Less WIP Without Less Output

Work in process is created by release timing, not by shop-floor speed. A job started three days before its next machine is free does not finish sooner: it becomes three days of inventory standing in a queue. EDGEBIC by User Solutions attacks WIP at the point it is created, by computing start dates from real downstream capacity and by letting operations overlap instead of waiting for whole lots.

This post names the mechanisms that move WIP, shows the arithmetic on documented worked examples, and is honest about the risk you take on when you carry less. For the general treatment of the metric, see WIP management. This post sits under the EDGEBIC results guide.

Where WIP Comes From

Three sources, in rough order of size in a typical job shop:

1. Early release. Material issued before the routing can absorb it. The job exists physically, ties up cash, occupies floor space, and moves nowhere.
2. Full-lot waiting. A downstream station waiting for all 200 pieces when it only needs the first 20. For the duration of the upstream run, the whole lot is WIP.
3. Queue behind contention. Jobs stacked in front of a machine that is genuinely busy. This is the honest kind, and the only one that shrinks by adding capacity or improving sequence.

Scheduling has a direct mechanism for each of the first two, and a partial one for the third. Notice what is missing from this list: shop-floor effort. Nobody on the floor created the WIP; the release decision did.

Mechanism One: Start Late on Purpose

Forward scheduling starts every job as soon as it can and accumulates slack at the end. That is the safe choice, and it is also the expensive one, because all the slack is paid for in material sitting on the floor.

Backward scheduling inverts it. The engine places the last operation to end at the due date, then works upstream in reverse order, right-aligning each step against the deadline its successor imposes. The slack ends up in front of the job.

The documented worked example, 40 valve bodies through a three-step routing:

The order's earliest allowed start (its floor) is Monday July 6. The due date is Friday July 17, and the product carries a two-day tail for outgoing inspection and freight, so all manufacturing must finish by the end of Wednesday July 15.

Forward scheduled, this job starts Monday July 6. Backward scheduled, it starts Thursday July 9. Three days of slack move from behind the job to in front of it, and the job is delivery-ready on the day it is due either way. Those three days are three days the material is not on your floor, not counted in WIP, and not exposed to damage or a design change.

Multiply by every order in your book that currently starts as early as possible for no reason other than habit, and you have the size of the prize.

Mechanism Two: Stop Waiting for the Whole Lot

The second source of WIP is a lot sitting complete at one station because the next station is waiting for all of it.

Lot streaming breaks that. The downstream step starts once a transfer batch has accumulated, using the documented formula: setup, plus the smaller of transfer batch and order quantity, times hours per piece.

The worked case: 100 shafts, half an hour of turning per piece, two hours of setup, a transfer batch of 20 pieces, and half an hour of handling delay. Drilling can start 12.5 hours into a turning run that lasts 52 hours, and the job's elapsed time compresses from 52 hours to about 37. The relevant WIP effect is not the elapsed time: it is that the lot stops piling up as finished-at-turning-and-waiting inventory, because pieces move forward while turning continues.

The lean extreme, a transfer batch of one piece, is documented too: 50 sub-assemblies through solder and test finish in about 13.75 hours instead of 18, with test starting 45 minutes in. That is the minimum-WIP configuration and it demands continuous material handling, which is exactly the trade the software cannot make for you.

Mechanism Three: Do Not Build What You Already Have

The cheapest WIP is the job that never enters the floor. Where a finished item is stocked, EDGEBIC can net demand against on-hand balance and build only the shortfall.

The documented netting case: 80 units on hand, a 60-unit order, and a 30-unit order behind it. The 60-unit order is fully covered from stock, so it never becomes a scheduled job at all, and the ledger records the issue against it. The 30-unit order draws on the remaining 20 and only the balance is manufactured. See how EDGEBIC nets demand against stock for the mechanics.

The WIP arithmetic here is blunt: one order out of two produced zero work in process because it produced zero work.

Mechanism Four: Shorter Queues Through Better Placement

The third source of WIP (contention) does not disappear, but it shrinks when jobs stop queueing behind a nominal machine while an equivalent one sits idle.

Work center groups let a routing step target a pool of interchangeable machines. In the documented three-mill example, when the fastest mill is booked solid until Wednesday, the earliest-completion strategy sends the job to an available slower mill and finishes Monday at 12:30 instead of Wednesday at 10:30. Two days less time as WIP, on a machine that was standing there the whole time.

The optimizer contributes the same way at the sequence level: in the documented three-job case, reordering compressed makespan from 17 working hours to 14. Shorter makespan is directly less time in process, and the optimizer is clamped never worse than the schedule you already have.

The Honest Trade-Off

Less WIP means less protection. This is not a detail: it is the whole reason forward scheduling remains the default.

A backward-scheduled job has zero downstream slack by construction. If the mill runs two hours long, the deburr deadline is immediately at risk. A forward-scheduled job carries its slack after the work, so the same two-hour overrun is absorbed silently.

EDGEBIC handles this by making direction a per-order choice with an automatic safety net. Backward is opt-in, and when the just-in-time plan cannot be honored (the documented infeasible case: a due date that leaves only 16 working hours for 34 hours of work plus a 4-hour queue) the engine rolls the backward attempt back and forward-schedules the whole order instead, recording exactly how late it will be. Under the prompt setting, nothing persists until the planner accepts, adjusts the dates, or cancels. See forward versus backward scheduling for how to choose per product family.

The practical policy most shops land on: backward for stable, well-understood routings with reliable suppliers, forward for anything with process variability or a customer who will not forgive a day.

What the Software Cannot Do Alone

It cannot stop material from being issued. The schedule computes a start date. If your stockroom issues to the floor on the old rhythm, WIP does not move. Release discipline has to follow the schedule, and that is a process change, not a software setting.

It cannot make your handling support small transfer batches. A transfer batch of 20 means someone moves 20 pieces mid-run. The formula is honest about the timing benefit and completely silent about whether your forklift schedule can deliver it.

It cannot carry the risk you removed. Backward scheduling trades inventory for exposure. When a supplier slips on a just-in-time job, the promise is at risk immediately, and the plan being correct is no consolation.

It cannot fix inaccurate on-hand balances. Netting demand against stock only helps if the stock figure is real. A netting decision made against a wrong balance produces a shortage instead of a saving.

It cannot decide your policy. How much protective WIP your plant should carry is a business judgement about variability, customer tolerance, and cash. The software gives you the lever and shows you the arithmetic on both sides.

Want to see how much of your floor is early release? Bring your open jobs and their due dates to a demo, and we will schedule them backward and compare start dates against what you released.