Jampp Case Study

Founded in 2013, Jampp specializes in improving the revenue advertisers generate from their apps. It does this with a suite of analysis tools, combining behavioral data and predictive technology to acquire and engage app customers.

Riding a wave of industry growth, Jampp has expanded by 600 percent in the last three years, and works with some of the largest apps in the market including Shutterfly, Boxed, Twitter, and Yelp. It has six offices in cities around the world including London and San Francisco, and it will soon open an office in Singapore to serve customers in Asia, including Grab, a ride-hailing app.

For most of us, mobile apps are such a big part of our everyday lives that we don’t give a second thought to the work that goes on behind the scenes to keep us using them. App publishers dedicate significant amounts of money and time to getting us to download their apps, and, more importantly, spend money through them.

Big data is imperative for a company like Jampp to continue to provide value to its advertisers through analysis of customer behavior. “We recognized from day one that understanding mobile behavior was the key to generating actual app customers,” says Chief Technology Officer Patricio Rocca. “Every day we process 75 terabytes of data and have a team of data scientists dedicated to operating and enhancing our capabilities. Powered by machine-learning algorithms, our platform’s accuracy and performance improves with each new piece of information about a user, as well as the billions of advertising impressions Jampp analyzes daily around the world.”

Jampp’s platform is split into two architectures: one that handles the real-time bidding (RTB) on ad exchanges, and another that tracks user behavior. This covers any in-app event that occurs, whether that’s booking a ride, ordering a pizza, or swiping right for a potential date on Friday night. Popular apps can generate billions of events each day across their massive user bases, and it’s up to people like Rocca to find ways to handle this constant stream.

Rocca says there’s a significant risk involved with every bid Jampp makes. “Every time we bid on an ad space and win an impression, that’s a cost to our business,” he says. “But our customers only care about the revenue we generate for them, so we have to make sure we’re paying the right price for a user who’s likely to repay that investment through their mobile purchases.”

Jampp had been using Amazon Web Services (AWS) since its inception, including Amazon Elastic Compute Cloud (Amazon EC2) instances with the open-source big-data processing tool Kafka. “When we took on one large client, we ramped up by an extra 35 million events an hour and our platform broke down because it couldn’t scale on demand,” Rocca says. “Kafka is great software, but it needed a lot of effort to maintain and fine-tune it.”

The Jampp team began to look at Amazon Kinesis as a data stream for their machine-learning platform. It uses Amazon Relational Database Service (Amazon RDS) to store 75 terabytes of daily bid data processed by Amazon Kinesis; AWS Lambda functions to enrich the events; and, finally, Amazon Kinesis Firehose flushes the data to Amazon Simple Storage Service (Amazon S3). Jampp uses Amazon DynamoDB to attribute and enrich unique events, replicating its data set across multiple regions in less than 40 milliseconds and providing single-digit millisecond read times. “We know certain apps will want to send users a discount code within five minutes of that user installing the app,” says Rocca. “So we need to process the in-app events and build segments to target those users in real time. That’s not possible with any other architecture than a stream-processing one. Amazon Kinesis came to our rescue. In the last 18 months, we’ve processed 250 times the in-app events compared to before.”

The more in-app events Jampp can process, the better it understands its customers’ users and the more accurately it can target them, ultimately resulting in more engaged app users, and more sales for advertisers. “The main benefit of choosing AWS has been the positive impact on our company’s growth,” says Rocca. “By using AWS, we can process more data than before, which means we can meet the needs of even the most sophisticated and data-demanding advertisers―and more of them. Since moving to AWS, we’ve boosted the number of app advertisers we work with by 250 percent, so we now do business with more than half of the top 150 global app advertisers, improving our bottom line and ability to accelerate growth across our entire business.”

As you might expect for such a data-driven organization, Rocca keeps a meticulous eye on the performance of Jampp’s AWS infrastructure. “We’re always looking to get the most out of what we do with AWS,” he says. “We use EC2 Spot instances, for example, to cut all our server costs by 66 percent right away. Not many people use them to that extent, but you can make immediate and significant savings.” Jampp also calculated the monthly cost of its tracking platform under Amazon Kinesis to be about one-third of the price of the previous one. “With our annual growth, Kinesis saves us about $15,000 each month,” says Rocca. “And that’s without taking into account the time savings we get by using a managed service, rather than one we have to invest time in managing.”

Rocca also speaks highly of the support Jampp received when migrating its tracking platform to Amazon Kinesis. “We were dealing with AWS Premium Support, which was crucial to the success of the project. The response time we get is excellent, and the quality of engineers is really high. They were able to guide us during the migration. Many of the lessons we learned and architectures we chose were recommended by AWS solutions architects,” he says. Jampp now dedicates just two DevOps engineers to managing its entire AWS infrastructure with more than 800 Amazon EC2 instances responding to more than 500,000 queries a second, as well as multiple databases and data streams.