These days, a lot of workers are worried about robots taking their jobs, but now robots are taking jobs that literally no one else was doing—like poring through every inch of people’s dense expense reports.
AppZen, a startup that provides automated auditing services, has just expanded its software, which can now not only read basic items on attached receipts, but also scan entire documents to look for clues about invalid charges. Anant Kale, the company’s CEO, tells Fast Company that the new offering, called ReceiptIQ, can audit 100% of the expense reports that employees submit, spotting them for “accidental fraud . . . or intentional fraud.”
The upgrade required moving beyond optical character recognition to computer vision that can understand the entire receipt, such as recognizing company logos. Many hotels put their logos on receipts rather than spelling out their name, says Kale.
ReceiptIQ analysis goes deeper by looking for context, such as whether car rental bills include a fuel service charge—a penalty for not returning the car with a full gas tank. Companies often won’t reimburse such penalty charges. On cell phone bills, the software can now determine if the employee has a family plan and is trying to get reimbursed for the kids or spouse. Machine learning helps the service better understand what to look for in all those PDFs, emails, and other attachments appended to reports.
Alcohol is another item often exempted from reimbursement. ReceiptIQ reads the names of beverages on receipts and looks them up on a bunch of websites that feature names and recipes of drinks and brand names for types of alcohol. AppZen often pings the web for information, such as checking sites like Yelp or Trip Advisor to verify that the restaurant named is a real place and that the amount on the bill seems in line with what other people spend there. It even checks news and government sites to ensure that the employee hasn’t treated a foreign government official to a prohibited freebie (aka a bribe).
AppZen debuted in spring 2015 and picked up $2.9 million of new funding in June 2016. The company has been testing ReceiptIQ, running 1,700 randomly selected expense reports from clients, totaling over $3.6 million, through its new system. Kale reports that the new tech found twice as many bogus charges as the company’s current software, and 10 times what spot checks by humans would have caught.
Expense report auditing is far from the sexiest aspect of running a business, and that’s exactly the appeal of AppZen. The service promises to not only save humans from drudgery, but it can also read far more than humans, and it can flag only the really suspicious stuff. “When we classify something as high risk, which is typically about 6% to 10% of all expenses, we give the formation back to the company,” says Kale. “And [we] say, ‘You need to look at that.'”
From there, it’s up to the client to figure out how to proceed. AppZen flags a bigger chunk of reports as medium risk and tracks the employee over time to see if sketchy patterns continue that need to be investigated. The majority of reports come out as low risk, says Kale, and AppZen recommends clients just pay them.
Although AppZen advises how a company should proceed with tackling fraud, it doesn’t take responsibility for making mistakes that could get a company in trouble with the IRS.
“We’re not guaranteeing that there is no fraud,” says Kale. “Even if you are in noncompliance, as long as you have the right controls at the company and have put in the right processes, that’s what matters.” The main point, he says, is that clients can audit a lot more paperwork than before. But clients have to decide their “appetite for risk” in setting thresholds for what requires review by a human.
AppZen, which emerged from the 500 Startups accelerator in 2015, has picked up some brand-name clients, including Bandai Namco, Cantor Fitzgerald, Comcast, and Equinox. In August, it became an approved app for Concur, the big expense report and invoice-paying software service. It also works with accounting systems by NetSuite and Oracle.