ForecastingKargo · 2024
Demand forecast v3
Weekly demand for 14k routes. MAPE down from 31% to 12%; saves about $4M a year in empty trucks.
Python / LightGBM / dbt
Live · Singapore
Wei Chen
Senior Data Scientist
Forecasts that people trust enough to act on.
01 · annual savings
$4M
02 · churn
−38%
03 · MAPE
12%
Summary
I build forecasting and experimentation systems for marketplaces. My favourite result is a model that gets switched off because the process it fixed no longer needs it.
Previously at Grab and Shopee; now leading forecasting at a logistics scale-up.
Results
Forecasting
ExperimentationKargo · 2023
Onboarding experiment
Six onboarding paths tested on 90k shippers; the winner cut 90-day churn by 38%.
SQL / Bayesian A/B
MLGrab · 2022
Delivery delay model
Predicted late deliveries 20 minutes earlier, so support could message customers first.
PyTorch / Spark
Track record
- 2023 → Present
Senior Data Scientist · Kargo
Lead forecasting; team of three.
- 2019 → 2023
Data Scientist · Grab
Delivery ETA and dispatch experiments.
- 2017 → 2019
Analyst · Shopee
Pricing and promotions analytics.
Toolkit
- Python
- SQL
- Time series
- Causal inference
- Experimentation
- dbt
- Spark
- PyTorch