Doordash and the Food Delivery Industry
DoorDash is a triple-sided marketplace and logistics platform that enables consumers to order food on-demand from restaurant merchants that are then delivered by a fleet of delivery people, called “Dashers”. Consumers are Doordash’s B2C customers, merchants are Doordash’s B2B customers, and Dashers are contractors employed by Doordash who deliver meals from merchants to consumers in their own time.
The food delivery space is highly competitive. Many large players, such as GrubHub, UberEats, Yelp’s Eat24 and DoorDash, vie for dominance, while the threat of new entrants from companies like Google and Amazon looms large. To stay competitive in the food delivery space, DoorDash needs to be better than its competitors at delivering against what the merchants and consumers care about. This includes selecting and surfacing the right restaurants to consumers, minimizing delivery times and making sure that the supply of Dashers meets the consumer demand for food deliveries at every given point in time. Machine learning is a critical tool that enables DoorDash to deliver against all of these key objectives.
Present Uses of Machine Learning
DoorDash uses machine learning to present a personalized selection of recommended restaurants for users based on their past search and order history. Like traditional recommender systems (e.g. Amazon’s product recommendations), DoorDash has to handle cases where new stores or consumers enter the system and manage the tradeoff between relevance versus diversity. Unlike traditional recommender systems, DoorDash’s personalization mechanism faces the issue of sparsity: not every consumer can see every store. By showing users restaurants they are likelier to order from, DoorDash has lifted its conversion rate from search to checkout by 25%.
Unsurprisingly, consumers care about both getting their food quickly and getting a reliable estimate of their delivery time. Predicting the delivery time enables DoorDash to assign their Dashers optimal routes and orders, which drives efficiency as measured by delivery time per order and Dasher utilization (the number of deliveries a Dasher performs per unit of time). Variability complicates the problem: DoorDash needs to take into account real-world conditions such as traffic and weather; Dasher supply and delivery demand are both constantly in flux; restaurants also vary in the time it takes them to prepare an order depending on how busy they are. The supply of Dashers available to make deliveries at a given point in time also affects delivery times. To ensure that the supply of Dashers meets consumers’ demand for deliveries at any given point in time, DoorDash uses machine learning to forecast delivery demand.
In the medium term, DoorDash is investing in autonomous robots to deliver food to customers. These robots are eventually meant to complement, rather than replace, the activity of human Dashers. While robots perform actions primarily based on reinforcement learning algorithms (separate from classic machine learning), machine learning is still a critical component of robots. Similar to autonomous vehicles, these robots run on LIDAR technology. LIDAR takes raw input from the robot’s camera and sensors and uses machine learning to recognize and classify objects in the robot’s surroundings. Having a fleet of autonomous delivery vehicles at hand serves as a great backup for when the supply of Dashers fails to meet consumer demand, which reduces the variability caused by inconsistent Dasher supply and dramatically cuts delivery times.
One short-term application of machine learning that DoorDash should do is empowering their local merchants with insights from machine learning to fuel operational efficiency. DoorDash can, for example, use machine learning to help merchants forecast demand, price more optimally, increase labor utilization, reduce food waste by optimizing inventory levels and select the right menu items to offer diners.
DoorDash can also use machine learning to prioritize the restaurants they choose to partner with. Consumers want to find their favorite restaurants on DoorDash. With machine learning, DoorDash can predict the types of restaurants attributes that will be popular with users and rank these restaurants accordingly.
In the medium term, DoorDash should continue developing their delivery robots to make them ready for mass adoption: the robots need to be able to reliably navigate themselves to the proper destinations, interface with local merchants (who are often not very tech-savvy) and recognize the correct orders. Moreover, DoorDash’s ability to grow its user base is currently limited by its Dasher capacity: if DoorDash grows too quickly, it runs the risk of failing to fulfill every order. Autonomous robots solve this problem and allow DoorDash to aggressively grow its user base.
One of the biggest open questions for DoorDash is an ethical one: the rise of delivery robots could eliminate the need for human Dashers, which runs counter to DoorDash’s mission of economically empowering the masses. DoorDash, then, will need to think about how exactly these robots should complement, rather than replace the work that Dashers do
 DoorDash. “Powering Search & Recommendations at DoorDash – DoorDash.” DoorDash, DoorDash, 6 July 2017, blog.doordash.com/powering-search-recommendations-at-doordash-8310c5cfd88c.
 DoorDash. “Welcoming Our Newest Robots to the DoorDash Fleet with Marble.” DoorDash, DoorDash, 16 Aug. 2017, blog.doordash.com/welcoming-our-newest-robots-to-the-doordash-fleet-with-marble-e752a85d6602.
 Hirschberg, Carsten, et al. “The Changing Market for Food Delivery.” McKinsey & Company, www.mckinsey.com/industries/high-tech/our-insights/the-changing-market-for-food-delivery.
 Marshall, Matt. “DoorDash Sees 25% Lift from AI Recommendations.” VentureBeat, VentureBeat, 18 May 2017, venturebeat.com/2017/05/17/doordash-sees-25-lift-from-ai-recommendations/.