DoorDash Technical Meetup Event: Exploring ML use cases that power DoorDash

This event is an opportunity to look under the hood and learn more about the models that power DoorDash's platform and up level our business. In this event we will be discussing our forecasting engine and techniques, new capabilities like image recognition and how we enabled quality merchant selection using machine learning.

Agenda:

12:00pm - 12:05pm: Welcome and Introductions

12:05pm-12:20pm - Keynote talk: Turbocharging Operations with Scalable Forecasting by Ryan Schork

Lightning Talks 12:20pm-12:45pm

• How DoorDash uses causal inference to improve forecasting accuracy by Chad Akkoyun
• The ML models that power our merchant selection by Lu Wang
• How DoorDash-quickly spins up multiple image recognition use cases by Chi Zhang

12:45 pm- 1:00 pm Q&A

Titles and abstracts:

Turbocharging Operations with Scalable Forecasting by Ryan Schork

Description: Forecasting is an essential part of operations but notoriously difficult to model. For applications like demand planning, there are constant fluctuations due to promotions, weather, outages, COVID, etc. that must be corrected for or modeled into the series to project accurate seasonal trends moving forward. At an hourly or daily level, these series lack the scale for big data techniques to be maximally effective. Furthermore, it is difficult to know future values of variables like weather - so you need to rely on forecasts or averages which add more uncertainty to the process. At DoorDash, we developed a novel platform based on our proprietary forecast-factory framework to not only systematically address these challenges but allow operators to seamlessly supplement the algorithms with their expertise via human-in-the-loop feedback ingestion.

How DoorDash uses causal inference to improve forecasting accuracy by Chad Akkoyun

Description: When generating forecasts at DoorDash, we leverage time-series methods to get the correct trend and seasonality abstracted from the temporary fluctuations affecting target metrics. The common approach to account for temporary fluctuations is to define features (e.g., precipitation for weather, one-hot encoding holidays) and incorporate those features in the model. However, some temporary sources of fluctuations (e.g., tax refunds, child tax credits) may not have a direct measure/feature, which makes them hard to detect and account for in our forecast. For these types of factors, we leverage causal methods such as difference-in-differences and preprocess the demand series so that we can generate accurate forecasts.

The ML models that power our merchant selection by Lu Wang

Description: An optimized merchant selection strategy has been one of the key factors that has enabled DoorDash to become an industry leader in US food delivery service. DoorDash was founded later than many companies in this industry, but we have effectively onboarded high-value merchants to ensure the selection in every market matches customer demand. We were able to do this quickly by developing machine learning models that identify which merchant characteristics do well on the platform while ensuring every market’s selection addresses customer food preferences and popular local merchants. These models provide intelligence to our merchant sales teams so that they can effectively find and onboard merchants that will delight customers and expand the total addressable market (TAM) for DoorDash by adding increased selection to the platform.

How DoorDash-quickly spins up multiple image recognition use cases by Chi Zhang

Description: DoorDash has rich image data collected by Dashers, our delivery drivers, that we use in a number of use cases. We can use this wealth of data for a number of things, including to check whether pizza bags are properly equipped, stores are closed, or catering has been set up, among many other possibilities. With the large volume of image data that comes in daily, it’s impossible for humans to examine images manually and confirm the information associated with them. It is therefore critical to create an automated, responsive, and reusable solution to extract important information from the images. In the past, DoorDash relied on third-party vendors to analyze incoming image data in a slow, expensive, and largely unscalable process. Our current image recognition solution described here enables us to spin up new use case models quickly, efficiently, inexpensively, and at scale.