Identifying cab users' individual usage patterns for an effective promotion strategy

Project Details




Sheel Jaitley, Sunny Sapra, Amit Phatak, Biswajit Pattnaik, Sourav Paul





The report describes an analytics approach towards designing better promotions for increasing the revenues of The two key questions answered by this study are
1. When to launch a promotional campaign
2. Which customers to target to have the maximum ROI on the promotional campaign

The answer the above question two set of forecasting models have been built
1. A linear regression model to forecast weekly revenues.
2. A logistic regression model to forecast the chance of making a booking the next week.

The data set available does not contain revenue per transaction hence revenues are estimated basis the time of travel calculated by the journey start and end time. Due to unavailability of quality data we restrict our study to the time period July 2013 to November 2013. Also, for the purpose of the report we restrict our analysis for the point to point business model.
The data series used for developing the forecasting model are
1. From and to dates
2. Number of booking per user

Forecasting revenues: The plot of the daily revenues shows no trend but a daily seasonality in the data. Also since the data small we use regression model based forecasting. On analyzing multiple models the best predictors for revenue emerged to be the last seven days revenues. Using these predictors we developed a regression model for predicting the revenue. These revenue forecasts can be used as indicators for comparing the revenue targets and the forecasts. In case of a falling revenue forecast appropriate promotional campaigns can be launched to increase bookings.
Forecasting usage pattern: In order to identify the usage pattern for each customer two approaches have been tried.

1. Developing a logistic regression model for each user: All repeat customers can be segregated in the data and for each user a simple logistic regression model is developed taking the lags of number of bookings in the last week as the predictors. This model generates a binary output signifying whether the said user will book a cab the next week.

2. Developing a logistic regression model for all repeat customers: The simple model developed for each user can be extended by using interaction variables for each user to develop a common model that can be used for predicting the possibility of making a booking for each user.

Application Area: