How Trainline is looking to predict your journeys before you plan them

Popular rail ticket booking app Trainline is starting to leverage its new cloud infrastructure and huge historic data set to deliver customers more personalised offers, including a new predictive pricing tool.

Since being acquired in January 2015 by private equity firm KKR, Trainline has been busy shifting to the cloud and building out its engineering and data science teams, to offer the kind of personalised shopping experience consumers expect in the age of Amazon and Expedia.

Speaking to Computerworld UK from their newly refurbished office in Holborn, London, Trainline's CTO Mark Holt explained how the company has finally admitted to being an ecommerce company, embracing the cloud and a DevOps culture to deliver regular changes to its user experience.

Predictive pricing

Armed with 15 years of search history and price data, the data scientists and data engineers at Trainline have been busy coming up with these smart features for its core consumer app.

"We have a massive amount of stuff in our labs, particularly around data innovation to create more customised, personalised experiences," Holt said. "We have both data scientists and data engineers that work in collaboration to make data available and then turn it into data product. So it is all very well coming up with an algorithm, but making it robust and reliable is a big part of that."

The aim of these projects is to increase a "mixture of conversion and attention", according to Holt, as the Trainline app acts as a popular information resource for hundreds of thousands of commuters every day, as well as a booking platform.

Read next: How was built on machine learning

For example, a predictive pricing tool is due to be rolled out to UK app users this month. The tool allows travellers to look into the future and see when advance ticket prices are set to increase over time. This makes it easier to compare ticket prices across a range of dates, like Skyscanner or Expedia does with flights.

Jon Moore, chief product officer at Trainline said as part of the official press release: “Our data scientists have used historical pricing trends from billions of customer journey searches to predict when the price of an advance ticket will expire. We now share this information in our app to allow our customers to get the best price possible for their journey.

"We’re introducing more advanced machine learning every day so naturally our predictions will get increasingly accurate. Our mission is to make train travel as simple as possible and price prediction is the first in a long line of predictive features we have planned to help customers save time and money.”

Holt says that internally, Trainline believes it can predict where you want to travel to a high degree of accuracy. "In more than 80 percent of cases we know where their destination is going to be, based on all sorts of dimensions, like day of the week, device, time of day," he said. "In 30 percent of cases we know which day they want to travel, based on booking history.

"Then you start to think about leveraging that to create great customer experience, that predictive capability of saying: 'we think there is a good price available today to go to Manchester next Friday night', for example."

Trainline says this tool could save customers an average of 49 percent (up from 43 percent before price prediction) on advance tickets.

For example: a standard class advance single fare from London Euston to Manchester Piccadilly route is £32 when booked 80 days before the day of travel, rising to £38 at 41 days before the day of travel, £87 at 2 days before and £126 on the day of travel itself.

Read next: How HackTrain is helping startups to solve the UK rail industry's biggest problems, from overcrowding to ticketing and delays

Cloud migration

All of this work has been enabled by a major change in technology and culture behind the scenes at Trainline over the past few years.

The first stage on this journey was to modernise the Trainline's underlying infrastructure, migrating from a private data centre in Rotherham into the public cloud with AWS. The 14-month migration was completed 18 months ago and has led to annual savings of £1.2 million a year in capital expenditure (capex), while operational expenditure has remained flat, according to Holt.

Trainline is now a headline AWS customer, with Holt's colleague Chris Turvil, head of cloud and platform agility, presenting on stage at the cloud giant's annual re:invent conference in 2016. So it's not surprise that his logic for choosing Amazon over its rival vendors comes down to them being "more awesome", as he told Computerworld UK.

By migrating to the cloud, and breaking up its monolithic application into smaller bits, the engineers at Trainline can focus on continuous delivery of incremental upgrades to the user experience, rather than massive updates.

"It used to be a six weekly release that landed with a thud and everyone would wince, so change was something to be afraid of," Holt explained. "If you go back to our experience three years ago, it is very different, but we have got there with a whole load of tiny increments.

"We test everything, so we do a lot of multi variant testing to ensure customers are using and liking the functionality we build. People are able to throw something out there, run a test and go from there."

Read next: Cloud and microservices help Expedia innovate quickly

Now Trainline developers roll out up to 200 changes a week across its apps, and can start to leverage the features coming out of the "innovation machine" of AWS, as Holt refers to the cloud vendor.

Holt says that Trainline now uses the whole AWS "toolbox", with data sitting in S3 buckets, moved around using Kinesis Firehose, and running code on Lambda to enable serverless delivery.

"In that world you are free as a developer and as part of the product team to move the experience on by tiny increments but really fast," Holt said.

Copyright © 2017 IDG Communications, Inc.

8 simple ways to clean data with Excel