In 2014, Bumble’s database was struggling to scale with the growing volume of data, and hundreds of AB tests became too many to manage without real-time feedback. Before running a test to “all women in the UK and France,” for example, Bumble needs to calculate the intersection with any existing experiments in a few seconds — on hundreds of millions of users — with the precise overlap percentage. If a user sees both versions of a test in error, the entire experiment is invalid.
Following a proof of concept (PoC), Bumble selected Exasol for security, flexibility, and performance — a 20-node cluster stores and analyzes 350 terabytes of data underneath existing BI infrastructure. They use Exasol in conjunction with Microstrategy’s SQL, and run Python, Java, and C++ directly in the database.
“Anything you can run in Linux, you can run in Exasol in parallel. We don’t need to set constraints on our users or pre-process a large query.”Artem Ivanov, Director of Business Intelligence
Bumble’s user interface and popularity is the result of thousands of tweaks over time from extensive user and AB testing
“Not only did it require little maintenance effort, the database’s internal views were amazing, Exasol has thought hard about how to make our lives easier and all the basics have been taken care of.” - Artem Ivanov, Director of Business Intelligence at Bumble
All bugs are fixed immediately, giving the Bumble team confidence to move quickly with upgrades and QA cycles. “The customer service is superb. It’s more of a partnership, we work together to suggest features and build a better product,” Artem says.