Rydely is a ride-hailing company that wants to predict rider retention. To explore the question, it has pulled a sample of the riders who signed up in January 2014. The data was pulled several months later. A rider counts as retained if they were active (took a trip) in the 30 days before the data was pulled.
The team wants to know which factors best predict retention, and how to put those insights to work.
The data
riders— one row per rider who signed up in January 2014.city: the city the rider signed up in (King’s Landing, Astapor or Winterfell).phone: the rider’s primary device.signup_date: date of account registration.last_trip_date: the last date the rider completed a trip.avg_dist: average distance in miles per trip taken in the first 30 days after sign-up.trips_in_first_30_days: number of trips the rider took in the first 30 days after sign-up.premium_user: TRUE if the rider took a premium-tier ride in their first 30 days.avg_rating_by_driver: the rider’s average rating from drivers, over all their trips.avg_rating_of_driver: the rider’s average rating of their drivers, over all their trips.surge_pct: percent of the rider’s trips taken with a surge multiplier above 1.avg_surge: the rider’s average surge multiplier over all their trips.weekday_pct: percent of the rider’s trips taken on a weekday.
Your task
- Define retention precisely. The pull date isn’t given: decide what date to assume, show how you inferred it from the data, and say how the retained share changes if you’re off by a few days. Describe the cleaning you needed (missing values, implausible values) and what fraction of riders were retained under your definition.
- Explore which factors relate to retention. Pay attention to when each field is measured: some cover only the first 30 days, others cover all of a rider’s trips up to the pull. Which fields could leak the outcome, and what does that mean for a model meant to predict retention early in a rider’s life?
- Build a model that predicts whether a rider will be retained. Explain why you chose your approach, what alternatives you considered, and what concerns you have about it. How valid is the model? Report the performance measures you think matter, and say how it would perform on a later sign-up cohort.
- What does the model tell you about what drives retention, and how much of that is correlation rather than cause? Separate findings that hold across cities and devices from those that don’t.
- How could Rydely use these insights to improve rider retention? Propose two or three concrete actions, say which riders each targets and when, and outline how you would test the most promising one before rolling it out.
What to deliver
A short deck (6–8 slides) for the Head of Rider Growth, with your findings and recommendations. Add an appendix for the data team covering your retention definition, cleaning decisions, model validation, and the queries or code behind each number.
All company names in this question are fictional.