gui yong nian -

Case study: How Trainline keeps customers satisfied

The Trainline is encouraging customers to discover cheaper rail fares for the same journey with its AI-enabled SplitSave option

It is a fact of life living in the UK that rail prices increase at the start of January, and that the rail fare system is incredibly archaic.

In fact, it is even possible for some passengers to find convoluted journeys that get to the same destination, on the same train, by buying rail fare tickets that are significantly cheaper than direct fares.

The phenomenon is called split ticketing, and it is something that Trainline has turned its attention to. It claims split ticketing could save UK train travellers up to £340m in 2020, compared with the cost of direct train ticket searches made via the Trainline app in October and November 2019.

The company says it wants to make it easier for passengers to find the cheapest fare. “We’re always striving to make train travel easier and more seamless for our customers, so one of our design goals was to default to SplitSave when cheaper tickets are available,” says Mark Holt, chief technology officer at Trainline. “Split ticketing has existed in the industry forever. It is one of the idiosyncrasies of the UK.”

But finding legitimate ticketed journeys that could be purchased using more than one ticket has been problematic in the past.

“We needed to work with the rail industry to make this a success. What we don’t don’t want to end up with is creating a problem for customers, because a rail guard says they cannot travel. The nightmare scenario is if someone is standing on a platform, asking the rail guard about which platform to go to for the next part of the journey, and the guard points to the train that is just pulling out of the station.”

So for Holt and the team at Trainline, the SplitSave functionality needed provide a no-compromise customer experience. However, the software also needed to work in a way that avoided overloading the systems from the rail operators that provide the fare information that SplitSave then crunches to find the cheapest split ticketing option.

“Legacy rail ticket providers often run brute force searches to find split ticket fares, and this drive massive loads on the rail operators’ systems. We needed something different that could provide the most likely split ticket savings, without overloading the rail industry,” says Holt.

Rather than process massive database queries to collect all possible fare options across the rail network, Trainline’s SplitSave has used the power of passengers, collating data from customer queries. It works like a recommendation engine for split ticket rail fares, according to Holt.

“Machine learning is a big piece of the app. We have a vast quantity of customer data. Some customers search for a ticket, for example, from Euston to Manchester, some may search London to Rugby, while others are looking for trains that go from Rugby to Manchester. Our code looks through 12 billion rows of data a day to figure out where to get the biggest savings,” he says.

So instead of looking for 10 possible tickets that may get a customer London to Manchester, Holt says the application is able to filter this down to one or two split ticket options that it has strong confidence will be the cheapest split ticket travel choices for the customer. It is these options that can then be forwarded to the rail operator systems to obtain the actual ticket prices.

Holt says the recommendation engine inside SplitSave to support split ticketing is constantly evolving and learns what rail travel routes customers look up. By using this customer behaviour and caching ticketing pricing, Trainline needs to send less data to rail operator systems.

Our code looks through 12 billion rows of data a day to figure out where to get the biggest savings
Mark Holt, Trainline

The software was developed using a cross-functional team, focused on delivering a good customer experience. Discussing the team effort, Holt says: “We always start with a customer problem. It isn’t enough to say, ‘I have a bunch of data, what shall I do with it?’. It is about leveraging data to create a good customer experience.”

Along with machine learning and caching to limit the impact split ticketing queries have on rail operator systems, Holt says the development of SplitSave required a stream of user interface improvements, clever enhancements to ticketing and fulfilment systems, and a lot of work from search teams to avoid overloading industry systems.

“After months of iteration, the feedback from our test customers has been overwhelmingly positive. We’re really excited to give the whole nation easy access to the great savings it offers,” he adds.

The project does not have a true end of development phase, even though the project to develop the SplitSave functionality is now being rolled out. Machine learning models constantly evolve, as do customer expectations.

Holt says there is a need to keep working across the IT industry on long-term projects. “Keeping people motivated after a launch is hard. There are peaks and troughs,” he adds.

But Holt believes people can remain inspired to do more and continue innovating on the projects if they believe they are helping others. “We have so much innovation across our organisation. There is always something new going on. It’s about the mission, and people are involved in something they can constantly see customers using,” he says.

Read more about Trainline

Read more on Artificial intelligence, automation and robotics

Data Center
Data Management