"Manager approvals of expense reports have to be one of the worst controls that exists,” said Oversight Systems president and CEO Patrick Taylor. Why, then, do many companies insist on managers approving every expense report? It exposes those who have violated policy. That’s valuable knowledge despite the effort required by managers and the friction caused for honest travelers. Artificial intelligence, though, can solve for all parties: the good, the bad and the managers.
Expense management controls several risks, including misuse, fraud, waste and abuse, according to Taylor. “If someone is committing fraud on an expense report, there’s a 75.6 percent chance they’re engaged in occupational fraud,” Taylor said, citing the Association of Certified Fraud Examiners.
Still, Oversight data shows that 70 percent of employees are “good travelers” partaking in no risky behavior. Travel managers can reduce friction for these employees, Taylor said. Moreover, a quarter of travelers commit offenses but aren’t considered high risk; their violations may be honest mistakes or owe to unfamiliarity with their companies’ travel policies. “This set of people are the ones you want to educate to make smarter decisions in the future,” he said.
A small number of employees intentionally cover up behavior they know is wrong—passing off a companion’s flight or hotel as a business expense, for example. These few are the ones you have to watch out for the most. “Five percent are engaged in high-risk activities, with a fraction of those engaged in fraud. Those people I don’t mind creating more friction for,” Taylor said.
AI allows managers to reduce friction for the good travelers, educate occasional offenders and thwart high-risk ones. It can catch potentially fraudulent transactions by comparing expensed items across company departments and against other companies. “There’s clearly an opportunity to make [expense management and auditing] more efficient and intelligent and even increase compliance rates,” Certify CEO Bob Neveu said.
Automation Hesitation
The typical expense management workflow requires a traveler to create and submit an expense report for his or her manager’s approval. If the company uses an automated expense management system, the traveler can upload receipts through an app. The manager then sends the report to the accounting or finance department for approval before the company reimburses the traveler. The process can take several days to several months.
Several expense management firms have automated many of these processes, including matching receipts to credit card charges, creating and submitting expense reports for users, approving reports on behalf of managers, and reimbursement. Some even use AI to auto-audit certain processes and flag irregular expenses or suspicious patterns for managers. Likewise, AI-based Xpenditure uses machine learning and predictive analytics to evaluate claims in real time and identify those that may require further investigation. The technology also can classify, organize and transform receipts and invoices into formatted data with “100 percent certainty,” claimed Xpenditure chief innovation and technology officer Moncef Khanfir.
Corporations, however, have hesitated to adopt advanced automation because they want the traveler and/or manager to confirm the accuracy of expenses before funds are reimbursed, Neveu said. They reason that a traveler or manager could blame the technology when a noncompliant expense gets through the system, he added. “We continue to struggle to see the adoption rates there,” he said. “Our customers want control around [those processes], not an AI layer driving that. … Everyone wants to be able to point to someone in this process if there’s an error.”
Expensify, a big proponent of AI and machine learning, similarly admits that a large portion of its customers use only some of its automation capabilities. Thus the expense firm is focusing on improving the manual expense reporting side of its system.
Simplifying Processes for Good Travelers
Still, there’s value in AI-based automation. While many companies believe the most effective way to catch fraud is to review 100 percent of reports, doing so is time consuming and ineffective, as most managers approve reports without scrutiny, especially for employees they trust. “That [approve] button is so easy to hit. It’s a waste of their time,” Taylor said. Others scrutinize a random sample of reports, which means some violations are slipping through the cracks anyway.
Expense management systems like Xpenditure and AI-based third-party audit systems like AppZen and Oversight can scrutinize every report, do it faster and catch violations a human could not. They also can analyze a traveler’s historical expense submission data to identify unusual patterns and recurring bad behavior. AppZen additionally scours the Internet to confirm acceptable merchants and identify suspicious meal or meeting participants.
Considering most business travelers do not participate in risky behavior, Taylor said, it makes sense to use AI to identify the good travelers, eliminate manager approval of their expense reports and reimburse them automatically. AI still can catch questionable expenses should such travelers stray from the straight and narrow. “I’m not ignoring the risk. I’m reducing the friction,” Taylor said.
Educating Offenders & Improving Travel Policies
Likewise, AI can categorize offending travelers based on their risk levels and respond accordingly. Oversight, for instance, can scan a traveler’s historical expenses and can tailor templated emails suited to the violation. It’s a recognition of the difference between a repeat offender versus a new traveler with no history or a long-time traveler with no prior offenses. “Because you don’t want to come down on that person like a ton of bricks,” Taylor said. The system could send repeat offenders a more stringent message and could copy a supervisor on the email.
Machine learning learns from humans, Taylor said, but it also can teach them. In the first quarter, Oversight tweaked its automated emails so they indicate which policy a given employee violated and a link to the travel policy. The idea is to educate travelers rather than reprimand them.
AI also can drive policy improvements, Taylor said. Data amassed from AI reviews of expense reports could reveal that certain violations actually constitute rational behavior; the company could consider updating those parts of its travel policy, such as allowing ride-hailing suppliers like Uber and Lyft. On behalf of one client, Oversight flagged individuals who repeatedly expensed in-room movies, but analysis revealed that those travelers were spending less money on meals, according to Taylor. “It turns out that the
$7 movie is a bargain,” he said.
Likewise, SAP Concur is working on an AI feature, Dynamic Policy, that incorporates a company’s travel policy, data from SAP’s intelligence innovation system Leonardo and other systems and automatically adjusts the policy within Concur as needed, according to Hendrik Vordenbaeumen, SAP Concur Expense global VP of product management for integration, globalization and cards strategy. For instance, a traveler might expense a hotel in New York City for more than the company’s $200 limit; if the data tells the system no hotels are available for less than $200, “the policy will be adjusted automatically to reflect that in New York City it will be higher,” Vordenbaeumen explained. That way, the traveler doesn’t get flagged unnecessarily and the expense reimbursement process isn’t slowed. While Dynamic Policy is still in the early stages, Vordenbaeumen said SAP Concur is a step closer to the idea of an expense report that “creates itself.”
Getting Buy-In
Moving legacy clients to new processes and technologies is more difficult, and so providing clients opt-outs from technologies is a must. “A lot of travel managers would have a heart attack, so it’ll be up to the travel manager to decide whether they want to use it,” Vordenbaeumen said of the Dynamic Policy solution.
But the benefits of AI-powered auditing and automation are evident for large corporations that process thousands of expense reports and need a more efficient process and for small and midsize businesses that are growing and don’t have the manpower to scrutinize expense reports. Some companies outsource these functions, and AI-powered auditing tools can bring that work back in house, eliminating back-and-forth with third-party auditors, AppZen CEO Anant Kale said.
Seeing is believing for his clients, he said, so AppZen typically compares a prospective client’s expense workflow and the App-Zen automated AI process side by side. In an automated expense ecosystem, “they simply have a look at trends and anomalies that keep appearing and make sure that behavioral changes are happening rather than involve themselves on every transaction,” Kale said. AI is “going to catch on as people realize AI apps are able to review and make those decisions around inappropriate, misuse or fraudulent expenses far better than what those managers can do.”