The biggest difference between our legacy Full Stack experimentation solution and Feature Experimentation is that all experiments, targeted deliveries, and Multi-armed Bandits (MABs) are based on feature flags. By eliminating the legacy A/B Tests in favor of Rules Engine and A/B Test rules, all A/B Tests are essentially Feature Tests (from legacy), and winning variations can be rolled out without developer involvement.
Today, all Feature Experimentation customers can create flags with up to one A/B Test rule and one Targeted Delivery rule. With this feature, we eliminate the cap on up to two rules per flag. This means that, with a single flag, customers can run several concurrent A/B Tests, MABs, and Targeted Deliveries.
This is a game changer for experimentation velocity: If Feature Experimentation customers follow the recommended workflow (see below), they can minimize engineering involvement and enable non-technical collaborators to run many experiments and targeted deliveries without using valuable engineering resources after the flag is implemented and tested.
This feature has the potential to double or triple a customer’s experimentation velocity alone. It also unlocks a new category of use cases for the industry: feature personalization, or the ability to tailor individual features to one or many audiences simultaneously.
Personalize your features
With Multiple Experiments and Deliveries per Flag, Feature Experimentation customers can follow the recommended workflow (below) to seamlessly run experiments, gain valuable insights, and roll out winning variants – all without needing a developer after the flag is implemented and tested.
Feature Experimentation users can run more than one experiment and one targeted delivery at a time, this capability lets you personalize features at scale by enabling the release of multiple versions of a feature to different audiences at the time same, all using the same flag.
How Rules Engine works
Rules Engine is unique to Feature Experimentation and enables users to create a series of A/B Tests, MABs, and Targeted Deliveries on a single flag.
It works by sequentially evaluating each rule. While the evaluation logic is complex, it was created intentionally to maximize the value of web traffic traffic by algorithmically bucketing users in the first experiment or Targeted Delivery for which they qualify, instead of disqualifying them at the first opportunity.
Order of operations
When a decision needs to be made for a user for a given flag:
Feature Experimetation users can get the most value from Rules Engine and Multiple Experiments and Deliveries per Flag by following this workflow for all features (and flags):
- A non-technical user or developer creates a Flag in the Feature Experimentation UI with variables for each part of the feature anyone may want to experiment on in the future. This can be thought of a feature or flag specification or spec.
- The flag spec and corresponding code snippet from the Feature Experimentation UI can be put into a Jira (or workflow tool) story for a developer to implement.
- A developer implements the flag following the flag spec, with sensible error handling (should there be an issue with the flag) and unit tests for the flag itself, including all its variables. Developers should test their features with multiple values for each flag variable, including empty or null values.
- Once the developer is confident that the flag works as expected, they may deploy the code with the flag off. This is sometimes called a dark release.
- No further developer involvement is required. Non-technical users may now add rules to the flag’s Rules Engine pane, including A/B Tests, MABs, and/ or Targeted Deliveries. They may turn the flag and rules on and experiment to their heart’s content, without the need to align with the Dev team.
- As A/B Tests are run and winning variants are identified, customers may pause or remove the A/B Test rules and add Targeted Delivery rules to roll out winning variations, all without needing a developer.
Q: Is Rules Engine available for Full Stack (legacy) projects?
A: No. This is only available in Feature Experimentation. There is a free of charge 1-click migration tool available to assist with Migrating from our legacy Full Stack to Feature Experimentation
Q: Is Multiple Experiments & Deliveries per Flag available for Full Stack (legacy) projects?
A: No. This feature requires Rules Engine and Rules Engine is only available in Feature Experimentation.
Q: Why is there a different order of operations for evaluating experiments and targeted deliveries in Rules Engine?
A: Experiment rules (A/B Test & MAB) are not evaluated the same way as Targeted Delivery rules. In both, if the user doesn’t match audience criteria, the next rule is evaluated. For experiments, if the user is not bucketed into the experiment’s traffic allocation, the next experiment rule is evaluated. For targeted delivery, if the user matches the rule’s audience but isn’t bucketed into its traffic allocation, the rule engine jumps to the last rule (Then, for everyone else...) instead of evaluating the next rule.