Variations

Variations are string values that we evaluate against features using Featurevisor SDKs in our applications.

They are often called A/B test experiments, which helps us test different variations of a feature and see which one performs better.

Variations are one of the 3 different types of values that we can evaluate against features, including:

• Flags
• Variations (you are here)
• Variables

Visual explanation#

Starting with initial weights#

We start with a feature that's:

• rolled out to 60% of all of our traffic targeting everyone
• with a 50-50 split between the A and B variations

Usually they are named as control and treatment variations in A/B test experiments.

The above feature could be expressed as follows:

description: My feature description here...
tags:
  - web
  - ios
  - android
bucketBy: userId
variations:
  - value: A
    weight: 50
  - value: B
    weight: 50
rules:
  production:
    - key: everyone
      segments: '*'
      percentage: 60

The bucketBy property makes sure that the same variation value is always evaluated for the same user across all of our applications, achieving a consistent experience for the user.

Learn more about complex variation definitions here.

Increasing the rollout percentage#

Let's say we want to increase our rollout percentage from 60% to 80%:

# ...
rules:
  production:
    - key: everyone
      segments: '*'
      percentage: 80

This will result in the following:

For the first and second users, the experience will remain the same as before, as they are still in the 60% rollout range.

But for the third and fourth users, the experience will change now that they are in the 80% rollout range.

Decreasing the rollout percentage#

If we decrease our rollout percentage from 80% to 70%, a re-bucketing process will be triggered for all users in the feature.

# ...
rules:
  production:
    - key: everyone
      segments: '*'
      percentage: 70

The impact will be the following:

You will notice that the third user would move from variation A to variation B because of this re-bucketing.

The same re-bucketing process would have happened if we changed the weights of the variations as well.

Increasing again#

From the latest state, if we increased the rollout percentage all the way from 70% to 100%:

We would still maintain consistent bucketing for all users in the feature because the percentage value is still increasing.

Usage of SDK#

If we take JavaScript SDK as an example, we can evaluate the feature's variation as follows:

import { createInstance } from '@featurevisor/sdk'
const f = createInstance({
  datafile: { ... }
});
const variation = f.getVariation('my_feature');

Learn about more advanced use cases for evaluations here.