Compared with other machine learning problems, our biggest challenge is lacking ground truth data, which is pretty common in the online-to-offline (O2O) business model. However, challenges arise along the way with its core development. There’s no other way to ensure accuracy without utilizing machine learning technologies. Time predictions through an order lifecycle Therefore, time prediction is constantly playing a critical role across the order lifecycle, which is shown in the figure below. On the other hand, arriving late will lead to unfavorable cooling of food. By arriving too early, they would take up the restaurants’ parking and dine-in spaces.
Ideally, they should arrive at the restaurant the moment food is ready. Next, we have to compute the perfect timing to send out delivery partners to pick up food. First, we need to set the right expectations by providing precise delivery time estimations in order to avoid frustrations in the case of delays. With the mission "Make eating well effortless, every day, for everyone" one of our top priorities is ensuring reliability. My recent talk covered how Uber Eats has leveraged machine learning to address these challenges. Additionally, time predictions are important on the supply side as we calculate the time to dispatch delivery partners. The ability to accurately predict delivery times is paramount to customer satisfaction and retention. Currently, it’s available in over 600 cities worldwide, serving more than 220,000 restaurant partners and has reached 8 billion gross bookings in 2018. Uber Eats has been one of the fastest-growing food delivery services since the initial launch in Toronto in December 2015. The estimated time-of-delivery prediction model is designed to be flexible enough to handle various scenarios due to its uniqueness of newly surfaced information in different stages.Our biggest challenge of lacking ground truth in O2O business model was tackled by inferring the label data via feature engineering work and leveraging the feedback loop for model retraining.Uber's in-house machine learning platform, Michelangelo, has provided tremendous help in simplifying the overall process for data scientists and engineers to solve machine learning problems.Uber Eats's dispatch system launched with a greedy matching algorithm, and by using a global matching algorithm powered by time predictions was able to be more efficient.We need delivery partners to arrive at restaurants the moment food is ready, in which time prediction is constantly playing a critical role across the order lifecycle. With the mission "Make eating well effortless, every day, for everyone", one of our top priorities in Uber Eats is ensuring reliability.