A 0.2-mile wall with no system
A Southeast distribution hub processes roughly 2,000 cartons per day for white-glove furniture delivery. Before cartons can be staged for their delivery trucks, they're temporarily stored along a 1,056-foot structure called the Wall — built right to left, with cartons placed at random regardless of which delivery door they belong to.
The result: floor associates searching the entire length of the Wall for each carton before staging it. Cartons get lost, damaged, or missed — causing the hub to reschedule 5% of orders daily, at an average cost of $376 per cancellation. From October 2022 to January 2023 alone, 23,529 cancellations totaled over $8.8M in losses. Of those, 52% were directly caused by associates being unable to locate a carton on the Wall.
From random chaos to five zones
The core insight was simple: if cartons are placed near the delivery doors they belong to, associates don't have to search. Drop Zones are designated sections of the Wall, each corresponding to a group of outbound delivery doors. Cartons are tasked as they're unloaded — not after the whole Wall is built — so every carton lands in its zone.
Three models, 30 runs each
The team built three full-scale Simio discrete-event simulation models of Hub processes, each run 30 times to generate statistically robust output. Model 0 replicated the current system and was validated against actual process times observed during an on-site time study. Models 1 and 2 tested 9 and 5 Drop Zone configurations respectively.
On-site time study & validation
Conducted a time study at the facility to measure actual process durations. Model 0's output (Hub Process 1: ~15 hrs, Tasking: ~2 hrs, Hub Process 2: ~9 hrs, Total: ~26 hrs) closely matched observed values, validating it as the baseline for comparison.
Triangular distributions
Given limited sample data, process times were modeled using triangular distributions parameterized from the time study. Each carton also carried a 1.5% probability of loss both in transit and on the Wall — calibrated from real cancellation data.
Paired t-tests at α = 0.05
All model comparisons used paired t-tests with a significance level of 0.05, generating 95% confidence intervals for each metric. Model 0 served as the true population mean baseline. Null hypotheses of no difference were rejected for all key metrics.
Model selection: why 5 zones?
Models 1 (9 zones) and 2 (5 zones) produced statistically equivalent results for Hub Process 1 (p > 0.05 for all metrics). Hub Process 2 was only 4.9% slower with 5 zones — a negligible tradeoff for a configuration that is significantly simpler to implement and train.
Before vs. after, by the numbers
All metrics averaged across 30 simulation runs per model. The 50% reduction in total process time was confirmed at 95% confidence (CI: [7.28, 7.36] hours decrease).
Before finalizing the recommendation, the client ran a small live test: one associate transported 4 cartons randomly from the Wall in 20 minutes (current system). A second, selecting cartons placed near their delivery doors (simulating a Drop Zone), staged 6 cartons in the same 20 minutes — a 50% productivity increase that directly corroborated the simulation results.
Put-Away Assignment Tool (PAAT)
Cancelled orders don't disappear — their cartons get put away into Hub inventory racks to await rescheduled delivery. Without a standard process, this put-away was disorganized, causing overflow and wasted rack space. A teammate developed PAAT: a two-button Excel/VBA GUI implementing a greedy algorithm to assign each carton to the optimal shelf position based on its dimensions.
The algorithm ranks cartons by volume (largest first) and attempts to fit each into 4-ft-deep shelf positions before falling back to 8-ft-deep positions, minimizing overflow to shelves A and B. Tested across 30 batches of 50 randomized cartons, PAAT saved an average of 1,079 ft³ of wasted space per batch — a 17.6% increase in rack utilization versus the first-fit baseline.
From operational chaos to quantified savings
The 9-zone model was technically marginally better than 5 zones — but not significantly so. The paired t-test between Model 1 and Model 2 failed to reject the null hypothesis for all Hub Process 1 metrics (p > 0.05), and Hub Process 2 was only 4.9% slower with 5 zones. Recommending the simpler configuration wasn't a compromise — it was the right engineering decision, prioritizing real-world implementability within the client's one-week training constraint.