time, the team can see where invoices get stuck and fix the root cause with the supplier.

Implementation

Running the pilot in parallel was important. The team could compare every automated result with what they would have done by hand, which built trust before switching over. We tuned the matching tolerances twice during the pilot based on their feedback.

The results

After three months in production:

• 92% of invoices are processed touchless, from email to posted entry.
• Average processing time dropped from about three days to 20 minutes.
• More than 1,200 hours a year were given back to the finance team.
• Month-end close is on time, and management reports include all costs.
• Early-payment discounts are now captured consistently.
• Duplicate payments stopped: the system caught 14 duplicate invoices in the first quarter.

Nobody lost their job. The two accounts assistants who spent most of their time on data entry now handle supplier relationships and cash-flow forecasting.

What's next

With invoices under control, the company is now automating the next steps in the same way: supplier statement reconciliation and expense claims from drivers, both using the same AI reading and approval engine.

Key takeaways

• Start with the process, not the tool. The biggest gains came from fixing how exceptions were routed, not just from reading invoices faster.
• Keep people in control. Approvers still approve; they just get the right information at the right time.
• Prove it with numbers. Measuring the baseline in week one made the result easy to show to management.