Industry Focus
All Aboard the AI Train
How Trainline and Virgin Trains are Using Artificial Intelligence to Automate their Operations
AI in its many shapes and forms is disrupting almost every market. Some sectors are early adopters, but the UK railway industry – on the whole – has been slow to keep up. While driverless trains are yet to take over, train companies are already using AI behind the scenes to improve the rail experience. Rob Scammell speaks to Trainline and Virgin Trains to discover how
Driverless trains have been around since 1981, when Japan opened the Port Island Line. There are now over a hundred driverless rail systems with varying degrees of automation.
Clearly, this type of narrow AI is not groundbreaking. The UK with its fragmented rail system supporting many franchises has been slow to keep up in this area.
The first driverless train was trialled in March this year. It is unlikely that AI will replace tube drivers any time soon, either, with Mayor of London Sadiq Khan branding the idea “madness”.
But that hasn’t stopped UK train operators and ticket providers adopting AI in other areas to improve their business operations.
AI at Virgin Trains
The Richard Branson-owned company has been leveraging AI for its behind-the-scenes administration under the helm of Virgin Trains CIO John Sullivan. He believes that AI will be critical for all business areas – especially rail.
“From my perspective, rail is behind the times in terms of digital innovation,” he says. “One of the beauties of my job is the opportunity that we've got: all train operators have that.”
And while Virgin Trains doesn’t currently have any plans to implement AI solutions into the running of the trains themselves, Sullivan says that that there’s a “huge amount we can do using technology in a creative way – it’s unlimited”.
“Put the rule base in the computer once, and then you're off and it's done – it's a yes or a no.”
One of those ways has been to automate customer service tasks with the goal of freeing up staff to spend more time on vital operational work.
The UK’s fragmented rail franchises has resulted in a complex ticketing system – one that guards have historically been expected to know in and out.
"Honestly, you have to be a real specialist to know some of these things about how complicated it [the ticketing system] is,” explains Sullivan. “Put the rule base in the computer once, and then you're off and it's done – it's a yes or a no."
Customers are unlikely to be aware that AI is being used; they will just notice that they will receive a response much more quickly.
Another area where Virgin Trains is using AI is to automate customer service call centres.
"One of the challenges in the rail industry is that there could be disruption,” says Sullivan. “We had a fire outside Euston about four or five months ago, so these are going to create some complaints. And how we deal with those mass complaints – the customers won't see it in terms of the technology, but what they will see is that there's a much faster response to those."
Image recognition in train operations
Using cloud content management and the file sharing service Box’s intelligent recognition feature, Virgin Trains can identify and tag an image with the appropriate metadata.
That means an image sent by a customer of their ticket to Virgin Trains can be identified autonomously, saving a human behind a desk from needing to check its validity.
Sullivan says that although some of his colleagues were initially hesitant, the response has been “excellent” – and now they want more AI in their workflow.
And economies of scale mean that all of this technology is getting cheaper over time.
“The technology stack that we've got is cheaper than it used to be beforehand, which for me is an indication of how dated and stagnant the technology is in rail,” explains Sullivan.
“The technology stack that we've got is cheaper than it used to be beforehand, which for me is an indication of how dated and stagnant the technology is in rail.”
Sullivan says he has a list of tasks that can all be automated to some extent, such as portals allowing customers to check the status of a complaint, which could one day involve chatbots.
Away from the administration side of Virgin Trains, Sullivan says the company is also looking at plans to harness the power of AI in security footage to automatically detect a disruption on a train and send that footage to the relevant authority.
“How far we'll go down that route in terms of process automation, I'm not sure. But we've got lots to do,” he adds.
How Trainline is using AI for ticketing
Away from rolling stock, but still very much enshrined in rail, digital ticket platform Trainline’s site receives around 70 million visits a month, while the app handles six million searches per day. Formerly owned by the Virgin Group until it was sold for £168m in 2006, it’s very much a technology company – and AI is playing an increasingly crucial role in its operations.
The most obvious use is in crunching vast amounts of data and turning it into a useful service for its customers. One of those is its price prediction, which lets users know if a journey is likely to go up in price.
“It truly is big data. We use our data platform to put that into a format that we can use.”
“Per search, there's probably about five to ten journeys, and every journey has on average five to seven prices available to it,” explains Fergus Weldon, director of data science at Trainline.
“So that results in many, many billions of rows – it truly is big data. We use our data platform to put that into a format that we can use.
“And then we use Google's TensorFlow framework to apply some deep neural networks to that to predict when the prices are going to change per tier."
Disruption alerts powered by AI
In addition to price prediction, Trainline recently launched a disruption alerts feature that harnesses the power of AI to pull together live tweets from train operators’ Twitter accounts. The feature is built into the Trainline app, as well as Google Assistant.
This provides personalised, real-time updates on a customer’s journey, with natural language prompts such as “how’s my commute doing” or “is this train running on time”.
It works by classifying data from train operators' Twitter feeds into levels of importance, putting it into context and calculating which stations are affected by a disruption. From this the AI model can determine how each individual train will be impacted, creating a real-time map that can be accessed through Google’s AI assistant, as well as Siri’s shortcuts.
However, Weldon says that there are no plans to widen the net to include public domain tweets because of the difficulties of a machine understanding the nuances of natural language, such as sarcasm, as well as the lack of control over what people post to social media.
That being said, he is in favour of building a community of “super users” that would allow “super advocates for trains” to provide more information about their journey and any potential disruptions.
The future of Trainline AI
Looking ahead, Weldon is focused on making sure that the models they build stand the test of time and get “better naturally over time as the data grows”.
That applies to the company's intelligent seating tool BusyBot, price recommendations and disruptions on the voice app.
He says that they are also carrying out research to solve more rail user problems, and for Fergus that research incorporates his own experiences of delays and disruptions as a commuter.
“It's nice that we live and breathe it. So we kind of have this connection to the product.”
"I know what it's like for everybody on the network because I experience it, my team experiences it and lots of people at Trainline experience it.
“It's nice that we live and breathe it. So we kind of have this connection to the product.”
More specifically, Weldon says, the company is looking into alleviating the lack of information around delays and choosing a reliable train.
"It's much more train-orientated and trying to help people make a better decision about what trains they should take and what time of day to travel."