The dataset consists of user navigation events throughout the onboarding funnel:

• user_id: Unique identifier per user.
• group: A/B group label – control or treatment.
• step: Funnel step (e.g., step_1, step_2, ..., confirm).
• timestamp: Timestamp of the event.
• Derived fields (created in preprocessing):
  • time_diff: Time difference between steps.
  • final_step: Last recorded step per user.
  • error_flags: Indicators for anomalies.

To ensure valid insights, we applied rigorous data preprocessing steps:

• Chronologically sorted user steps.
• Removed sessions with:
  • Repeated steps (e.g., multiple step_2s).
  • Backward jumps (e.g., step_3 to step_1).
  • Zero-second step transitions.
• Labeled sessions that did not end in confirm as abandonments.
• Engineered features for:
  • Time spent per step.
  • Funnel depth reached.
  • Navigation consistency.

These cleaning rules were based on domain assumptions. In a production setting, more session metadata and UX feedback would further guide this logic.

The EDA compared control vs. treatment groups on multiple dimensions:

• Completion Rate – % of users reaching the confirm step.
• Drop-off Points – Common exit steps.
• Navigation Errors – Frequency of repeated or reversed steps.
• Time Metrics – Total and per-step time comparisons.

Key visualizations:

• Funnel diagrams by group
• Step-wise conversion rates
• Session length distributions

We ran statistical tests to validate whether differences observed were statistically significant:

• ✅ Z-test for proportions (completion rate comparison).
• ✅ T-test / Mann-Whitney U test (time differences).
• ✅ Chi-squared test (distribution of final steps and errors).
• ✅ Sanity checks for group balance and random assignment.

Assumptions tested:

• Normality (Shapiro-Wilk, histograms).
• Equal variances (Levene's test).

Statistical significance ≠ business impact. All insights were contextualized with user experience and operational considerations.

• ✅ Higher completion rate in the treatment group (statistically significant).
• 🔄 Fewer navigation errors post-redesign, especially backward transitions.
• ⏱️ Time efficiency slightly improved but not statistically conclusive.
• 🧩 Users followed a more linear path in the redesigned flow.

• ✅ Roll out the redesign to the broader user base.
• 📊 Monitor funnel metrics continuously to detect regressions.
• 🔬 Run additional segmented tests (e.g., new vs. returning users).
• 🧠 Gather qualitative UX insights (e.g., via surveys, heatmaps).
• ⚙️ Improve experiment design with longer run times and controlled traffic splits.

This analysis was developed as part of a Data Analytics Bootcamp.
It reflects industry-standard approaches to experimentation, data cleaning, and interpretation of A/B tests within digital products.