Trainline's machine learning could save travellers £340 million this year

The SplitSave feature uses machine learning, with Trainline's massive data repositories, to efficiently calculate the best way to save customers money

There are many complaints that travellers like to repeat about the British rail system. It is overcrowded, infrequent, prone to breaking down and - the one on everyone's lips in January, when rail fares traditionally increase - expensive. Online ticketing platform Trainline can't do much about the crowds or the level of service, but it's using big data and machine learning to save passengers money.

Trainline CTO Mark Holt tells Computing's Evert Lombaert, "Split ticketing is a feature of the UK rail market. In some journeys…rather than buying a ticket from London all the way to Manchester, it is possible to buy a ticket from London to Rugby and then Rugby to Manchester. That - buying two tickets instead of one - is often significantly cheaper than buying the single journey all the way through."

SplitSave is Trainline's new feature that takes advantage of this, splitting a trip into multiple tickets in order to lower the cost of a journey. The company predicts that SplitSave will save UK travellers around £340 million in 2020, an average of £360 per traveller (this is a correction to the video, which states it will save £360 million and £270 per traveller).

Holt is keen to stress that there is no change to the journey for the passenger: splitting the ticket means staying on the same train, and even the same seat. The only difference is there are two tickets on the Trainline app instead of one.

"It brings together everything that we've learned about customer experience in rail and a lot of our data assets to be able to do this. So with what we know about the rail industry and the way that rail industry systems work, we were able to create a system that held the seat reservation all the way through on the same ticket."

Other ticketing providers may offer something similar, but Holt says that their approach uses "brute force" search. These send an "enormous amount" of requests to rail operators, eating up valuable resources. Trainline's machine learning algorithm only sends two.

"By putting all of that data together [about customer experience and Trainline's data assets], we were able to come up with a machine learning algorithm that was able to guess pretty accurately - super accurately, in fact - where the best place to split in any given journey was… Doing that has enabled us to offer it as a service to every single one of our customers, not just the select few that understand split tickets."

SplitSave processes 12 billion rows of data every day, before customers search for split tickets, to save time, says Sam Taylor, Trainline's Head of Data Science. The feature uses machine learning to calculate the best place to split a journey, based on a variety of factors such as time of day, available ticket types and what Trainline knows about when tickets tend to sell out.

"We've built a lot of time-saving and money-saving features - for example Price Prediction and BusyBot, which helps you decide where to get on the train. If you add up how much [money] this is saving customers and how much time, that's significant. It's amazing to think how many lives we're changing.

"That may sound a little bit corny, but being the scale Trainline are at, it is millions of customers, so it is significant."

Watch the full video interview with Mark Holt and Sam Taylor now.