LNER is on a mission to reduce “marketing effort waste” by exploring ways AI can help the business “be smarter” about targeting.
One way it has already done this is by building a machine learning model to improve its email marketing strategy by predicting and targeting customers when they are in the market to buy.
Speaking at the Data & Marketing Association’s (DMA) Customer Engagement Future Trends 2025 event today (27 February), LNER’s CRM lifecycle lead Lauren Hobson said the team wanted to move away from sending high-volume emails and be more “precise” in showing up to customers at the right point in their journey.
With customers travelling with LNER on average just twice a year, Hobson said the business had a “frequency challenge”.
Having the [in-house] skills to build these models faster, to be more agile and to iterate faster is great.
Omar Jouda, LNER
“When we couple that with the statistic that, on average, only 5% of customers were in-market to make a purchase at any one time, we resorted to high volume settings with some surface level personalisation to really drive our revenue targets,” she explained.
At the time, one of its key segments was also comprised of third-party data, which was “no longer driving value for the brand”.
To solve this, the marketing team approached LNER’s in-house machine learning team to understand how predictive machine learning could be used in segmentation.
Building a model
LNER set out to target customers ready to buy, using a machine learning model they built and trained to segment audiences by purchase likelihood – high, medium, or low propensity. This is updated daily depending on a customer’s intent to purchase.
LNER defined “in-market” as customers intent to buy within a 90-day window, which is around 95% of its customers.
Before building the model, the machine learning team took a “fail fast” approach, working in a six-week cycle to refine their strategy. They defined key business challenges, established clear success metrics for the proof of concept (POC), analysed available data and tested different machine learning models.
“Our goal was to make sure that machine learning is the way to go,” explained LNER’s machine learning product manager, Omar Jouda. “A machine learning model is only as good as the data that goes into it, which is why we spent some time finding the right data.”
The team analysed journey data (booking frequency, ticket types, in-train purchases), transaction data (spending habits, average purchase value), and engagement data (customer interactions with marketing campaigns).
Understanding the results
Once the model was ready, the marketing team created two email variants instead of its usual monthly “revenue driving” email campaign.
One featured information about its first-class tickets and was sent to the high-propensity group, with the intention of driving up their average order value. The second was sent to the medium-propensity group, who were on “the fence”.
“We sent them something a bit more inspirational, some ways to save and reasons to travel by train, and then we fully excluded the lower propensity group, because these were our cold audience,” explained Hobson.
Hobson claimed that the business achieved a 78% prediction accuracy compared to its current CRM system offers, which deliver a 66% accuracy. The emails sent to its segmented audience also achieved higher engagement and conversion.
The high group had a 20% conversion compared to the medium group, which had a 4% conversion. Meanwhile, by choosing to exclude the low propensity group, LNER predicted it would send 300,000 fewer emails per month, which over a 12-month period could make carbon savings by sending millions of fewer emails – the equivalent of a car driving 5,000 miles.
The pivot to using machine learning in its CRM strategy is part of LNER’s “decades long conquest” of moving from volume based email marketing into more predictive targeting.
Hobson noted the brand is still on the testing journey but has since identified four additional applications for the model.
“The ultimate goal is to have something which is automated, so it will trigger at the point of a customer becoming in market in the medium decile,” she says.
Meanwhile, Jouda added that having an in-house AI team makes a “huge difference”.
“Having the [in-house] skills to build these models faster, to be more agile and to iterate faster is great.”
