MediaMath MediaMath is on a mission to build a more accountable, addressable supply chain that drives better outcomes for marketers and publishers while delivering better experiences for consumers. As the leading independent programmatic company, MediaMath has steered the evolution of the Demand-Side Platform (DSP) sector since 2007. Their business has become a globally scaled, enterprise-grade DSP and Data Management Platform (DMP) offering that delivers personalized digital advertising across all addressable touchpoints. With advertisers in 42 countries around the world, MediaMath does business across all verticals, including retail, CPG, travel and finance. Notable customers include IBM and Uber, and a range of marquee clients who want to elevate their marketing by innovating at scale with new ideas, tools and trends.
MediaMath helps advertisers optimize decisioning and determine how to best spend their media budget across display, mobile, video, audio, digital out of home and connected TV. They then help clients analyze the data from their campaigns and find the best ways to improve performance. To remain a leader in the media buying field, MediaMath needed to provide customers and its own analytics team with greater flexibility to create any kind of analysis they want, without limitations. And they needed to be able to run those queries on billions of records at once.
To address this billion-record data discovery challenge and gain the flexibility they needed, MediaMath turned to Qubole in 2013. The Qubole Platform supercharges productivity through intelligent automation–enabling a 1:100+ admin-to-user ratio—and delivers faster cycle times with a self-service infrastructure for all users. There is zero waste, as workload-aware autoscaling and aggressive downscaling optimize compute usage. Plus, customers are always able to find the best buy via Amazon EC2 Spot shopping (for lower-cost compute instances) to drive down compute cost even further.
“A full year of data for our larger clients is in the billions of rows of records,” says Mike Rancourt, Senior Product Manager of DMP at MediaMath. “We have three major tables which hold differing but all-important data sets that users compare in their analyses. For a large client, some of those tables may contain two to three billion records. It’s a massive amount of data.”
There was obvious demand for answers to questions that the data would offer, and Qubole provided us a tremendous amount of flexibility in seeking those answers.
John Slocum, VP, Data Management Platform, MediaMath
MediaMath’s first use of Qubole was internal. Their analytics team used it to understand their clients’ audiences and the impact MediaMath’s platform was having on client campaign performance. But soon they began using Qubole to power the self-service analytics products they were making available to clients. Today, MediaMath continues to expand its use of Qubole, putting its power, flexibility and ease of use at the disposal of its customers.
One of the ways in which clients can optimize their media budget more intelligently using Qubole is via a data product called Enriched Identity, which extends and augments clients’ first-party audiences. This lets advertisers enrich their user data and model new audiences before pushing them to the MediaMath DSP for targeting.
MediaMath developed its own DMP within its existing DSP, so the integration is truly seamless, enabling zero data loss and latency for advertisers interested in performance. “With Enriched Identity and Qubole supporting DMP analytics, we can model custom audiences to target on and off of our TerminalOne DSP, driving media performance in any relevant channel,” Rancourt says.
Another MediaMath offering that helps clients optimize their spend is Back Test Analysis. This supports performant audience discovery without actually targeting or testing specific segments. It takes data clients have served previously or are currently serving, finds the conversion activity of those impressions, overlaps it with MediaMath segmentation taxonomy and determines which segments those converters most likely fall under. It finds actual CPAs—cost per acquisition—of the users in that segment. Qubole enables this through highly efficient simulation and scenario testing, which saves not only costs but time.
“With Qubole, Back Test Analysis allows us to overlap campaign and audience segment data to calculate performance by segment in sample populations. This highlights the segments advertisers should (and should not) be targeting… without actually targeting the segments being tested,” Rancourt says. “So, advertisers can very easily determine which segments they should target without ever having to spend precious media dollars to test them.”
To use MediaMath’s Enriched Identity, Back Test Analysis and other products that let them buy media more intelligently, clients need a target audience. Ideally, they’d like to have a list of prospects with traits that indicate a high likelihood of conversion. That’s where MediaMath’s Audience Creation product comes in.
With Audience Creation, users don’t have to determine the traits of a high-propensity audience through testing. Instead, their analysts can use Qubole to query their audience data and find an audience they know has a high propensity to convert. Then, through MediaMath’s Audience Creation template, they can send the output of that analysis—a list of MediaMath user IDs—directly to T1 for targeting.
“Not only is this a time-saver but, more importantly, it’s more accurate. It’s more effective targeting,” Rancourt says. “You’re making inferences and using those inferences to tweak your target in T1. You’re able to target the exact audience you want.”
MediaMath offers clients a choice of how they can work with their data. They have the option of integrating their data with their own infrastructure via API if that’s what they require, or they can query it through Qubole. “We make clients aware of this API any time we talk to them. But more often than not, they choose to use Qubole,” says John Slocum, MediaMath’s VP of DMP. “Most prefer the packaging, ease of use and support they’ll have in Qubole, rather than supporting a homegrown solution they need to manage and maintain.”
We see ease of use—ease of access to data—as being one of the drivers to success with this data platform.
John Slocum, VP, Data Management Platform, MediaMath
Slocum and Rancourt say they are continually improving upon and expanding the ways in which they use Qubole.
“A few years back, the only people utilizing Qubole for us were data scientists and heavy analysts,” Rancourt says. “As we evolved, more and more users and different groups started using it. We see a good opportunity out in the marketplace, as well as internally, to utilize more of Qubole’s tools, like the dashboard and notebook functionalities. Anything that would allow us to put this tool into the hands of a larger set of users—not just our heavy technical users—is something that will be very important for MediaMath over the next couple of years.”
Breadth and Size of the Data Set that can be Queried
Flexibility to Query Data in Whatever Way the User Chooses
Powers Many of MediaMath’s Self-Service Analytics Products
Helps MediaMath Win New Business
Free access to Qubole for 30 days to build data pipelines, bring machine learning to production, and analyze any data type from any data source.