Sign up to newsletter

Magazine subscription

BBT March/April 2019 cover
March/April 2019

Expense management: the emergence of AI

How much impact will AI have on expense management?

Automation of travel expenses would be a welcome application of AI, but robots aren’t taking over just yet

Be afraid, be very afraid. Artificial intelligence (AI) is coming to get you. That is what a report from consulting firm PwC would have us believe. The company’s March 2017 UK Economic Outlook forecast that 30 per cent of jobs in the UK, many of them office roles, could be at risk from automation in the next 15 years.

PwC’s chief economist John Hawksworth said at the time: “No industry is entirely immune from future advances in robotics and AI.”

It’s not entirely bad news. “Automating more manual and repetitive tasks will eliminate some existing jobs,” he said. “But it could also enable some workers to focus on higher value, more rewarding and creative work, removing the monotony from our day jobs.”

Nowhere is this automation going to be more keenly felt than in the area of expense claims, regularly named as employees’ number one hassle. In fact, AI cannot come soon enough for some business travellers.

Chris Baker, managing director of UK enterprise at Concur, says: “We see AI as the next fundamental leap in technology; companies across the travel space and beyond are hard at work developing and furthering AI.

“Although we are not quite at the stage of being able to offer ‘true AI’ yet, we’re seeing a lot of machine-learning systems being built into chatbots, scheduling tools and financial analysis; a trend that is only set to continue picking up pace.”

At the heart of Concur’s approach is the idea of a perfect trip, says Baker, “one that is seamless, connected and of benefit for the traveller, travel manager and company”.

“This is what has led us to develop not just an expense report that writes itself, but an entire connected ecosystem of applications that automatically feeds back information into Concur, allowing companies to capture booking and spend information really easily.”

The company is currently adopting a combined approach of augmenting human brain power and machine-learning. “Ultimately, the point of using these systems is to speed up processes for those in travel and finance, rather than give them a complex tool to learn how to run and analyse,” says Baker.

But, he says, at this moment in time, no matter how intelligent a system being used is, it cannot be a substitute for human intuition and analysis – man and machine need to work side by side. “Fundamentally, what AI and machine-learning bring to the table is the heavy lifting. As with any kind of automation, these systems are able to analyse great swathes of data that would take a team of employees weeks to sift through, in a matter of minutes.”

Optimised opticals
Bob Neveu, president of Certify, which recently merged with Nexonia, ExpenseWatch and Tallie in a US$125 million deal, says that while many companies claim to automate the expense management process fully, the reality is a little different. “You can take a picture of a receipt, have the system get the vendor name and location using optical character recognition and, if it is a hotel, the check-in and out dates, and it will say instantly whether it is in or out of policy. We do that incredibly efficiently. Others have gone along that road; most do not do it in real-time.”

The Certify system then takes that data and matches it to the corporate card information which is automatically brought into the system. “We fully automated that,” says Neveu, “so on the first of the month we can say, ‘Bob, your expense report is ready for you to review.’”

One interesting area of AI for Certify is in benchmarking. “One of the big challenges for first-time adopters is expense policy,” says Neveu. “We have millions of users and billions of dollars of receipts and we have the ability to anonymise and share that back with our customers in the form of benchmarking policy and tools.”

Dan Fitzgerald, Traveldoo’s chief product officer, says that AI’s promise is in making good decisions around multidimensional problems that a person would struggle to manage in a reasonable time. “AI definitely has the potential to make things easier,” he says. “But the hype is ahead of the reality right now.”

The goal at the end of the road, believes Fitzgerald, is the ‘CEO experience’, where the traveller would literally have to do nothing to submit an expense claim. “No one likes doing expenses,” he says.

AI coming into expense management will be a gradual process, he believes. “The successful companies will be those that build tools and can test scenarios more effectively than others – and be wrong faster than anyone else,” he says.

Fyle, which launched in early 2016, is an automated expense management platform aimed at SMEs that puts AI at the heart of its platform. Its marketing tagline says: “We automate your expense reports, so you can do better things.”

Yashwanth Madhusudan, Fyle’s cofounder, says the idea came from his own frustrations as a frequent business traveller. “Over five years, my travel increased a lot and the time I spent on travel-related expenses increased. I had a feeling that I lost a lot of my own money.”

Fyle works in a different way to many other expense management systems. Its key product is an AI-powered plug-in that works with Gmail, and it has recently rolled one out for Outlook 365 – the first such product approved by Microsoft. Users install the plug-in and then click on emails to identify them as expense-related; the system then extracts the information from the unstructured data in the email.

For receipt images captured through its mobile app, Fyle compares the image to a large database of standard receipts from which it knows where to find the relevant information, such as the merchant, date, location and amount. For images not in the database, the system uses machine-learning to make informed guesses, which get better over time.

Death of paper receipts?
There is some attachment by tax authorities around the world to paper receipts, but even this sector is seeing some liberalisation, says Madhusudan. He cites those till receipts that only last a few days before fading so badly they are unreadable. “Tax authorities are also cognisant of this,” he says. “All they really want is a verifiable trail of when the expense was incurred, what it was for, who approved it and when it was paid out.”

“Receipts will disappear,” says Traveldoo’s Fitzgerald. “Why do you have to carry around a receipt?”

But, he says, AI currently has a problem with trust. Do you trust a system enough to take a photo of your receipt and then discard it? “It is important to think about when you tell people they no longer need to keep a receipt, particularly if you are going to remove choice,” he says.

But the power lies in the database of receipts Traveldoo has already collected. “We have millions of receipts and correct analysis of those entered by the user. It is the perfect database to drive smarter, predictive analysis.”

Voice recognition
Voice recognition technology looks like an interesting new frontier whose boundaries are being pushed by advances in AI but again, the challenge of paper receipts remains. “The big challenge with voice is company policy around the receipt,” says Certify’s Neveu. “Here in the US, if the expense is for less than US$75 and it’s not a hotel charge, then you don’t need a receipt. Yet most companies will never adopt that.” But he does predict that electronic receipting will become standard.”

Traveldoo’s Fitzgerald thinks voice has some way to go, too. “It annoys me because I am talking to something and all of my expectations are aligned with having a proper conversation. Voice for me is still in that underwhelming phase.”

What does the buyer side feel about AI? Chris Pouney, of Severnside Consulting, says that great strides have been made in automating expense management, such as being able to photograph expenses and automated upload, but it has not yet lived up to the hype. “There is much more to do for it to be effective for most business travellers,” he says. “Will we ever see hotel folio data transmitted straight to a credit card, for example, which has been talked about for many years?

“Intuitive approvals are another area where companies can benefit [from AI]. I see companies who install complex pre-trip approval processes, but what are they trying to achieve in reality? Most are trying to reduce certain trip types – high -value transactions or high lost savings – so approvals should be targeted in this area. I would like to see more emphasis on how many trips are approved or rejected. If all are approved then where is the value?”

So is PwC’s prediction of one-third of us being out of jobs in 15 years’ time realistic? Concur’s Baker thinks not. “The overall problem with AI comes with the categorisation of it in all its various formats,” he says. “At the moment, it’s heading the same way as the cloud or describing things as ‘next-gen’.

“Fundamentally, there isn’t a company in the world at the moment that is offering true AI. Instead, there is machine-learning, augmented intelligence, chatbots and business intelligence.

“The past few years have seen a surge of interest in programs and solutions that can think and adapt themselves, with many – if not all – of these terms being used interchangeably.

“For now, what does need clarity is exactly how autonomous these technologies are – and how much work is needed from employees analysing the data alongside them, allowing businesses to make smart decisions about how they approach reporting and insights.”

It seems we can sleep easy in our beds for the time being.

Add new comment