How Conversant Uses Data Science to Bring Ultra Transparency to Online Advertising
John Wanamaker, the 19th century department store magnate, once famously said: “Half the money I spend on advertising is wasted; the trouble is I don't know which half.” He wouldn’t have had that problem if he’d worked with Conversant. Sometimes timing is everything.
Conversant is a 21st century Chicago-based personalized digital marketing company. It uses analytics and data science to help its clients in retail, telecom and other industries place the right online ads in front of the right people at the right time. Conversant does this by collecting and analyzing massive amounts of online behavior data to create unique user profiles, then applies predictive models to match those profiles with targeted messaging in real time.
“Maybe Jane Doe is a soccer mom, maybe she drives a minivan, maybe we have that type of information available to us,” explains John Conley, Vice President of Data Warehousing at Conversant. “We can use that to customize and tailor her profile and then make sure she is receiving relevant content on the Internet.”
The more relevant the content, the more likely someone is to take an action, like making a purchase. And that’s what advertisers are ultimately after.
Whatever use case we can dream up and whatever ways we can think of to better understand the user, Greenplum allows us to do it.”John Conley, Vice President of Data Warehousing, Conversant
But Conversant doesn’t stop there. It also uses analytics and data science to measure the effectiveness of each ad it places in front of users and reports that information to clients. This gives clients insight into which ads most often lead to purchases and which ones don’t so they can adjust their creative direction and ad spending accordingly. Wanamaker would be envious.
“Digital allows for us to share with our clients measurement that they've never been able to get out of another channel,” said Conley. “That's the kind of measurement that the entire industry is just incredibly thirsty for, that our advertising solution brings to the table, and that Greenplum is really the basis for how we do all of that.”
Conley is referring to Pivotal Greenplum, an open source-based, massively parallel processing (MPP) analytical database that Conversant uses for a variety of data-intensive workloads. Conversant chose Pivotal Greenplum in part because it provides the company a single, next generation platform to support the different needs of business users, analysts and data scientists alike, according to Conley.
“Greenplum is just a very versatile analytics platform that allows us to do a number of different things,” Conley said.
The company’s data scientists, for example, use Pivotal Greenplum to develop the sophisticated predictive models that are then run against live data in real-time to match the most relevant ads to users. Conversant’s business users and analysts also rely on Greenplum to support the company’s internal reporting and business intelligence capabilities. And, of course, Greenplum enables Conversant to understand and report to clients the effectiveness of the ads themselves.
“Whatever use case we can dream up and whatever ways we can think of to better understand the user, Greenplum allows us to do it,” Conley said. And in those cases in which Greenplum doesn’t support a particular analytics function out-of-the-box, Conley and his team often build extensions to the database to handle the task.
“I’ve got the best technical team in the business. So we’re able to build in additional functionality as needed because of my team’s capabilities and Pivotal Greenplum’s open source nature,” Conley said. Pivotal Greenplum is based on the open source project Greenplum Database. Conley and his team developed its own unique HyperLogLog implementation, an algebraic routine that enables Conversant to get an extremely accurate transaction count without overtaxing the system, for example. Conversant often contributes its homegrown Greenplum extensions to the open source project, so other users can benefit from them.
Versatility and extensibility weren’t the only characteristics that drew Conversant to Pivotal Greenplum. Scalability and performance were also key. “Our business is based on data. It's the Internet ... there's a lot of it,” said Shaun Litt, a principal architect at Conversant.
Conversant needs to quickly analyze petabytes of data to support its operations. Pivotal Greenplum’s MPP architecture allows it to scale out horizontally, rather than scale up vertically like many competing data warehouses. This allows Conversant to easily grow the platform to accommodate ever-increasing data volumes simply by adding more commodity nodes to its Greenplum cluster. Data is distributed across the machines in the cluster and analyzed in parallel, providing the processing speed and efficiency Conversant needs.
“Many companies worry about loading millions of rows of data. I don't start freaking out until its … hundreds of billions of rows of data!” said Litt. “That's just the world that we live in. So, we needed a solution that scales. Greenplum offers us that solution to be able to load and analyze our data at massive scale.”
GE customers are also building applications on Predix for use cases as diverse as optimizing wind farm performance, helping doctors make more accurate diagnosis, and improving aircraft engine efficiency.
Conversant clients - which include the likes of Cracker Barrel, Staples and Cabela - see a 10 times incremental return on ad spending, on average, according to the company. And with the insights provided by Pivotal Greenplum and Conversant, clients can quickly zero in and double down on the ads that are working … and scrap the ones that aren’t performing, pushing their return on investment even higher. Wanamaker himself couldn’t ask for more.