Implementation Engagement · The Capacity Model

A model the GM can read, and a Monday meeting that runs the week

The capacity model behind the engagement: a forecast written in plain numbers, measured labor standards for six activities, one weekly capacity-plan table where overtime gets chosen instead of suffered, and a two-meeting cadence. One workbook. One small refresh script. Nothing else.

Executive framing

The fix was not software. It was a model the GM can read, plus a Monday meeting.

Before the model, the schedule came from last year plus instinct, and overtime absorbed every miss. The model converts client volumes into required labor hours, compares them to the schedule, and forces a decision on every gap on Monday morning, five days before the gap becomes a Friday scramble. The GM owns it end to end.

1 + 1

one structured workbook, one small script that refreshes the forecast from WMS exports. That is the entire system.

50 min

of meetings per week: 30 on Monday to set the plan, 20 on Friday to check it. That is the entire ceremony.

5 days

of schedule visibility for staff, up from day-before and sometimes same-day.

Deliberately low-tech

No platform. No agent. On purpose.

Nothing else in this operation was stable enough to deserve automation yet. Standards were unmeasured, the forecast did not exist, and the weekly decision had no owner. Automating on top of that would have automated the chaos. So the build automates exactly one stable thing, the forecast refresh, and puts everything else where the GM can see it, question it, and change it. When the fundamentals hold, automation can earn its way in. Not before.

new platforms. No workforce-management suite, no new logins, no per-seat licenses.

black boxes. Every number in the plan traces to a measured standard or a client volume.

100%

of the model is legible to the GM without Diazovate in the room. That is the handoff test.

The model, in four parts

Forecast, standards, plan, cadence. Each one earns its place.

Every part exists to answer one operating question. Nothing here is infrastructure for its own sake, and each part is detailed below exactly as it is delivered.

1 · The demand side

A forecast the GM can read

Trailing 8-week baseline by client and activity, seasonality factors, and a one-page promo-intake rule: clients declare peaks two weeks out. Statistical and explainable. Explicitly not a black box.

2 · The supply side

Measured labor standards

Units per labor hour for the six core activities, measured over two weeks of floor observation. The GM's instinct, written down and tested against a stopwatch.

3 · The decision table

The weekly capacity plan

One table: forecast units to required hours to scheduled hours to gap to action. Every gap gets a named action on Monday: planned overtime at chosen crews, a shift swap, or a temp request.

4 · The operating rhythm

The Monday / Friday cadence

Monday, 30 minutes: plan set, published, schedules out five days ahead. Friday, 20 minutes: forecast vs actual, standards drift, next week's risks. The model lives because the meetings run.

Component 1 · The readable forecast

Three inputs, no mystery: history, season, and what clients actually tell you

The forecast is a page of numbers the GM can argue with, not a model output the GM has to trust. Each Monday at 6am the refresh script pulls the prior week's WMS export and rebuilds the baseline. The GM reads it over coffee before the 8am planning meeting.

8 weeks

of trailing volume by client and activity forms the baseline. Long enough to smooth noise, short enough to track the business.

2 weeks

of notice on promo peaks, required by the intake rule every client signed. Surprises move from Thursday to two weeks out.

27% → 11%

weekly volume forecast error, before vs by the end of the first quarter on the model.

Trailing 8-week baseline Average weekly units by client and by activity, with the most recent two weeks weighted heavier. Recalculated every Monday from the WMS export. Any line the GM disagrees with can be overridden, and overrides are logged.

Seasonality factors A small set of multipliers for known patterns: month-end surges from two subscription clients, the pre-holiday ramp, the January returns wave. Written as plain factors on the forecast page, not buried in formulas.

The promo-intake rule One page, agreed with all client brands: promotional peaks are declared two weeks ahead with expected unit ranges. Declared peaks enter the forecast as line items. Undeclared peaks get handled, then show up in the client's quarterly review.

The only code in the engagement is the refresh script: it reads the weekly WMS export, rebuilds the baseline table, and flags any client whose declared promo window opens inside the next two weeks. Roughly a page of code. The GM has rerun it alone since week one.

Component 2 · Measured labor standards

Six activities, two weeks of observation, one number each

Before the engagement, "how long does picking take" lived in three supervisors' heads with three different answers. Two weeks of floor measurement produced one agreed standard per activity. The standards drive every required-hours number in the weekly plan, and drift is checked every Friday.

Activity	Unit measured	Standard (per labor hour)	Observed range	Note from measurement
Pick	Units picked	60	54 - 66	Steady across shifts. Zone layout, not people, drives the spread.
Pack	Units packed	45	41 - 49	Drops on kitted SKUs with fragile inserts. Standard is the blended rate.
Kit	Kits assembled	30	26 - 33	Most variable activity. Re-measured per new kit design.
Receive	Cases received	50	44 - 56	Dock congestion, not labor pace, causes the low end.
Returns	Units processed	12	10 - 14	Slowest and most judgment-heavy. Kept in-house, never rushed.
Ship	Cartons shipped	80	72 - 88	Carrier cutoff times bunch the work. Standard assumes level loading.

Illustrative standards from two weeks of observation across both shifts. The Friday review compares actual units per hour to these numbers; three weeks of drift in the same direction triggers a re-measurement, not a guess.

Component 3 · The weekly capacity plan

The centerpiece: one table, and every gap gets a decision on Monday

This is week 6 of the first quarter on the model, as set at the Monday planning meeting. Forecast units divide by the measured standard to give required hours. Required minus scheduled is the gap. Every gap line carries a named action, chosen Monday, published with the schedule, five days ahead.

Planned OT Shift swap Temp request Hold

Activity	Shift	Forecast units	Standard /hr	Required hours	Scheduled hours	Gap	Action (chosen Monday)
Pick	Shift 1	27,600	60	460	400	-60	Planned OT60 hrs across pick crews A and B: 12 volunteers at 5 hrs each, confirmed in the Monday meeting.
Pick	Shift 2	15,600	60	260	240	-20	Planned OT20 hrs, pick crew C. Same crew that volunteered last promo week.
Pack	Shift 1	14,400	45	320	280	-40	Planned OT30 hrs, pack crew A, Swap in10 hrs from receive, shift 1 (cross-trained).
Pack	Shift 2	9,000	45	200	180	-20	Planned OT20 hrs, pack crew B.
Kit	Shift 1	3,300	30	110	110	0	HoldLevel. No action.
Kit	Shift 2	1,500	30	50	70	+20	Swap out20 hrs to ship, shift 2. Kit crew is cross-trained on the ship lanes.
Receive	Shift 1	5,000	50	100	110	+10	Swap out10 hrs to pack, shift 1.
Receive	Shift 2	2,000	50	40	40	0	HoldLevel. No action.
Returns	Shift 1	960	12	80	56	-24	Temp request24 hrs: 3 temps at 8 hrs, agency booked Monday morning instead of Friday afternoon.
Returns	Shift 2	240	12	20	20	0	HoldLevel. No action.
Ship	Shift 1	33,600	80	420	370	-50	Planned OT50 hrs, ship crew, volunteers confirmed Monday against the carrier cutoff schedule.
Ship	Shift 2	15,200	80	190	170	-20	Swap in20 hrs from kit, shift 2.
Week totals				2,250	2,046	-204	Covered exactly: 180 hrs planned OT + 24 temp hrs. Swaps move 30 hrs and net to zero.

Illustrative volumes against the measured standards above; required hours are forecast units divided by the standard, exact on every row. Scheduled straight time of 2,046 hours sits inside the 2,080 hours that 52 warehouse staff at 40 hours provide, with the difference being approved time off. The week closed at 248 total OT hours: the 180 planned here, plus 68 reactive. Before the model, a week like this ran near 320 OT hours, almost all of it bought on Friday.

Component 4 · The cadence

Two meetings, 50 minutes, and the model stays honest

The workbook does not run the operation. The meetings do. Monday turns the forecast into a published plan; Friday audits the plan against what actually happened. Both are short because the table does the arguing.

Monday · 30 minutes · set the week

Refreshed forecast on the table: baseline, seasonality, declared promos.
Walk the capacity plan: every gap gets an action and a name. Planned OT goes to chosen crews who volunteered, not whoever is standing near the door on Friday.
Plan published to supervisors by 10am. Staff schedules out five days ahead, every week.
Temp requests placed Monday morning, when agencies still have people to send.

Friday · 20 minutes · check the week

Forecast vs actual, by client and activity. Misses over the tolerance get one line of explanation, not blame.
Standards drift: actual units per hour against the standards table. Three weeks of one-direction drift triggers re-measurement.
Next week's risks: undeclared promos, dock schedule conflicts, PTO clusters.
Reactive OT reviewed by cause. The goal is not zero overtime. It is zero surprise overtime.

The boundary

What we did not build, why, and what would change the answer

Each of these was considered and declined, in writing. Each has a stated trigger. The discipline is not "never automate"; it is "automate when the fundamentals have been stable long enough to deserve it."

No workforce-management platform

At 52 warehouse staff, two shifts, and one site, a workbook the GM fully understands beats a platform the GM has to trust. A platform would have added licenses, training, and a vendor between the GM and the numbers, while fixing nothing the workbook does not already fix.

Build it later when: the workbook stops scaling: a second site, a second planner, or a plan that takes more than the Monday half hour to set. The model migrates as-is; the structure is the asset.

No machine-learning forecast

An 8-week baseline plus declared promos cut forecast error from about 27% to about 11%. The biggest error source was never the math; it was clients announcing peaks late, and the intake rule fixed that. A model the GM cannot interrogate would trade legibility for a decimal point.

Build it later when: forecast error plateaus above target for a full quarter with clean promo intake, meaning the readable method has hit its ceiling and complexity would now pay rent.

No automated scheduling

Supervisors carry knowledge no export contains: who is cross-trained, who volunteered last peak, who cannot stay late Thursdays. Auto-generated schedules would spend that trust to save minutes. The plan tells supervisors the hours; people decide the names.

Build it later when: a third shift is added and the swap combinations outgrow what two supervisors can hold in their heads.

Scope

What this engagement is, and what it is not

The definition of done is behavioral, not technical. The engagement ends when the operation runs the model without Diazovate, not when the files are handed over.

Included

The four-part model, built and live: readable forecast, measured standards for six activities, the weekly capacity plan, the Monday / Friday cadence.
Two weeks of labor-standards measurement on the floor, both shifts.
The forecast refresh script, documented so the GM can rerun and adjust it.
The one-page promo-intake rule, drafted for client signatures.
Four weeks of coached cadence, then four weeks observed from a distance.

Not included

No workforce-management platform, no machine-learning forecast, no automated scheduling. Declined in writing, with triggers, above.
No WMS replacement or configuration work. The WMS stays as it is; the model only reads its exports.
No headcount recommendations. The model prices gaps; staffing levels stay a GM decision.

Client inputs

Weekly WMS export access for volumes by client and activity.
Timeclock data for scheduled and actual hours by shift.
Two weeks of floor access for standards observation, both shifts.
Promo-intake agreement signed by the client brands.

Definition of done

The GM runs Monday planning and Friday review for four consecutive weeks without Diazovate in the room.
Schedules published five days ahead, every week, no exceptions.
Every gap in every published plan carries a named action chosen on Monday.

Pricing: Scoped per engagement.