Scheduling a debris-box operation to its real capacity

This entry describes a five-day kaizen held in September 2020 on the debris-box operation that Recology’s Golden Gate and Sunset Scavenger companies run in San Francisco — the roll-off containers dropped at a construction site or a house clear-out, exchanged when full, and pulled when the job is done. The brief was to reduce the rescheduling that was straining customer service and cost; the kaizen’s finding was that the rescheduling was the symptom, and an overbooked schedule the cause. Recology’s people are credited by role rather than by name; the numbers in the results block are the targets the team set on the closing day, not confirmed 90-day outcomes.

A debris box is a large open roll-off container. A customer — a contractor, a landlord, a household mid-renovation — calls or emails a customer-service representative to order one. The representative schedules the delivery; a dispatcher routes it into the day’s work across the fleet; a driver drops it, exchanges it for an empty when it fills, and pulls it when the job is done. The operation runs on work orders, one for each drop, exchange, or pull. Eighteen drivers were each carrying about seven a day.

The problem the managers put to us was rescheduling. There was too much of it, and it was hard to do in the system. Boxes were being scheduled wrong. And the customer-service floor and the dispatch floor had drifted far enough apart that billing disputes were surfacing mid-call, while a dispatcher was still trying to work out with a customer where a box should go. The charter asked us to make rescheduling easier, to cut it from five steps to three, and to reduce the volume of it by a fifth.

The pivot came when the team set the scheduling process next to the fleet’s actual capacity, and the two numbers refused to reconcile. Eighteen drivers at seven work orders each is 126 a day. Add the eight expedites and last-minute orders the operation absorbed as a matter of course, and the fleet could genuinely serve about 134. But the schedule was closing each day at 176 — forty-two more than any driver had the hours to serve, nearly a third over.

The operation knew it. Over time it had built and lettered seven standing contingency plans to absorb the overflow. Drivers took an hour of overtime. Drivers were pulled off the garbage routes. Dispatch skipped customers, pushed work to the night shift, or rescheduled yard work — and, the plan that closed the loop, called customers back to reschedule the box they had been promised that morning.

So the rescheduling the charter set out to reduce was not a defect in a workflow. It was the operation reconciling an overbooked day to a fleet that could never serve it, one apologetic phone call at a time. The boxes were being rescheduled because they should never have been scheduled.

Seen that way, the seven contingency plans were the tell. A contingency is what an operation reaches for on a bad day; these were reached for every day. The exception had become the operating model. A permanently overbooked day was being treated as a fact of the weather — rather than a number the operation was itself writing each morning.

Underneath the overbooking sat the mechanics of how a box got scheduled. Each representative booked into the same coveted early-morning slot, with no cap and no shared view of what the fleet — or the other representatives — had already committed to. Box sizes were estimated by eye, so the wrong box arrived and had to be exchanged. Location and access went uncaptured, so a driver reached an address with nowhere to set the container down, or nothing yet to pull. The single largest of the seven contingencies was exactly that: a driver arriving to find the debris box empty, nine wasted trips a day.

From the middle of the week the team walked the schedule back. The redesign put a ceiling where there had been none.

Cap the schedule, and check the fleet before promising. The number of early-morning boxes any one representative could commit was capped, and the representative was routed through a live check of dispatch availability before a delivery was promised to the customer. The schedule stops closing at 176 because the fleet’s real capacity, not the phone queue, now governs how many boxes can be written.

Widen the delivery window from three hours to four. An hour of additional slack in the promised window gives dispatch room to sequence the day’s work around traffic and exchanges without breaking a promise and triggering a reschedule. It is a small concession to ask of a customer and a large one to give a dispatcher.

Standardize the box-size estimate. A single online box-size calculator, linked to every representative, replaced the by-eye guess, so the box that arrives fits the job and the mis-scheduled exchange never has to happen.

Standardize the work order itself. The box’s location, access, and position are captured in the comments field every time; billing information is taken at the point of scheduling; the customer opts in to a text notification. The driver arrives knowing where the box goes, and the customer is reachable — which takes the trip-and-turn-back and the same-day cancellation out of the fleet’s day.

Simplify the reschedule, and take billing off the dispatch floor. When a reschedule was genuinely needed, its path was cut from a five-step relay between representative, dispatch, and driver to a clean three-step move across the drop, pull, and exchange cases. The team also held itself to a standing rule: the new front-end discipline must not simply shift the load onto Dispatch. Billing was to be resolved and captured upstream at scheduling — on a target of halving the aging on accounts over $5,000 — so that a dispatcher or a driver was no longer relitigating an invoice at the curb.

What the team drew by the Friday was an operation in which the schedule could not be written past what the fleet could serve, the box that arrived fit the job, the driver arrived with what the drop required, and a reschedule — when one was truly needed — was a short, clean transaction rather than a scramble across three floors. The KPIs the team set against that design, to be read at 90 days:

Schedule vs. capacity: Closing at 176 work orders against a real capacity of 134 — a 42-order daily overbooking absorbed by seven contingency plans → the schedule capped to the fleet’s capacity, with a live dispatch-availability check before a box is promised.
Rescheduling: Target: reduced 20% in 90 days.
Mis-scheduling: Target: reduced 50%, via a shared box-size calculator and a standardized work order.
Reschedule steps: Five → three, across the drop, pull, and exchange cases.
Billing: Target: billing-related issues reduced 20% in 90 days; aging on $5,000-plus accounts halved; billing captured at scheduling and taken out of the dispatch-customer conversation.
Delivery window: Three hours → four, giving dispatch routing slack to absorb variability without a reschedule.

The visible brief was rescheduling, and the reschedule path did get shorter. The structural finding was quieter and sat one level up. Scheduling had no capacity ceiling: any representative could promise a box into a day that was already full, because nothing in the process compared what was being promised against what the fleet could do. So the day was written past its capacity, and the overflow came home in the afternoon as reschedules, skipped customers, and overtime — the very costs the charter had named as the problem to solve. No amount of making the reschedule easier would have fixed that, because the reschedule was the reconciliation, not the fault. The leverage was in the rule that governed how many boxes could be promised, and a rule is a cheaper thing to change than a fleet. The operation had been trying to bail the water faster; the week was spent finding the hole.

· · ·