AP Automation Myth #3: It’s Inaccurate & Error-Prone

Accounts payable automation is an increasingly popular way for businesses large and small to create positive change. Its benefits are real, proven and repeatable. Unfortunately, a few persistent myths are holding many companies back from automating AP. In this exclusive BerkOne blog series, we’re investigating one each month; this entry explores the accuracy of the data extracted in an automated AP workflow.

There’s a part of the human brain that assumes its own infallibility. After all, if you want something done right, you have to do it yourself. But that’s not necessarily true when it comes to accounts payable. Sure, a seasoned AP professional can deliver very good data quality – but they’re unlikely to ever match the accuracy generated when the best of human intelligence is combined with the latest technology.

Let’s start by examining the assumption that humans are the best translators of accurate data. That’s simply not true. To enter information into a system, we need to type. That means a mistake – causing delays, eliminating the potential for early payment discounts and costing the AP department credibility – is always only a single keystroke away. That’s bad.

So, is it best to go the totally opposite direction and automate the process – invoice arrives, invoice is scanned, data is automatically extracted, approval workflow begins – completely? Not exactly. Fact is, an automated AP workflow running entirely on its own probably would introduce errors into the process. Unstructured optical character recognition (OCR), the process by which capture software identifies fields and then “reads” words and numbers on a scanned invoice, is not – and never will be – 100 percent accurate.

Rather, a properly automated accounts payable workflow shouldn’t simply rely on an unstructured OCR engine to capture totally accurate data every time. That would be foolish. Instead, when AP automation is done right, the capture engine attaches confidence scores to each extracted field. Below a certain confidence threshold, a human technician steps in to verify or correct the software’s work. This results in most data being properly interpreted by the OCR software, and human intervention required only when recognition confidence is low. Much of the potential for inaccurate data to work its way into the accounting or ERP system is eliminated.

Even better is the fact that OCR doesn’t operate in a vacuum. By incorporating other databases – a vendor master file, for instance – a data capture system can significantly increase its hit rate. Imagine an invoice from Widget Co., vendor No. 2315, is being scanned. If both these pieces of information – the vendor name and the vendor number – are captured, the OCR system can run them against the vendor master file and develop an extremely high degree of confidence that the information is right. Conversely, if only one is legible on the invoice, the other can be automatically populated, at a slightly lower degree of confidence.

Could a human reference the vendor master file in this way? Yes … but since it’s a constantly changing document, the time taken to do so would be problematic. Software can do it instantly.

Remember, when it comes to data extraction and entry, automated processes aren’t perfect. Neither are humans. But when the two are smartly combined, amazing things happen. Are automated AP workflows prone to inaccurate data? Hardly – if they’re designed to leverage the best of both worlds.

Want to learn more about business process automation? Download our free guide!