Published on 24-Jan-2012
Validated on 07 Oct 2013
"Terabyte-scale data analysis is the new weapon for competitive advantage on Madison Avenue, and the IBM PureData System for Analytics provides MediaMath with the infrastructure to optimize and serve billions of daily impressions — establishing a technical foundation for their long term success in this industry." - Brad Terrell, IBM Vice President and General Manager, Digital Media
Media & Entertainment
PureData System for Analytics (powered by Netezza technology), PureSystems, Big Data, Data Warehouse, Smarter Computing
MediaMath, a New York City startup with an idea to help ad buyers capitalize on the rapidly evolving marketplace by providing services and tools for ad agencies — the most active “demand side” players — to identify, bid on and buy just those impressions most likely to yield the results they sought for their clients. Implementing the idea in 2007, MediaMath created a hot market segment now known as demand side platforms, or DSP.
Best of breed data analytics to enable the largest, most sophisticated ad buyers to get and use all information needed from every aspect of every ad campaign. Select and deploy new solution within three months with minimal internal resources.
A purpose-built, high-performance data warehouse appliance that makes advanced analytics on very large data volumes simpler, faster and more accessible.
Transparent view of every impression and factor affecting performance of more than 13 billion ad impressions per day; one client achieved campaign goals while reducing CPA from $170 to $80; requires half the manpower to deliver 10x the output with 3x more advertisers, more media channels and clients.
The supply of online display advertising inventory skyrocketed in the mid-2000’s with the rapid growth of social media networks, blogs and content-sharing websites — everything from Facebook, YouTube and Flickr to millions of individual blogs. An urgency to monetize billions of new ad impressions spawned a slew of new ad networks and ad technology innovation that culminated in the creation of the online advertising exchange.
By enabling many networks, publishers and advertisers to connect with one another on a unified, auction-based media trading platform, the ad exchange promised to automate much of the complexity out of the online display advertising process. The hope was that automated impression- level bidding might be just what sellers of display advertising needed to ignite demand from more buyers for their surplus inventory. More exchanges, new networks and tech providers rushed in, initially to help publishers — the “supply side” — generate more revenue from abundant and inexpensive inventory.
Enter MediaMath, a New York City startup with an idea to help ad buyers capitalize on the rapidly evolving marketplace by providing services and tools for ad agencies — the most active “demand side” players — to identify, bid on and buy just those impressions most likely to yield the results they sought for their clients. Implementing the idea in 2007, MediaMath created a hot market segment now known as demand side platforms, or DSP.
Exchanges had also made it easier to use anonymous cookie data and third-party sources to track and reach millions of visitors across the internet. The ensuing avalanche of data started a mind shift among ad buyers away from the simple procurement of commodity impressions to the algorithmic buying of audiences likely to be most receptive to an advertiser’s messages. MediaMath needed serious analytical power to enable sophisticated buyers to harness and channel data to drive optimum performance for any advertiser, campaign or marketing objective, using any combination of data inputs. Ultimately the new platform had to ingest and analyze massive volumes of data from multiple sources to make ad optimization and delivery decisions in milliseconds.
During MediaMath’s startup phase, the company used and outgrew MySQL. They tried Oracle Standard Edition with about five terabytes of data but determined it lacked scalability. According to Chief Technology Officer Roland Cozzolino, it was difficult to ingest and store data from 50 million daily transactions, let alone handle their growth to 350 million transactions per day. It took “tons of partitions to summarize and break data into vertical buckets by advertiser for analysis,” said Cozzolino, and the Oracle platform limited any critical ad-hoc analysis capabilities that were required to understand data value and to “gain a horizontal view of the business.”
That horizontal view is essential in online ad decision making, and speed is critical to ad agencies responsible for investing their clients’ budgets to achieve specific performance objectives. If something in a campaign goes wrong — or if something exceeds expectations — the sooner the buyer knows, the sooner he or she can optimize and adjust how and where to direct the flow of ad dollars. If gaining and acting upon campaign performance insights could be automated substantially, all parties to the transaction would benefit.
Getting to speed-of-thought analysis
Seeking a solution to their data challenges, Tom Craig, VP of Information Strategy at MediaMath, tested databases while Cozzolino tested in-memory coding. Code was fast, but difficult to maintain. They needed something they could deploy quickly, and that would scale to meet the demands of their fast-growing business.
From an initial consideration set that included Aster Data, Hadoop, Infobright, Oracle, Teradata and Vertica, the finalists were IBM® and Greenplum. Cozzolino reported, “We selected IBM PureData™ System for Analytics, powered by Netezza® technology, because it offered the best ROI, the fastest time-to-market of any solution, ease of use and a low total cost of ownership.” TCO was key, because there would not be a lot of internal resources available to build applications and support the solution selected. Plus, competition was heating up and MediaMath insisted on bringing the new platform live that quarter with few resources allocated to it.
Craig noted, “We knew where we wanted to take this market, but were unable to execute on that vision with our current tools. Through our proof of concept, IBM PureData System for Analytics clearly demonstrated it provides speed-of-thought analysis that helps us extend our ‘market leader’ status.” In addition to fast deployment and low resource requirements, MediaMath needed their new solution to enable:
• Flexible reporting, including dashboards and campaign diagnostics
• Capability to use all of their data in the company’s proprietary optimization algorithm for fast, thorough decision-making
• Special applications such as cross-channel attribution analysis enabling advertisers to gather and de-dupe all user data across display, email, and search, from a multitude of sources and technical platforms
• Internal/financial reporting
• Dynamic interval reach and frequency
• Purchase funnel analysis
• Deep site analysis and classification
• Fraud detection at an IP level
• Near real time (e.g. 15 min) reporting and attribution
IBM PureData System for Analytics met MediaMath’s requirements while adding analytic capabilities that were impossible previously. IBM PureData System for Analytics could recast and strengthen MediaMath’s deliverables because “we could give it more data, and faster,” Cozzolino said.
“Before IBM, we had to be far more deliberate in our data tracking and reporting because our analytic computing power was limited,” according to Craig, who had worked with Netezza previously at AOL. “With IBM, we pour through hundreds of millions of rows, with as many dimensions as are available, to look at and consider all the data. This intensive data mining enables better ad decision making. Forget about aggregating and bucketing data. It makes us and our customers smarter. You can pull out all the facts you want.”
Ad exchanges provide access to billions of buying opportunities each day. MediaMath has proven to be the best at matching those buying opportunities to their client goals. With the IBM PureData System for Analytics powering the algorithmic trading engine, MediaMath is able to listen to and act on more of those opportunities and drive up campaign performance. “PureData System for Analytics enables MediaMath to deliver on the promise of impression level bidding in real time,” Craig said.
Cozzolino added, “We asked our clients for every possible fact they wanted to see from campaigns running on our platform. The top of the list is simply, ‘transparency,’ visibility into the very granular data exhaust from the buying process. This data could only be delivered with the capacity and capabilities of the PureData System for Analytics.” This includes, for example, identifying patterns in the data over much longer time periods to understand true statistical flow. Without IBM PureData System for Analytics, MediaMath could not view 30 days of data at once. “Now we can look at, for example, what happened 12 months ago and see how it relates to today.”
On any given day MediaMath may see more than 13 billion impressions (and growing). “We can tell you about every single impression we see every day. Seeing and knowing where every impression originated helps us create the most effective machine learning algorithm and best overall performance of any DSP,” according to Cozzolino.
With IBM PureData System for Analytics, “everything is done in real time,” Craig said. “We can value impressions and determine pricing on raw data. We can tell clients if they’re seeing a difference in average price or standard deviation or median for different time periods. We couldn’t capture and report that data before.”
Said Cozzolino, “PureData System for Analytics today is the hub of everything we do. If I removed it and there was nothing in its place, the only knowledge we would have of what is happening would come from staring at log files. We’re the first and largest DSP and we innovate continuously. PureData System for Analytics reduces cycle times and the learning curve for faster development.”
Data and analytics as competitive weapons
The DSP market has become crowded with companies wanting to help advertisers invest an estimated $8 billion in online display-related ad spending (IAB/PriceWaterhouseCoopers Internet Ad Revenue Report, 2009 Full Year Results). MediaMath executives believe that their proven results and unmatched scale have kept them ahead of the pack. Now they are also “the insight DSP,” according to Craig, which enables them to continue to set the pace for the category as the one platform that most effectively lets large, sophisticated ad buyers derive maximum value from data. “Our MathClarity product [illustrations below] brings in all the data you want and presents it exactly as you want to consume it. This enables clients to make intelligent data-based decisions that improve performance and profits.”
MediaMath’s ambitious roadmap includes more industry firsts that tap the power of IBM PureData System for Analytics, such as introducing predictive analytics for replaying past events while changing one variable at a time to see how it would have changed the outcome. Said Brad Terrell, IBM Vice President and General Manager, Digital Media, “Terabyte-scale data analysis is the new weapon for competitive advantage on Madison Avenue, and the IBM PureData System for Analytics provides MediaMath with the infrastructure to optimize and serve billions of daily impressions — establishing a technical foundation for their long-term success in this industry.”
About IBM PureData System for Analytics The IBM PureData System for Analytics, powered by Netezza technology, integrates database, server and storage into a single, easy-to-manage appliance that requires minimal setup and ongoing administration while producing faster and more consistent analytic performance. The IBM PureData System for Analytics simplifies business analytics dramatically by consolidating all analytic activity in the appliance, right where the data resides, for industry-leading performance. Visit: ibm.com/PureData to see how our family of expert integrated systems eliminates complexity at every step and helps you drive true business value for your organization.
About IBM Data Warehousing and Analytics Solutions
IBM provides the broadest and most comprehensive portfolio of data warehousing, information management and business analytic software, hardware and solutions to help customers maximize the value of their information assets and discover new insights to make better and faster decisions and optimize their business outcomes.
For more information
Help IT make the shift to the strategic center of your business, and leverage proven expertise to take the lead. To learn more about the PureSystems™ family and the PureData System for Analytics, contact your IBM representative or IBM Business Partner, or visit the following websites: ibm.com/PureSystems/PureData or ibm.com/software/data/ puredata/analytics/.
Products and services used
IBM products and services that were used in this case study.
© Copyright IBM Corporation 2013 IBM Corporation Software Group Route 100 Somers, N Y 10589 U.S.A. Produced in the United States of America May 2013 IBM, the IBM logo, ibm.com, PureData, and PureSystems are trademarks or registered trademarks of International Business Machines Corporation in the United States, other countries, or both. If these and other IBM trademarked terms are marked on their first occurrence in this information with a trademark symbol (® or ™), these symbols indicate U.S. registered or common law trademarks owned by IBM at the time this information was published. Such trademarks may also be registered or common law trademarks in other countries. A current list of IBM trademarks is available on the Web at “Copyright and trademark information” at ibm.com/legal/copytrade.shtml Netezza is a registered trademark of IBM International Group B.V., an IBM Company. This document is current as of the initial date of publication and may be changed by IBM at any time. Not all offerings are available in every country in which IBM operates. The performance data and client examples cited are presented for illustrative purposes only. Actual performance results may vary depending on specific configurations and operating conditions. It is the user’s responsibility to evaluate and verify the operation of any other products or programs with IBM products and programs. THE INFOR MATION IN THIS DOCUMENT IS PROV IDED “AS IS” W ITHOUT A N Y WARR A NTY, EXPRESS OR IMPLIED, INCLUDING W ITHOUT A N Y WARR A NTIES OF MERCHA NTABILITY, FITNESS FOR A PARTICULAR PURPOSE A ND A N Y WARR A NTY OR CONDITION OF NON-INFRINGEMENT. IBM products are warranted according to the terms and conditions of the agreements under which they are provided. Actual available storage capacity may be reported for both uncompressed and compressed data and will vary and may be less than stated.