eBay Case Study | Firebase

eBay Motors uses Firebase ML to quickly categorize images, reduce costs and improve user experience

Introduction

eBay Motors allows users to search and find cars for sale in their area. Users can also upload pictures of cars that they want to sell, using a simple and easy-to-use inteace through the mobile app. The app is targeted towards auto enthusiasts: customers who care about cars. These are folks that see a car as something much more than a utility that helps them get from point A to B.

Challenge

This type of buyer expects a more detailed description of the car and insights into its history. For sellers, it's of the utmost importance to have well-taken pictures uploaded for their car listings in the app. However, tagging each photo as an exterior, interior or engine requires a lot of manual effort, which can delay posting a car for sale, and ultimately prevent more people from listing their cars through the app.

The eBay Motors app development team spent a lot of cycles trying to figure out how best to automate the tagging of these photos, but still didn't have a solution they were happy with.

Solution

The team realized that they did not need to build all of this functionality in-house and turned to AutoML Vision Edge to help them with this user experience challenge. "There were two key considerations: how do we provide a seamless upload experience for sellers to reduce friction for them when creating listings, and how do we reduce the amount of cost, both in terms of engineering effort and server costs, to support this type of feature," said Jake Hall, Head of Native Apps for eBay Motors. AutoML addressed both issues head on, "allowing us to improve the time it takes to create a listing through the app, which obviously has clear implications for us in terms of seller ROI," added Hall.

Thanks to AutoML Vision Edge, the eBay Motors team was able to go from concept to prototype within a week using their own data set. With just a few hours of data labeling, they were also able to improve the model to a level that they feel comfoable using in production.

Results

1 week to go from concept to prototype (using their own data set)

20 hours of data labeling needed to use the model in production

"AutoML allowed us to improve the time it takes to create a listing through the app, which obviously has clear implications for us in terms of seller ROI."

- Jake Hall, Head of Native Apps, eBay Motors

eBay Motors uses Firebase ML to quickly categorize images, reduce costs and improve user experience

Introduction

Challenge

Solution

All Firebase products

Build

Run