2. Who am I?
Ankit Jain | Data Scientist
Finance Data Science
Primarily work on trip forecasting at Uber
Previously
Data science at BofA, Facebook, and a few VC
backed startups
3. Why I’m here
● Discuss the applications of AI in logistics & forecasting
● Data and signals are our eyes and ears into the world (past, present, and
future)
● But extracting information from them to drive business efficiency is not
straightforward
4. “AI is going to affect most aspects of our lives. It's
similar in impact to other revolutions that have
occurred in human history, like the agricultural
revolution, the industrial revolution and the
computer revolution.”
-- Zoubin Ghahramani, Chief Scientist, Uber
5. There are mainly two types of Logistics businesses:
● On Demand
● Scheduled
Types of Logistics Business
6. Objective
A key aim of any logistic business is :
● Minimize Estimated Time of Arrival (ETA)
● Improve Reliability
At minimal costs AI
7. Data
● Location Data
● Anonymous User (Driver/Rider) Activity data on the platform
Huge amounts of data is processed everyday.
10. Forecasting- Problem Types
Long-Term Forecasting
Uber uses forecasting to make optimal business decisions. Long term forecasting
is useful for financial planning and onboarding new drivers
Short-Term Forecasting
Logistic businesses care about forecasting at a far more granular level though.
Knowing supply/demand imbalances before they happen ensures marketplace
stays efficient
Real-time Forecasting (anomaly detection)
Having minute-by-minute forecasts of all major metrics allows us to immediately
detect outages and issues
11. Pricing
● AI enables smarter and more efficient pricing for any
logistic business by anticipating trends in the market
and adjusting accordingly.
● Few input features to the ML algorithm:
○ Origin and destination of the trip
○ Day of the week and hour of the day
○ Weather, holidays etc.
○ Type of vehicle
○ Current/expected supply and demand gap
12. Dispatch and ETA
● ETA: Some factors which affect ETA are:
○ Current Traffic Conditions
○ Origin and Destination locations
○ Past traffic conditions of same hour and day of the
week data
○ Weather
● Dispatch is essentially efficient matching of drivers and
riders. It is based on a combination of factors including
○ ETA, Driver/Rider ratings etc.
○ For carpool, it is based on % route overlap for trips
13. Food Delivery
● Restaurant/Dish Recommendation
○ Past user history
○ Popularity
○ Offers
● Estimated Time to Delivery
○ Pick-up Time
○ Food Preparation Time
○ Delivery Time
14. Self-Driving Cars
● One of the most challenging problems
● Lots of data recorded by sensors
● Extremely high accuracy requirement
● System should be able to handle
uncertainty
● Three major use cases of AI:
○ Perception (Computer Vision)
○ Prediction
○ Motion Planning
15. And...
● Customer Support
○ Help customer support agents respond to tickets
● Destination Prediction
○ Predict the destination of riders based on current location and time of the
day
● Fraud
○ Payment Fraud
○ Collusion Fraud (Driver/Rider, Driver/Restaurant)
16. “The future is not some place we
are going, but one we are creating.”
-- John H. Schaar