Engineering drives innovation when you understand whether or not you are solving the problem.
I had the opportunity to speak with Nick Kephart, V.P. of Marketing at Split Software which provides a feature delivery platform to manage feature flags, monitor release performance, and experiment to solve business problems and achieve KPIs.
Feature Monitoring enables enterprises to release features without rollbacks and hotfixes. By ingesting performance data, the platform shows performance and specific user issues. Problem features can be pinpointed with the automated detection of significant errors and automated attribution to each feature in the release. Alerts notify specific team members responsible for the feature and there’s an instant kill switch to reduce poor user experience (UX).
This enables companies to find failed features before customers experience them. Errors are identified within 30 minutes and tied to specific features. Failed features can be turned off instantly without the need for rollbacks or hotfixes. Companies can limit exposure by gradually rolling out features to a well-defined portion of their users.
Feature Flags (aka feature toggles) are changing developer workflows. Every enterprise has an approval process/pull request for code changes in production and feature flags are included in the process. Feature flags and gradual rollouts make changes to UX in production even faster than code changes. Feature flags need to be integrated with approval processes. SOC2 is not feasible without this.
Approval Flows bring workflow standards to feature flags. Email notifications request individuals or groups to submit, approve, withdraw, or reject making it easy to adopt feature flags with workflow engineering teams. Providing reviews and comments on production changes prevents risky changes from making it to production. They include full audit information about submitter, approver, and status to stay compliant with company policies and audits. Users can see all past and pending changes making it easy to see why a change was made.
Enterprises are using Customer Data Platforms (CDP) to standardize data flows. Split has bi-directional feature integration with market-leader Segment mParticle. Inbound, they enable feature delivery with the existing data pipeline and outbound they send feature data to long-term storage for further analysis.
Google Analytics integration reduces risk on KPIs and business metrics with bi-directional feature integration. Inbound features are delivered with existing web data and outbound feature data is collected for segmentation.
Streaming Architecture enables real-time feature updates with multi-service applications and widely-distributed clients. Streaming architecture propagates changes in near real-time. SDKs are notified of changes fetching the latest change from the CDN with an option for environments with strict firewall rules due to streaming service open connection.
Use Cases
Vistaprint avoided 4X load time degradation when they rolled out a feature that provided early signals that would have hurt exit rates. They used Signal to kill the feature.
Go Daddy increased signups by 8% by running dozens of experiments, seeing what’s working and what isn’t, and optimizing the signup process.
Takeaways
Deploy continuously and then choose the time to release so you can coordinate between teams and put dependencies on other flags.
During release, have the ability to find problems quickly in production with small groups of end-users.