Machine learning is changing the face of invoice automation

As anyone who has seen the Terminator movies knows, machines becoming smarter is not always a positive trope. In the real world, however, machine learning has made a much more positive impact turning what used to be static data and analog processes into automated and actionable intelligence. Exhibit #1: invoice automation.

Traditionally, processing invoices is a day killer for the accounts payable staff. These documents take an inordinate amount of time to capture and classify the data contained within them and often require complex matching, coding, approvals, and room for exception processing. This time adds up quickly for AP staff, preventing them from working on more enjoyable and, most importantly, more valuable tasks for the finance team and wider business.

The good news is that machine learning is stripping away the wasted time and frustration associated with those processes, creating efficiencies previously unthinkable only a few years ago. And when supplemented by timely guidance that expert accounts payable personnel can offer, today’s learning-based solutions can elevate the department to an entirely new level.

Always learning, always improving

The key for invoice automation solutions that employ machine learning is to provide a high floor and an even higher ceiling for efficiency and performance. Imagine a scenario where your AP department could:

• Track and learn vendor submission habits to ensure future straight-through processing via a tri-level learning model that automatically reconciles the invoice to the correct company and vendor account within the system of record. It could also apply coding, routing, pre-approval, or matching rules.
• Extract key “header” data such as invoice number, invoice date, subtotal, tax total, invoice total, purchase order number and freight total (if applicable), regardless of format or structure, with built-in controls to ensure the data is sound.
• Extract purchase order invoice “line” data for automated matching to the purchase order, dynamically respecting the rules consumed from the system of record for validation. The data extraction is accurate, regardless of how it is structured within the invoice.

For those invoices requiring approval, the system generates automated reminders to nudge approvers without nagging and can reroute the invoice based on pre-set delegation options, further increasing automation even when invoices might get lost in the shuffle, such as during quarterly reporting.

Invoice automation can also provide analytics around the average time to process or approve an invoice, the accuracy of matching, and more. AP departments need to know how much time and money they save overall versus having invoices captured and matched by a full-time AP clerk working through thousands of invoices a month.

The human role in machine education

In my experience, and executed correctly, automation achieves more than 90% improvements on all these points. However, sometimes companies will get stuck on that 10% that escapes digital processes. For example, an invoice with missing or ambiguous references requires a nudge from the discerning “accounts payable” eye to get things straightened out. To that end, when necessary, a state-of-the-art invoice automation solution intuitively presents exceptions for human response to help reinforce the learning models and achieve the optimal balance of control and automation. For instance, even clicking a data location on an invoice can provide additional intelligence that the learning engines utilize to improve automated extraction and decisioning. A few seconds to educate, when the exceptions are presented, pays off with considerable time savings down the line.

Repurposing learned data

Invoices contain valuable data that can serve purposes other than classification or predictive outcomes when mined through learned extraction. They often have the most accurate information about the vendor including email, phone, URL, named contacts and physical/remit addresses. They can also help glean supply-chain insights when compared across the vendor network. Here’s another example: It can present new vendor contacts, mined from the invoice to the payer system of record for one-click verification and add. This helps fill in the gap in cases where the vendor profile contacts are incomplete or stale.

The Bottomline: This represents just some of the benefits delivered by invoice automation, provided that the vendor has a solid track record in other areas of AP automation. If there’s one essential takeaway here it’s that invoice automation feeds the “machine” with more data. When there’s high-quality, accurate data available from invoices, the entire AP system is elevated. To paraphrase a quote from The Terminator, the future has not been written but it can be learned.