Developing an AI model to prevent customer churn within the public transportation industry
Imagine having the tools to accurately predict your customers’ behavior, and know which actions will sway them from discontinuing your products or services. This European public transportation operator wanted to know exactly that and was able to gain real insights into customer behavior through AI.
Churn prediction for a public transport operator
Growing the number of users is a fundamental goal for public transportation companies. Analyzing “customer churn” – or the number of customers that deflect a service – has served as a tactic that supports transportation organizations in better understanding customer behavior and helps reach their desired growth. Preventing customer churn is necessary to remain competitive and cost-efficient. According to Forrester, acquiring a new customer can cost up to 5 times more than it does to retain an existing one.
This use case tells the story of a European public transportation operator, and how they took steps toward overcoming their challenges in predicting customer churn through implementing AI into their organization. The company is one of the largest public transport operators in Scandinavia, operating several modes of passenger transportation.
For many years, the organization had been running its operations, offering services to passengers without using a data-driven approach to report on customer behavior. Without these data insights, the organization lacked critical information about its customers’ wants and needs, and most importantly, insights into which customers are most likely to churn and their reasons for churning.
One of the main challenges, in this case, was to discover and outline information known about the operator’s customers. We worked together with the management consultant company, Valcon, to figure out which variables typically influence transportation operators, and created a hypothesis that could be tested and investigated through AI. We focused specifically on the commuter customer segment, which was previously thought of as an unconditionally loyal segment. The challenge was to extract information about these commuters from the operator, and with this information, create useful insights for new initiatives to prevent churn.
A churn model was developed for the operator that provided insights as to which customers were likely to churn, and why. Customers were grouped together by churn probability to identify the key drivers, using the average drivers across a customer segment to explain global trends that cause a particular segment to churn.
The model was aimed at the commuter segment, which is a valuable segment for the operator. The model leveraged historical commuter information, such as the type of commuter card, seniority, means of purchase, and so on.
When an AI model is fed this information, it accurately predicts if a commuter is likely to churn or not, and presents the prediction as a probability between 0 and 100 percent. The prediction became the likelihood the customer had of repurchasing for another period, based on the length of the previous period he or she had purchased.
The overwhelming conclusion was that customers that bought a commuter card for more than one month at a time were less likely to churn. The model also revealed that customers who had been with the company for more than three months on average were more likely to stay. This data suggested to the operator that there could be an onboarding period.
In the future, the goal is for the model to further process historical time-series data, such as weather trends, carried out campaigns, planned track work, and car sales to yield a commuter’s probability of churning the following month.
The overwhelming conclusion was that older customers who bought multiple services were less likely to churn. In comparison, younger customers with no other link to the company, rather than the targeted service, were more likely to churn. The model also revealed that customers who had been with the company for over a year were more likely to stay. This data suggested to the operator that there could be an onboarding period.
Three key takeaways from this project
→ By applying AI, the organization was provided with data-driven customer insights.
→ By focusing on the most valuable segments of the commuter group for the operator, the AI model yielded information on what factors impacted churn most.
→ In the future, the operator can take strategic and commercial actions based on data-driven insights provided by 2021.AI.
As a result, the operator can use the predictive probabilities to prevent customer churn. Based on specific insights into the factors that influence churn, the operator can take strategic and commercial actions for the future based on insights from previous customer behavior, as opposed to driving change through best guesses. Provided with this insight, the operator is equipped with a fantastic tool to improve overall customer experience, lower the churn rate, and thereby optimize cost, and grow revenue.
In this case, the operator can get to know their most loyal customer segment, and be proactive in improving their services, ensuring customer satisfaction and longevity.