Uber Eats

Data Science, 09/2021—11/2021, Berkeley, CA

In fall of 2021, I worked with a Voyager Consulting team to consult with Uber Eats, Uber’s food delivery platform. The company dealt with many types of users, from consistent daily diners to occasional late-night snackers. One of its new offerings, the Eats Pass, gives users access to discounts and special promotions. Uber Eats wanted to understand how to best market the pass to different groups of eaters and navigate the tradeoff between retention and conversion. The data science project sought to cluster eaters based on behaviors in order to design promos and optimize conversion and retention metrics.

I contributed to the data querying and clustering analysis. I wrote a batching script to split up a 30M+ row query for eater attributes (trips taken, average order size, etc.). I trained a random forest model to predict a user’s promotion status (whether they were in a trial period or paid subscriber) and benchmarked it against other classifiers: extreme gradient boosting, light gradient boosting, logistic regression, and decision trees.

We then used the random forest to conduct segmentation. I extracted a proximity matrix on a subset of users by computing the pair-wise co-occurance of users in the same terminal nodes of trees. We then used principal coordinates analysis to reduce the matrix to two dimensions and conduct K-means clustering. We used violin plots and conversion/retention curves to characterize the users of each cluster and recommend promotions.