TechPrufer
Get started
Grab
Grab logoGrab · Feature trail

Demand Forecasting and Peak Shifting

1 post · Feb 2019 – Mar 2019

What this trail covers

Grab has developed a 'Travel Trends Widget' that leverages historical supply and demand data and machine learning to forecast pricing trends for the next two hours. This aims to help passengers plan their trips to avoid peak demand periods, thereby 'peak-shifting' demand. The widget displays aggregated historical surge data in an interpretable visual format, providing tips and information to encourage users to adjust their travel times. The development process was agile and collaborative, involving the Analytics team and product managers.

All Grab trails
  1. Recipe for Building a Widget: How We Helped to “Peak-Shift” Demand by Helping Passengers Understand Travel TrendsThis post details the development of the 'Travel Trends Widget' which aims to 'peak-shift' demand by helping passengers understand travel trends. The widget uses historical supply and demand data and machine learning to forecast pricing trends for the next two hours. The development process was agile and collaborative, involving multiple teams. The widget pulls together historical imbalances between supply and demand for the user's current location and nearby time periods, displaying aggregated data visually. The design team drew inspiration from Google's Popular Times and focused on an intuitive tapping gesture for interaction. The data science team worked on deducing historical surge trends and ensuring reliable data storage and access. The widget was developed and released within two weeks, with ongoing iterations based on user feedback.
  2. Understanding Supply & Demand in Ride-hailing Through the Lens of DataThis post details the analytics initiative to measure the supply and demand ratio in ride-hailing at any given area and time. It defines supply as an online, idle driver and demand as a passenger checking fares. It explains how supply and demand are mapped using geohashes and inversely weighted by distance, and how 'effective supply' is calculated. The post also describes the aggregation of these metrics into 'Supply Demand Ratio' and 'Supply Demand Difference'. It highlights the computational challenges of processing this data at scale and introduces the 'Travel Trends Widget' on the passenger app to inform users about predicted demand distribution, encouraging time-insensitive demand to book later. It also mentions the updated 'Heatmap' on the driver app to guide drivers to oversupplied areas.