Artificial intelligence uncovers outrageous employee expense reports

AI platforms are flagging expenses ranging from first-class airline tickets to strip clubs, according to AppZen.

Guns and bullets. Bulletproof vests. Nail salon excursions, jewelry and hangover kits. These are among the expenses employees bill to their companies, according to AppZen, a provider of financial software. 

You've got to hand it to them, employees are certainly creative in their perceptions of what constitutes business expenses. Businesses process an average of 100,000 expense reports and 700,000 invoices a year, and that means suspect expenses are falling through the cracks, according to AppZen's The State of Business Spend report for Q4 2019.

The problem is that most modern invoice automation systems can look out for two invoices with the same invoice number, or the same amount, and stop a payment that appears to be a duplicate, according to AppZen. But they don't find typos in invoice numbers, duplicates across expense and AP systems – or, ahem, interesting items employees bill to their employers.

SEECheat sheet: Artificial intelligence (free PDF) (TechRepublic)

"Nothing shocks me. It's just how much is going through—that's what surprises me—how far people can take it," said Anant Kale, CEO of AppZen.

Companies spend an average of $4,134 on a so-called high-risk expense report, which includes out-of-policy items or merchants, or non-itemized receipts.

Artificial intelligence can be the panacea, Kale said, and a system embedded with AI can audit 100% of expense reports and invoices.

Other accounts payable gotchas

Some of the most egregious examples of waste, errors, and fraud AppZen found in AP spend in the last quarter are:

  •          Debarred suppliers—invoices sent from a supplier that had been barred by the federal government

  •         Overcharges—charges for a premium web conferencing service that wasn't ordered

  •          Duplicate invoices—the same invoice sent to different divisions within the same company

  •          Double payments—monthly office space rent being paid out twice via expense reimbursement and again by invoice

  •          Fraud—fraudulent supplier invoice with a residential address

  •          Expired contracts—a strategic contract that expired nine months ago.

Delving into spending habits

A trend Kale is seeing more of is groups of people exhibiting similar behaviors. "If one person is buying suitcases from Walmart and expensing that … and no one is checking that, then others start doing that," he said. "I see that kind of trend happening where if this is not stopped right away, it keeps growing and the sky's the limit."

Another issue is that every company has its own culture and policy on what an employee and supplier is supposed to do when submitting an expense report or invoice and what is allowed, he said.

"We've seen a big disconnect between the two. The policy says one thing, but the company doesn't enforce it," Kale said, depending on the type of business. "I find employees expect certain things to be OK company to company."

A system with AI embedded will show analytics and the fact that two people out of 10 on a team tried to buy things they weren't supposed to, for example, he said.

"When you provide this information to managers, they're having conversations about behaviors that need to change," he said.

AppZen's data comes from 1,500 companies in 35 countries and in different verticals that use its platform, including Amazon, Facebook, Novartis and Airbus, he said. 

The AppZen platform uses natural language processing and language augmentation to find if a submitted expense is "non-compliant or someone's just trying to be opportunistic," Kale said. "Our technology is able to catch that and find out if there's a risk associated with that expense."

Because the data a manager or accounts payable person receives is unstructured—meaning it is a contract, invoice, or receipt, people are expected to know what to look for in a transaction and the background information before they become "rubber stamp approvals," he said.  

Humans simply don't have time to look at all these data points, Kale said. "Our goal is to eliminate workloads and human interventions and when a problem is found, it is flagged."

Typically, 90% of the transactions AppZen sees don't have anything wrong with them, he added.

While it varies by company, Kale said the AppZen platform can reduce a company's reimbursement spend by up to 5%.

Also see

screen-shot-2020-01-21-at-1-18-19-pm.png

AppZen has built a machine-learning platform for finance departments that allows customers to review all transactions from invoices to expense reports.

Image: AppZen