Next-Gen Expense Management

2016 was the year expense management players got more aggressive, strategic and intelligent. While the sector has unarguably consolidated, that hasn’t stopped new players, armed with intuitive technologies and consumer-style interfaces, from launching solutions that could turn into prime acquisition targets for established providers looking to leapfrog competitors.

Last year also continued the shift from submitting expense reports associated with trips or time frames to real-time expense reporting. Increasingly sophisticated data capture through optical character recognition and card integrations have moved this transition along. Machine learning may prove to be the ultimate enabler, however, as systems are configured to identify duplicate charges and to learn to identify outlier transactions. It makes for an increasingly accurate and automated process from data capture through reimbursement.

Most vendors now use OCR and some providers use machine learning technology to scan, auto-fill and categorize expense details, pair receipts with corporate card charges and submit completed expense items to the expense system for reimbursement. Employees have less manual work to dread and are reimbursed quicker. Likewise, corporations get quicker spend visibility.

Even so, true automation requires some nerve on the part of the corporate client, and proficiency varies by provider.

Automation Migration

Abacus began the push to eliminate the expense report in 2013, treating each charge as an individual transaction. That system fills in expense details from a receipt using OCR, geolocation and machine learning to categorize the expense and pair the receipt with a corporate card charge. That same year, Certify launched ReportExecutive, which goes one step further by automatically submitting complete and compliant expenses according to a set schedule. Expensify branded and popularized the term Realtime Expense Reporting after it introduced similar capabilities in 2015. It made RTER its default setting last year. The more receipts/data that go through any of these systems, the more the tools learn how to categorize receipts using machine learning.

The Advisory Board Co. VP of information systems Steven Mandelbaum, who also runs T&E for the firm, has been on the leading edge of this concept. He dropped the company’s post-trip expense reporting system in 2015 and moved The Advisory Board to Expense Now, a proprietary expense system in which managers approve transactions. It pulls expenses from the employee’s corporate card and pre-populates an email-like inbox. The employee chooses a purpose for each expense from a drop-down menu of meeting dates populated by Salesforce and then submits each expense for reimbursement. On the back end, each manager can see all transactions associated with his or her team and can filter expenses in various ways—including by employee, date or expense type—for approvals.

After nearly 18 months, Mandelbaum said the tool has proven that “expense reports are artificial groups.” The change has cut the number of unsubmitted transactions in half. He notes, however, that the tool works not in isolation but in conjunction with a comprehensive travel policy, established procedures and effective corporate card management, among other factors.

According to Expensify CEO David Barrett, adoption of RTER and submission of individual transactions for approval have been “incredibly high” among new clients that never knew anything else. He said “tens of thousands of companies” use RTER. Yet, he also said established clients tend to turn the setting off and revert to reports. Those reports are valuable to the finance department, Barrett said; batching an individual’s expenses together, whether monthly or by business trip, reduces the number of reimbursements an accountant must reconcile with the general ledger.

Michelle De Costa likes the idea of transaction-based, automated expense submission. “In theory, I love the concept because it lends to the ease of experience for travelers. Plus, the faster you get in expenses, the faster you can bill out to clients,” she said.

Automated expense reports are one thing, but automated expense approvals are a different story for De Costa. She piloted Abukai’s automated expense submission when she managed T&E for Sapient. The pilot allowed Sapient travelers to capture receipts using OCR, and the system auto-submitted completed expenses. De Costa said the test run was “pretty successful,” but as The Advisory Board did, Sapient stopped short of automating approvals.

In her current role as head of corporate travel for Liberty Mutual Insurance, as well, that would be a step too far. “You need the checks and balances in the middle,” she said. She underscored that Liberty Mutual would be wary of systems that bypassed human approval, noting that at the very least, the company would need to implement auditing processes to scour a percent of expenses for fraud or out-of-policy charges.

Rules Engines & Then Artificial Intelligence

Certify president Bob Neveu, though, maintains that automated approvals do not have to be rubber stamps. “There’s a big difference between [being able to] flag exceptions, incorrect or fraudulent expenses versus just auto-approval and instant reimbursement. ... No customer is looking for that.”

Rules engines can analyze expenses against policy, such as merchants that the company does not allow, and artificial intelligence can go farther to identify expenses that are abnormal for the traveler, job description or kind of trip. Certify and Expensify, for example enable companies to upload their travel policies and set exceptions for certain employees or scenarios. The systems’ rules engines then flag duplicate expenses and those that fall outside policy.

AI then can look across an expense-report-free universe and can view every transaction, comparing it to an individual’s history or other employees’ histories. Or AI can group transactions by employee, date, merchant type, expense amount or another characteristic to search for irregularities or suspicious patterns. Certify, for example, lets “the robots take over” to review and compare a given company’s spend data, including submissions that are in progress, explained a spokesperson. Expensify, meanwhile, aims to combine AI with its Big Data stack to compare expenses to past ones so the system can flag risky or abnormal submissions, such as higher-than-normal spending for a particular employee. Barrett called it “finding the needle in the haystack for you without giving you the haystack.”

According to Oversight Systems, which audits expenses on behalf of corporate travel clients, removing manager approval from the process makes it easier to identify the expenses that are not OK. In a January whitepaper, Oversight Systems argued that managers are more concerned with approving reports quickly than with scrutinizing expense submissions for anything other than the most obvious errors or problems. A typical manager, for example, would not pull out previous expense reports to identify patterns of behavior. And after manager approval, the second line of defense against rogue expenses would be an audit team, which reviews just 10 to 20 percent of expense reports.

Oversight Systems accesses its clients’ expense data via integrations with expense reporting providers like Concur, Oracle and SAP and then combines machine learning with analytical tools to determine whether an employee submitted the same dinner expense twice, for example, or two employees submitted the same dinner. It additionally detects out-of-policy charges based on keywords, merchants and suspicious patterns. The company claimed its clients cut in half the time and effort it takes to review, audit and resolve expenses.

The key, according to the company, is to review line-item expenses as they come in against policy, company data and external data and to flag them or approve them for reimbursement. Then, the manager can “devote more time to addressing risk instead of searching for it.”

Smart Bots & Other AI Applications

Some of the most notable expense acquisitions of late involved companies that have AI capabilities: Coupa acquired Spend360 in January 2017, and Deem acquired Olset in November 2016. American Express Global Business Travel acquired KDS the month prior, though KDS uses AI only on the booking side. The rush suggests a proliferation of upcoming AI-enabled capabilities.

Several expense providers are experimenting with AI bots, more genially known as expense assistants, to communicate with customers and facilitate expense reporting via messaging platforms and voice command.

In February, Concur announced a chatbot pilot with messaging tool Slack that allows mutual clients to request travel itineraries, view expense report summaries, upload receipts and submit quick expenses by typing “@Concur” before each command. Real-time expense reporting software Abacus launched a chatbot with Slack in June, and Abacus updated it in October to allow users to submit expenses, which appear in a Suggested Expense list, within Slack. Also within Slack, users can add details as needed.

Last year, Expensify started using machine learning for customer service through its Concierge engine. Barrett admitted that customer service chatbots can be frustrating because they’re typically built with natural-language processing while people chatting online typically use a text message dialect. Barrett said Expensify built its chatbot for a world “where words are always misspelled and there’s never any grammar cohesion.” If the system still doesn’t understand, the question routes to a human—and the system remembers the answer for next time.

Personal assistants like the voice-activated Amazon Alexa and Google Home have become popular in the consumer world, and Coupa general manager of expense Sunny Manivannan believes voice recognition will gain more traction among expense technology providers this year. Coupa launched the capability in 2015 along with Traveldoo, while Oracle Fusion Expenses has had voice capability since 2012.

Meanwhile, AI is helping expense providers as they inch into travel and related functions. The new TravelBank expense app uses AI to predict costs before trips are booked, giving companies and their travelers more flexibility and opportunities for savings, though the app does not provide any shopping or booking capability.

And Expensify plans to use AI to identify multiple employees traveling to the same location so they can organize. The company also aims to use AI for bookkeeping and accounting for companies that don’t have those functions internally.