Split Lifecycle Management

As you scale utilization it becomes increasingly important to have a strategy for how to best  "clean up" splits.  This should be a consideration from the very first split you create.

When creating new splits 

• There are a variety of reasons for placing a change or new feature behind a split. The three main types of splits to consider: 

Examples of Release Management splits 

• I want to expose a feature to a subset of my traffic that meet a certain criterion (location, age, plan type) 
• I want to safely do a progressive or gradual rollout for a potentially high impact feature, expose to 10% at first then gradually increase 

Example of Experimental Splits 

• I want to learn if the new version of an existing feature, or an entirely new feature, has a positive or negative impact on my business metrics 

Example of Operational Control

• If this feature were to have an issue, it could potentially impact revenue or conversions, I want the ability to quickly revert back to the previous steady-state 

When to Retire a Split 

Now that you have splits in production, what are the indicators you might establish for when to retire a split. 

• Splits that have not been modified within the past 30 days = eligible for retirement 
• Splits that have received no traffic in the last 7 days = eligible for retirement 
• Metrics show that you the split has proven the feature should be rolled out to everyone or removed from code = eligible for retirement 

To surface the Splits that fall into the first two criteria, within the Environments tab in your Split dashboard you have the option to filter by; 

• Active Splits 
• "Killed" Splits 
• Splits that have received traffic in the last 7 days 
• Splits that have received no traffic in the last 7 days 
• Splits that have seen no traffic at all 
• Spits that have been modified in the last 7 days 
• Splits that have not been modified in the last 30 days

For Consistency 

• Before removing a split from code, ensure that everyone in the population is receiving the same intended treatment, essentially rolled out to 100%

Incorporating into your existing process: