What machine learning has done for the Virgin Velocity program
- 16 April, 2018 07:00
Applying machine learning to the Virgin Velocity Frequent Flyers program has already seen communication effectiveness increase by 10 per cent and given teams the ability to apply advanced analytics at 10 times the pace, its data analytics chief says.
Virgin embarked on a data transformation program a little over 12 months ago, work that’s running alongside a wider digital transformation program across the organisation. The emphasis is twofold: Enhance customer experiences across the Velocity customer loyalty member base by improving redemption offers that are personalised and relevant; and lift the team’s ability to understand and attribute what communications and digital activities are supporting this quest.
As part of this overhaul, the data analytics team adopted DataRobot’s machine learning platform to bolster predictive modelling capability. Oliver Rees, general manager of Torque Data, the data analytics arm of Virgin Australia, told CMO the group’s application of machine learning is about driving personalised customer experiences, and at pace, by being able to run data analytics and modelling more accurately and faster.
“This is so that in a digital environment, we can respond to the information customers are giving us much more quickly,” he said. “In the old days, you’d send a direct mail, or put an ad in print, and wait to see something happened. Then you’d do analysis a few weeks later, try and figure out what works and send again. Now this is happening in fractions of a second. We need to be able to build process and capability to run the analytics in seconds. That is the journey we’re on.”
The ambition is relevance and personalisation. “The better we can get at personalising offers, the deeper we’ll drive engagement with members,” Rees said.
“But with the tools now available to present offers and do outbound and inbound marketing, the complexity of offers is increasing exponentially. We have 400 partners in the loyalty program, and deciding what is the most relevant offer to deliver to someone is very important.”
In a rules-based environment, data analytics teams could try and write a rule that says if a person is 35 years old, has three kids and lives in Manly, they’re likely to be interested in going to Fiji.
“You could then serve content and offers because you recognise someone is interested in a holiday in Fiji. But as soon as they’ve taken that holiday, you have to work out what’s next for that consumer,” Rees explained. “And that happens very quickly. Add in what other experiences they could be interested in, if that consumer has a credit card or insurance, and the permutations go up very quickly. You could write that first rule but you need an engine to build the rest and that is where machine learning fits in.
“It’s delivering the next-best action as close to real time as we can for every one of our members.”
Legacy tools for number crunching or existing infrastructure environments just didn’t cut it or allow the Virgin team to execute fast enough, Rees continued.
“We’re doing stuff far more quickly and with more data, which means a better outcome for members. We’re not wasting the most value commodity we have, which is their time,” he said.
Virgin is predominantly using a combination of transactional data and a series of bespoke segmentation variables developed with Torque to drive redemption models. The attribution models include additional inputs relating to both traditional and non-traditional activity drivers such as media and response variables.
“Whether it’s an email, outbound, other channels – we need to be respectful of their time by presenting something that is meaningful," Rees said. "Each time we can see whether someone is or isn’t interested, we need to use that in a continual learning process, and plug in that response as additional variables. Machine learning in general helps us go down that path.”
With any technological change comes cultural angst, and Rees said the transformation program at Virgin has needed to focus on people first, technology second.
“Teams are familiar with using legacy technology, so how do we to take them on a journey so all are comfortable in the environment? It’s like swapping from manual to automatic car – 20 years ago, if you asked a racing car driver if he’d drive an auto, he’d say never,” he commented.
“So yes, there’s investment in technology and we’re going through that, and we’re going through a major digital transformation. As part of that, we are integrating AI. But the real story is this data transformation journey. And that requires teams of people and analysts to not only apply new tools, but work in different ways.”
Getting staff to embrace new automated technologies requires leadership and sensitivity, Rees said.
“There is magic in that, it’s a leadership role, and ensuring people don’t feel threatened,” he said. “It’s not about replacing their roles with bots, it’s making people more effective and collaborating with technology for better outcomes. That’s an important communication to make.
“This program relies on that view of AI making people more powerful and engaged. And you’re more likely to have success.”
Machine learning started as a proof of concept, which proved the Virgin analytics team could do things faster and better, as well as undertake tasks it’s not been able to do previously. From there, it’s turned into an integration exercise that Rees ultimately hoped will bring about a new era of “citizen data scientists” right across the organisation.
“This data transformation journey means everyone in the business is more aware of how we use data for better outcomes for members,” he said. “That gives us a bridge into creating more citizen data scientists across the business who are more comfortable with modelling and using data to make decisions. We’d like to make modelling as ubiquitous as Excel is.
“There is a saying that marketing is too important to be left to marketing departments. Well, data is way too important to be left to the analysts. It has to be prolific throughout the business and used to deliver better outcomes.”
Achievements to date
Machine learning for Virgin started in the redemption space and how to make more relevant redemption offers to members to get more value from the program. The second area was how to drive more powerful attribution models throughout the business, to see what does and doesn’t work.
“It’s about more targeted offers around encouraging people to redeem in areas that make more sense for them. That’s the key to program and our point of difference: Outstanding redemption opportunity for people. Therefore we apply analytics to be stronger in that area,” Rees said.
Data transformation is a journey without an end game, Rees added. “We’re setting our own path on this, and it’s about continuous improvement.”
To date, the Virgin analytics team has proven they can perform predictive modelling at 10 times the speed of processes before. The business has also seen an uplift in models of about 10 per cent.
“We’re 10 per cent better, seeing 10 times faster, and we’re doing stuff we could not do before,” Rees said. “In other words, the effectiveness of what we deliver is being maintained and improved and we’re doing it faster. We are able to genuinely to attribute results, a process which requires us to mine vast amounts of data to make out where the gold lies. With machine learning, we’ve been able to bring in a wider array of data in that modelling to find the gold nuggets.
“So there’s far more data, we’re working at higher volumes and we’re getting a deeper understanding of where impact is coming from.”