Publicis Media Automates Audience Segmentation Using Machine Learning on AWS

Publicis Media creates value for its clients through global media agency brands and scaled capabilities across investment, strategy, insights and analytics, data and technology, commerce, performance marketing and content. The company leverages solutions from Amazon Web Services (AWS) to maximize that value for its clients. In an AWS on Air Voice of the Customer session from re:Invent 2020, Patrick Houlihan, SVP of Decisioning at Publicis Media, describes how the company’s data science team used AWS to build a machine learning pipeline that provides media buyers with highly-accurate recommendations on audience segments. 

Publicis Media built a machine learning pipeline that trains models with petabytes of audience datasets called the Decision Sciences Framework. Its solution uses Amazon Simple Storage Service (Amazon S3) for storage, Amazon EMR clusters for parallel processing, Amazon Redshift to produce custom attributes and segments, and SparkML for custom end analysis. The pipeline “acts as a general-purpose solver and decisioning layer,” says Houlihan, by scoring data such as profiles, audiences, data providers, keywords, ad impressions for propensity, persona extraction, and other characteristics. The solution is also designed to be portable—Publicis Media can spin up the Decision Sciences Framework inside its customers' AWS accounts to analyze data locally without moving it.

According to Houlihan, the tool processes petabytes of data easily using AWS. “That’s the beautiful thing about scalability of Amazon EMR,” says Houlihan. “For certain client engagements, the outcome might require a small cluster; another client project could be a very large cluster. The flexibility of AWS really makes it extremely scalable and versatile so that we can tackle as many different client datasets as possible.”

By building its framework on AWS, Publicis Media is able to focus its data science resources on developing proprietary algorithms rather than managing manual processes. “The value add at the end of the day is going to be what people are doing internally with their data science and proprietary algorithms," says Houlihan. "We lean on cloud providers such as AWS to get the latest processors and memory."

For data science that took at least a month with a substantial client team to do manually, Publicis Media can now service multiple clients in parallel and cut out many of the manual processes that were labor intensive in the past. Leveraging machine learning for media planning benefits Publicis Media's clients by surfacing what audience attributes will be the most impactful for ad performance. 

Learn more about AWS for Advertising at aws.amazon.com/advertising-marketing.