Even with the rise in eInvoicing and electronic B2B payments, disconnects, miscommunication and erroneous transactions plague the accounts payable department. Finding where those errors are via manual auditing can be a headache, too.
Supply chain management company APEX Analytix recently released a new solution, APEX Archimedes, that aims to not only automate the AP recovery audit process, but to enhance its capabilities via machine learning, artificial intelligence and other next-generation technologies. Danny Thompson, senior vice president of market and product strategy at APEX Analytix, recently spoke with PYMNTS to explain why such tools are necessary to solve a frequent problem.
“There are a lot of really common mistakes,” he said, alluding to errors in accounts payable that can lead to either overpaying suppliers or, in many cases, even paying them twice. According to Thompson, those mistakes have a variety of factors behind them.
The most common, he said, is that a single vendor will land on multiple vendor master files. When an invoice from one of these vendors is sent and processed in the AP department, it may get processed against one copy of the vendor master; if that vendor sends another invoice, it could get processed against the other vendor master, meaning the same invoice may be in the system and get paid without anyone catching it.
“We think roughly 30 percent of duplicate payments are the result of some sort of vendor master issue, either as a duplicate vendor or some other error associated with vendor master,” the executive explained.
One reason why it’s so difficult for an AP department to catch a duplicate invoice is that they aren’t always exact copies of each other.
“If a supplier has set up electronic invoicing but they also send through a paper invoice,” for instance, Thompson said. Or, if a supplier with access to a client’s supplier portal doesn’t see their own bill in the system yet, or if they haven’t been paid by the agreed-upon date, that vendor will often send an invoice again. But that duplicate invoice may have a different date on it, or a different invoice number – and any information on the document that’s changed means ERP or other software solutions may not always pick up that it’s a bill already in the system.
In yet another way, Thompson explained, that a company may overpay a vendor, there are instances in which a supplier sends a summary statement of outstanding balances, but the AP department may actually process that as an entirely new bill.
If a company has recently gone through some type of merger, or if their back-end systems are being upgraded and consolidated, a bill may live in two ERP systems and see duplicate payment in that way too.
There are even cases, he added, in which a supplier will send fraudulent invoices in hopes of getting paid extra, ”but they are very rare,” Thompson noted. “These are mostly just errors.”
Clearly, there is a lot of room for error. Recent research from Tipalti found that this is an especially large problem for companies that are operating with suppliers across borders.
In order to prevent AP professionals from having to manually go through troves of invoice data and match them with finances, APEX is deploying machine learning, artificial intelligence and other data analytics capabilities that can identify overpayments.
Basic analytics tools won’t always cut it, Thompson said, because sometimes key information can be found buried within conversational emails or in supplier contracts that can’t be easily identified and aggregated. In these cases, APEX uses IBM Watson to be able to capture this type of information from natural language more efficiently, he explained.
“I think the real value of deploying machine learning in this process is the value it brings to processing large amounts of data,” he said, adding that audits can often involve assessing data from a couple of years worth of invoices, supplier contracts and contact information, and other documents. AEPX uses the information found within its own clients’ systems, in addition to APEX’s own collected data, meaning something as straight-forward as identifying which supplier contact information is the best to use can be automated.
Even automating the data matching or reconciliation process can miss errors, Thompson noted. AP departments must go further, and machine learning, AI and other sophisticated technologies can get them there.
“What we’re doing with machine learning is doing some more sophisticated scoring of suppliers and invoices,” he explained. “We’re looking at a lot of characteristics of those transactions that we see are common across invoice and suppliers that result in claims. There are so many different characteristics of a supplier that can be taken into account other than just a direct comparison of invoice-to-invoice or invoice-to-contract.”