Browsed by
Month: June 2018

Optimizing Ad Serving with Machine Learning for cost, performance and quality improvements

Optimizing Ad Serving with Machine Learning for cost, performance and quality improvements

About our client

Their Ad-serving technology is optimized for user experience, and is integrated in over 450 mobile and web applications. They serve more than 1 billion ad requests every day to millions of users with a minimal latency. They embraced the Cloud-First spirit by being 100% hosted on AWS from the beginning and they are using over 50 services to create a cutting-edge technology solution.

The challenge

Nowadays, Ad Tech companies have to handle huge traffic volumes to deliver Ads. They focus on high quality Ads by controlling traffic and bidding on partner programmatic platforms. This strategy reduces the payout per Ad request and therefore limits the revenues of the company. Learning user behaviors and optimizing traffic quickly became inevitable, so they gathered a R&D team to build Machine Learning algorithms.

The solution

User data is processed using an Amazon EMR cluster, running every month and processing tens of billions of user events. Amazon S3 is the primary choice for storing raw data (hundreds of terabytes). An Amazon EC2 instance, packaged with Machine Learning libraries, uses the monthly generated dataset to update the currently trained model and prepare a blacklist, that is loaded in Amazon DynamoDB for high traffic reads.

Machine Learning

The benefits

  • The ML algorithm detects and filters 60% of total traffic being not relevant. The R&D team continuously improves its accuracy.
  • AWS infrastructure costs are reduced by 45%. The company is reinvesting this money into new services and quality improvements.
  • Ads get rendered 7x more often, meaning users have a better experience while watching ads in mobile applications.
  • Higher Ad partner trust and Ad revenues. A direct impact is a visible growth in partner integrations and market visibility.