A revamp on the internal tool for A/B testing within the organization

Calculon

Problem Statement

Today, we find that Calculon has many different results pages which causes a lot of discrepancies in the UI, issues in UI implementation, and velocity problems.

The goal will be to make the results page a single page so that we can better provide information to the users, minimize re-use and prevent UI errors.

Objective

  1. Help Agodans run better quality  experiments and make better decisions

  2. Provide accurate measurement of true value

  3. Enable experimentation in new areas

  4. Provide a smooth UI/UX for our tools (Calculon & FE Analytics)

  5. Make sure our data, results, and tools are trustworthy

Help Agodans run better experiments and make better decisions


  • Improve signal-to-noise of experiments

  • Better recommendation

  • Show more relevant data to users (consider the lifecycle of the experiment)

  • Educate users

  • Promote learning from past experiments

Provide accurate measurement of true value for Agoda


  • Improve how eval run is done

  • Measure other benefits besides UBI

Enable experimentation in new areas


  • Experiment design and validation of new metrics, stats calculation, …

  • Scale up in terms of supporting teams (e.g. not just consultancy, but also providing tools)

Provide a smooth UI/UX for our tools (Calculon & FE Analytics)


  • More tooltips to explain how things work

  • Faster search

  • Unified UI

Make sure our data, results, and tools are trustworthy


  • Improve consistency between Calculon & FE Analytics

  • Display correct data and results in UI

  • Reliable and robust data pipeline

  • Minimize escaped bugs

Design Revamp

  • Navigation

    Detailed design navigation to help user locate features

  • Summary Card

    Enabling users to subscribe to experiments only from their team provides. overall view of running experiements

  • Experiments List

    Display as much relevant data points at front helps user to have a good glance of the experiment performance