Users performing software experiments will soon have a new option to store back-end data in an existing data warehouse system after an acquisition by LaunchDarkly this week.
LaunchDarkly's software supports feature management, a practice that separates the release of software features into production and those features' exposure to live end-user traffic, using software toggles called feature flags. Proponents of feature flags, which are also offered by DevOps platform vendors such as Harness and CloudBees, say they make software deployments less risky by exposing changes to a small group of end users before fully rolling them out. LaunchDarkly's software also supports rapid rollback for problematic features.
As generative and agentic AI apps hit the enterprise IT mainstream, however, software experimentation -- in which organizations release multiple slightly varied versions of a change to groups of end users and gauge their response before performing a full deployment -- is gaining interest among LaunchDarkly's customers, according to the company's chief executive.
Dan Rogers
"Our customers said to us, 'We love what you're doing on [making] releases [less risky] ... [and] progressive rollouts,'" said Dan Rogers, CEO at LaunchDarkly. "'But as it comes to experiments, as it comes to launching these AI features, a lot of our data is already residing in Snowflake. Can you please work on that territory instead of us having to send you unique metrics, unique events and unique [user] sessions?'"
That's where Houseware.io comes in. It's a startup based in San Francisco and acquired by LaunchDarkly for an undisclosed sum this week. In 2022, the company won the Snowflake Startup Challenge for a no-code app it developed using the Snowflake Native App Framework. That app weighs transactional, performance and user experience data against product and business objectives. Now called Warehouse Native Experimentation, the app is in private preview for LaunchDarkly customers and used internally at LaunchDarkly, Rogers said.
The new app, and a newly inked partnership with Snowflake, will expedite feedback on LaunchDarkly experiments using a centralized data store that many customers already own, he said. Rogers also didn't rule out future integrations with other data warehouse vendors such as Databricks.
LaunchDarkly has been strong on engineering impact. Houseware may help them improve their ability to run experiments that show business results.
Andrew CornwallAnalyst, Forrester Research
Linking IT and business data for software experimentation was an area LaunchDarkly was less focused in the past, in favor of software engineering features, according to one analyst.
"LaunchDarkly has had the ability to deliver features and experiment by [end-user] segment. By adding Houseware.io's technology, they give their customers the opportunity to discover segments that have been overlooked," said Andrew Cornwall, an analyst at Forrester Research. "LaunchDarkly has been strong on engineering impact. Houseware may help them improve their ability to run experiments that show business results."
LaunchDarkly experiment platform competitors include Spotify's Experimentation Platform, Microsoft's ExP, Uber's XP, the Netflix Experimentation Platform, Optimizely and Amplitude.
Users envision experimentation expansion
LaunchDarkly customers said they appreciate that the company chose not to create its own duplicate data warehouse.
"LaunchDarkly is connecting product analytics and business intelligence," with this acquisition and product, said Artie Lee, director of engineering at Climate LLC, a subsidiary of Bayer that makes digital farming software. "In this current climate of cost-consciousness, that’s a very compelling proposition."
Parent company Bayer already uses Snowflake, whereas Climate does not, Lee said. But "Houseware is definitely interesting for ramping up a company's maturity level in data-driven decision making, including for Climate," he said. "It's one thing to simply have data, but that data is only valuable insofar as it is being analyzed and compared against hypotheses, so having a tool that automatically provides those insights can save a lot of time and effort."
Lee hasn't had the chance to test Houseware's software, but he said he can imagine potential uses for the Warehouse Native Experimentation feature that measure user engagement for Climate's FieldView product.
"I can envision experiments around figuring out not just what changes or features drive engagement, but giving us a better understanding of why those things drive engagement, and adjusting our product roadmaps to aim in those directions more," Lee said.
Another LaunchDarkly customer, supply chain logistics software company SPS Commerce in Minneapolis, anticipates it will have fresh opportunities to put out its own software in the Snowflake marketplace as a result of the LaunchDarkly partnership.
"Having that data accessible in existing warehouses makes it much more convenient to incorporate into existing systems without manual extraction or loading in place," said Travis Gosselin, distinguished software engineer at SPS Commerce. "This might be a timely boost in helping SPS Commerce get LaunchDarkly data directly integrated into some of the operational confidence reports that [technology] teams build out."
Feature flag cleanup pain remains
A common challenge with feature flags is managing when to remove them from production code, and LaunchDarkly has added features to support more systematic forms of flag cleanup, including for mobile app environments. But there's still room for improvement, according to Gosselin and Lee.
"As longtime users of LaunchDarkly, we have a lot of mature teams that clean up and use flags very appropriately, but we are still onboarding new teams, and some teams are not as far along in the journey," Gosselin said. "While LaunchDarkly brings a number of capabilities to governance and compliance, I think we need more tooling to help enforce hygiene and best practices around flag usage."
Specifically, Gosselin said he's asked for an enforceable limit on the number of feature flags a project can use before developers are required to clean them up; a more structured approach to the ownership of flags and their associated codebases using automated groupings; and automated flag cleanup based on new insights from data warehouse integration, including direct pull request submissions to teams when LaunchDarkly detects that a flag can be removed.
Lee said he would like to see "stats on how often flags get evaluated so we can make better business decisions on things such as if or when we want to force upgrades so we can [retire] features, support, and back-end services."
A LaunchDarkly spokesperson didn't immediately respond to a request for comment on these items as of press time.
Beth Pariseau, senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.
Dan Rogers
