01 The Problem
A logistics company ran training across four regions with no idea what was working — virtual vs in-person, course order, employee type, or which of two new programs to roll out. I set out to answer all four with data.
02 The Approach
Data Merging
Combined 6 datasets (employees, enrollment, courses, offices) into one clean table using shared IDs.
Employee Segmentation
Ran PCA and K-means clustering on a 12-question survey to group employees by motivation type, then confirmed the groups were real with a chi-square test.
Format & Sequence Analysis
Compared virtual vs in-person courses and tested which course order drove the biggest score gains.
A/B Testing
Ran a controlled test on 593 employees across 6 offices, using Difference-in-Differences regression to isolate the real effect of each program from noise.
03 The Result
Collaborative Optimizers 33%, Efficiency Seekers 25%, Goal-Driven Achievers 17%, Process-Oriented Stabilizers 13%, Social-Growth Enthusiasts 12%
Unevenly spread by region (p < 0.001).
Moving one advanced course from first to last raised its score improvement from 0.63 to 0.74
Sequence alone moved the needle — no new content required.
Several core courses did just as well online as in person
Format wasn’t the tradeoff leadership assumed.
Program A +4.93% on practical skills; Program B +1.73% on proficiency
Both beat the current program — on different outcomes.
04 Business Impact
Personalized Plans
Train by motivation type instead of one-size-fits-all.
Go Virtual
Expand virtual courses since they don’t sacrifice results.
Program Rollout
Program A for hands-on roles, Program B for knowledge-heavy roles.
Free Fix
Reordering courses boosts results without new content spend.
05 What I Learned
Raw before/after numbers can lie. The DiD test taught me you need a real comparison group to know if a program actually caused the result.