Lean Experimentation and Decision Science with Humans in Warehouses
summary
A picker-scheduling system combines stacked regression time-series forecasts with a mixed-integer programming linear optimizer to recommend shifts that match, for example, a 1,000 man hours demand profile; deployed logic targets reduced operational expense by raising picker utilization and cutting idle hours across 800 warehouses in 70+ countries. Experiments use A/B rollout between control and test warehouses, focused-group surveys to fix low-quality logs, and statistical hypothesis tests; results require repeated iterations because human errors, regional rules and stochastic receiving often break initial assumptions.
AI-written summary · about
chapters (17)
- 0:00Lean experimentation in warehouses
- 1:27Picker scheduling in quick commerce
- 3:54Picker scheduling with demand forecasting
- 5:37Inventory and regional constraints
- 7:22Operation research for inventory chaos
- 10:19Experiments with human pickers
- 13:26Biases and seasonality in decisions
- 15:35Embracing chaos in experiments
- 17:22Lean experiment stages and MVP
- 19:50Hypothesis testing with humans
- 20:55Actionable hypotheses
- 22:43Scientific business experiments
- 23:25Decision intelligence in uncertainty
- 24:48Biases in experiment decisions
- 27:17Data roles for decisions
- 30:25Pitfalls in experiment-driven decisions
- 31:08Summary
related talks
29mDecision sciences in Python
01h 03mExperimentation in Data Science
18mHow Data Scientists Broke A/B Testing (And How We Can Fix It)
35mOptimize your schedule with prescriptive analytics
31mFrom stockouts to happy customers: Proven solutions for time series forecasting in retail
01h 30mBayBE: A Bayesian Back End for Experimental Planning in the Low-To-No-Data Regime
30mA physicist's approach to Data Science
50mAnalytics – Balancing ‘Advanced’ and ‘Actionable’