# Personalized vs Fixed Next-Step Suggestion A/B Test

Page: https://alidemirbas.com.tr/en/lab/ab-test-library/tests/personalized-vs-fixed-next-step-suggestion-ab-test

## Hypothesis

If users see a next step that fits where they left off instead of the same fixed suggestion everyone gets, they may find it relevant and act on it. As long as status data is accurate, that could raise the recommended next-step completion rate.

## When to run this test

Analytics showing users glance at the fixed next-step suggestion and move on, people with finished profiles still told to complete them, or support tickets about pointless tasks all mean the suggestion does not fit where users are. If personalization works, users complete the action that fits their status more often and reach their first meaningful action more easily. If status data is missing or late, fix that first; base status only on data users have provided, and avoid copy that suggests you know something they did not share.

## Primary metric

**Recommended next-step completion rate**: Classify users before assignment based on known status and compare completion of the same predefined next action over the same period.

## Other metrics

- **First meaningful action rate**: Check whether users reach the product's core value.
- **Irrelevant recommendation rate**: Audit recommendations against recorded status and eligibility rules.
- **Seven-day retention**: Check whether retention improves.
- **App abandonment rate**: Make sure recommendations do not pressure users into leaving.

## What to check before and during the test

- **Logic**: Compare one fixed recommendation with a predefined status-based match.
- **Status accuracy**: Validate input data before assignment and update it consistently.
- **Fallback**: Define a recommendation for users whose status is unknown.
- **Presentation**: Keep recommendation count and visual presentation fixed; multiple options are a separate test.
- **User groups**: Review new and active users within the same experiment.

## What to watch when reading the result

- Higher recommendation completion is not a win until you check whether users reach the product's core value. Read first meaningful action rate, seven-day retention and app abandonment together against your measurement plan, and do not ship if a guardrail metric worsens past the limit you set in advance.
- The effect you measure belongs to this status-to-action mapping and this presentation; a recommendation with several options is a separate test. Review new and active users separately; the suggestion should not override a path users chose themselves, so check that in session recordings.
- An unclear result calls for two checks first: that both arms count completion against the same status-to-action mapping, and the irrelevant recommendation rate, since inaccurate status data can mask a real effect. Once the data is fixed, plug the current completion rate and the smallest lift worth acting on into the [sample size calculator](https://alidemirbas.com.tr/en/calculators/sample-size-calculator). Divide the sample per variant by daily eligible users per arm to size the new round.

## Frequently Asked Questions

### How can completion be compared when suggested steps differ?

Define user states and the appropriate next action before assignment. Apply that same mapping and follow-up window in both arms, including control.

### What happens if the state information is wrong?

It can suggest an irrelevant or already completed task. Validate inputs and maintain consistent updates. Audit recommendations against recorded state and eligibility rules.

### What should users with unknown states see?

Use a predefined fallback. Do not imply access to undisclosed personal information or override the user’s explicit route choice.

### Does more recommended-step completion mean success?

Use the defined next-action completion as the primary outcome. Activation, retention and exits have their agreed diagnostic or guardrail roles. Do not ship if a guardrail crosses its predefined harm limit.