Summary of "Auto-Code Non-PO Invoices Using AI / ML" training
A Waypoint 23 short course · AI in Finance series
The challenge
Most supplier invoices flow through smoothly with three-way matching (PO, goods receipt, invoice). Non-PO invoices are different. Utilities, subscriptions, professional fees and one-off services fall out of auto-processing and land back on someone's desk, where they have to be coded, routed and approved by hand.
The idea
Your finance team has already done this coding work hundreds of times. Machine learning can learn from that history and predict the right GL codes, cost centres and approvers for new invoices. People then only need to review the exceptions, not every line.
What you'll learn
How an ML layer fits into the existing non-PO workflow, at the coding stage and before business or SME review
A practical three-step approach to get started:
Prepare the non-PO invoices that were rejected from auto-processing
Extract historical invoice data and ERP journal templates (more history means better recommendations)
Prompt: write clear instructions so the AI matches against your reference data and outputs the coding fields you need
How to scale from a workflow one person can run to agentic or API-connected automation
Why AI doesn't "learn from its mistakes" on its own, and how to improve results on purpose through better data and better prompts
Who it's for
Accounts payable and finance operations staff, finance managers, and ERP or process owners who want a practical first use case for AI. No coding background is needed.
Our approach
Each step is taught the way finance teams already work: what triggers the process, what changes when AI steps in, where the control check sits, and what that means for your control environment.
Keep the human accountable. Let AI carry the repetition.