How do you find the bottleneck on a shipping dock?

Pandemic demand had orders arriving faster than the dock could ship them, and every obvious fix cost money: another crew, more people per crew, more lift trucks. Before leadership paid for any of them, it wanted two questions answered.

Which resource limits how many trailers ship in a day? Add to the wrong one and the number stays where it was.

Does the loading sequence matter? The crew could alternate trailers between the two product lines, load all of one line before the other, or mix the two.

The company's logistics advisors brought in SimWell to answer both with a model of the future-state dock that the client's own planners could run.

Loading a trailer is a chain of steps. Metal parts come out of paint, plastic parts arrive by trailer from the second plant, everything waits in staging, and the crew carries or lifts each part into a trailer that gets harder to fill as it fills. Any step in that chain can hold up the rest, and which one does shifts through the day. An order heavy with oversized parts, giant corkscrew slides among them, keeps the lift truck busy while the crew waits on it. A nearly full trailer reverses the picture: the crew hunts for space and the lift truck sits. A long paint run leaves staging thin, and then nobody has anything to load.

A time study can say how long each step takes. The study cannot say which step is holding up the dock on a given afternoon, or what another crew or another lift truck would buy.

SimWell built a hybrid agent-based and discrete-event model of the shipping area in AnyLogic. Trailers and orders are agents with state charts: a trailer cannot move from Loading to Loaded until every order on it is loaded, and an order cannot reach Loaded until every one of its parts is on the trailer. The logic reads the way the dock works.

The model shipped as a standalone application the client runs on its own computers, with every operating assumption exposed as an input: the target trailers per day, part quantities and sizes by product line, which parts need a lift truck, process times for painting, for picking up oversized parts, and for finding a place in a full trailer, the number of crews on a shift, and the people and lift trucks in each crew. Planners with no simulation background change a number, run it, and read the result.

On sequence, the answer was no answer, which was the useful part. Monte Carlo runs over every loading pattern showed no effect on trailers shipped per day. Management could set the sequence on other grounds, delivery dates or crew preference, without worrying about throughput.

On capacity, the runs that loaded only one product line told the story. If every order were from the faster line, the dock would ship 40% more trailers. When the two lines' load times went up on the screen in the executive readout, the presentation stopped for ten minutes while leadership worked out why. The faster line ships most of its parts in a crate, so a lift truck sets the whole crate in the trailer. The slower line is hand-carried, one part at a time.

Why not crate both? A phone call to a product representative answered it: the faster line arranges for a lift truck to be at the customer's park or school to unload the crates. The slower line does not. The 40% was sitting behind a scheduling habit at the delivery end, not a limit on the dock.

Loading sequence has no effect on trailers per day, so the client sets it on other grounds. The product-line mix drives throughput: crated loading ships 40% more trailers than hand-carried loading with the same crew and the same lift trucks. Leadership's plan came out of the readout itself: arrange lift trucks at customer sites for the second product line so its parts can ship crated too, a change made at the delivery end rather than on the dock.

The model stayed with the client as a standalone tool, so the next question about the dock gets answered without SimWell in the room.