In 2024, the team at LaunchDarkly was busy developing a new product: Warehouse Native Experimentation. It would be the company’s first product that would feed experiment results directly into customers’ data warehouses.
“Data scientists want to be able to run their own analyses,” said Jonathan Wellons, Senior Engineering Manager at LaunchDarkly. “With Warehouse Native Experimentation, data scientists can join test results with any other data set in their data warehouse. For instance, if a customer wants to filter test results by demographics, all that data is immediately available in the customer’s warehouse.”
Product development was divided between two teams: one to build the warehouse native product and the other to build data export. Wellons was a leader on the data export team.
“Our priority was Snowflake,” Wellons said. “We have a lot of customers on the platform. We also have a data export product we built ourselves for Snowflake. The problem was that it only supported one region. We needed to support every major region on day one and to be ready to add destinations like BigQuery, RedShift, and Databricks soon after.”
LaunchDarkly had a decision to make: Should the team expand its existing proprietary product or switch to Prequel’s Data Export Platform?
“We started investigating what it would take to enhance what we already had,” Wellons said. “First, we thought about function. We were already familiar with the data sharing features most platforms offer. Databricks took an afternoon to set up and test. Then, we researched costs. We calculated the cost of supporting every major region with Snowflake and assumed the others — BigQuery, RedShift, Databricks, etc. — would be roughly the same.”
“One of our biggest concerns was that if we built this ourselves, we would need to load data into every destination and cloud region ourselves.”
“That would create opportunities for errors and discrepancies that can take significant time and customer involvement to solve. The full scope of the product would need to include consistent ownership, management, and instrumentation through the long tail of destinations, not just the handful of regions and destinations everyone uses, to ensure we can sell to and support all potential customers.”
“All of that work would take a big chunk of our time, and we would be committed to it for the product’s lifetime,” Wellons said. “As a leader, I asked myself, is this the highest and best use of my team’s time for the foreseeable future?”
While he pondered that question, the team got to work evaluating Prequel.
“Our engineers were able to validate Prequel quickly,” Wellons said. “The development team was delighted. The tools were designed well, the platform is complete, and the team provided excellent support throughout.”
“Outsourcing data export to Prequel means we wouldn’t have to load any of the data ourselves,” Wellons said. “Their system reads from our source, replicates the data wherever it’s needed, and immediately forgets the data it just moved. The process is fully instrumented, so we always understand data integrity and have all the tools we need to deliver an excellent customer experience.”
“Prequel can also support the volume we need. We have customers that need to sync tens of billions of rows per month.”
“When we compared the two options — building versus buying — Prequel was the clear winner,” Wellons said. “Prequel would unlock the market, free the team up to focus on business value instead of moving data bits, and save us several quarters' worth of work. It was an easy decision.”
We asked Jonathan Wellons, Senior Engineering Manager at LaunchDarkly, for his advice to anyone considering building data export. Here’s what he said:
1. What’s the long-term cost of support and maintenance?
It's very easy to set up one destination and think, “Hey, this is simple. Ship it.” But when you do it yourself, you have to load data into each region and destination, which adds complexity and opportunities for painful errors that impact customers.
2. What’s the opportunity cost?
In our case, buying Prequel meant that we gained two months where the team could be actively prototyping instead of worrying about data transport.
3. How many data targets do you really have?
When a big prospect comes along who wants a destination we don’t support yet, I don’t have to tell the sales team no or spend months on a custom build. We can just turn it on in Prequel and ship it right away.