Swiss banking giant UBS came out of a five-year travel program strategy in 2015 with a globalized program that optimized policies and contracts. But, said UBS global travel lead Mark Cuschieri, the only data available for decision-making was coming from the agency, which limited both the travel team and UBS executives in numerous ways moving forward.
While agency data is rich with detail, it often trails actual travel activity by at least 30 days. Plus, a single data source of any kind produces only a partial picture for travel. True visibility relies on aggregating and integrating sources like global distribution systems, suppliers, payment solutions and expense tools, said Cuschieri, but also sources outside travel like market data, hotel availability data and catering.
“We weren’t accessing all the data sources we had to better inform UBS businesses,” he added. “Plus, we were only offering historical data, which is nice to identify trends but the business wants to know, ‘What am I doing today? And what if I change what I’m doing today or tomorrow? What impact will that have on my goals, my targets?’”
UBS had contracted with PredictX on a broad shared services technology initiative, including travel. “Travel [data] is very difficult to manage, comparatively, and there’s a healthy appreciation of that at UBS. The work of a bank is to make decisions from information, so it’s quite natural that they were open to that promise on the travel side,” said PredictX CEO Keesup Choe.
UBS wanted deep visibility into total trip cost, and the team wanted actionable pre-trip data, as well as a way to predict the future impact of travel on UBS businesses. One of the challenges peculiar to travel is that the company is quite often not the owner of the data it needs to analyze its own programs. Instead, it’s the travel management company, the GDS or the suppliers that own the data. As a result, the corporation is not the one that defines the data format, the content or the accuracy.
“That’s unlike any other core system used to managed other vertical within the organization. Customer relationships, human resources, enterprise resource planning ... they’re operated and maintained by the enterprise,” said Choe.
Leveraging Machine Learning
As a result, travel departments have a unique challenge to re-create total trip spend. Even if the traveler uses all the approved channels and payment mechanisms, there are ancillary hotel and airline charges, ground transportation costs and meals, and there could be multiple legs of the trip. “An executive assistant, perhaps, could manage this for a single trip, but it’s very difficult to do at scale,” said Choe.
At UBS, machine learning algorithms are picking apart the data from multiple sources and grouping transactions based on probabilities to create the total trip cost. Prior to machine learning, this type of data might be aggregated using a static rules set that would miss elements like whether a hotel room transaction was actually part of a meeting expense. Machine learning algorithms can figure out new exceptions and possibilities without manual intervention.
“It can automatically identify and correct data without having to go back to the data source,” said Choe. This is machine learning at its most basic, and it is facilitating more advanced automation and more strategic action.
The Analysis Story
UBS has taken its machine learning-enhanced data set and transitioned it into AI-enhanced analysis and reporting for UBS executive management and business leads, said Cuschieri. As machine learning identifies spending trends, it automatically pulls them out of the data mine in a visualized presentation designed for key stakeholders, who want to understand its effect on UBS business.
Because these stakeholders aren’t interested in travel, per se, “you want something very clean and very clear,” said Cuschieri. “You have to have a report that shows how much you’re spending and where. This is how you are spending it and what you’re doing against historical levels.”
Direct data feeds into the tool allow UBS to work with very close to real-time information, so business leads know what is happening “right now.” At the same time, machine learning is correlating and grouping data in the background to ensure accuracy without manual intervention.
Cuschieri said UBS also is working on a predictive data set, which means the reporting will show a visual representation of current spend and a prediction of whether business leads will hit their targets, come in below them or are on a trajectory to beat them. The analysis includes pre-trip data so the information is actionable. “It allows us to get ahead of the curve, rather than trying to explain what happened at the end of the year,” said Cuschieri.
The automated analysis and projections are impressive in themselves, but there’s more to this AI story: The reporting uses natural language processing capabilities to write itself.
“[The technology] engages senior management with more visualization and less text, but the text is actually generated by the machines,” said Cuschieri. Also, UBS offers businesses a self-serve model in which leaders go into the tool themselves, where they can look at current data and model what-if scenarios.
Asked what is left for the travel managers to do, Cuschieri said there is plenty, but it does change the skill set required. “The businesses can depend on the travel team much more as subject matter experts,” he said. “Too many people go, ‘Here’s the report. Go away.’ That’s not what we do. We support the business in better understanding their goals and driving reports that [show] meaningful actions … around behavior management, demand management, etc.”
Machine Learning & Managing Travel
Machine learning is powering the business of managing travel as well. Integrating broad and enhanced data sets has not only provided a better picture of traveler spend patterns but also of their motivations. The improved visibility has increased negotiating power for UBS buyers, according to Cushieri.
“Managed travel has been the only industry where we agree to enter into negotiations where the supplier knows more than [the buyer],” he said. “How many times do you actually get your negotiated rate [at preferred hotel properties], and how often do travelers have to book away because of availability? We now have that kind of information on historical trends, variances against what we actually contracted and a much better understanding of compliance at the property level. We have far more information that enables us to go back to suppliers and say, ‘Wait a second. From what we’ve agreed, we’re not getting the availability.’ Machine learning is helping us identify those things far better and quicker than humans could.”
At a certain level, said Cuschieri, there’s an element of trust because humans can’t process the amount of data that goes into producing those insights. “We’ve seen successful models in other industries. You have to jump in if you want to move forward.”