Insights

Where your shop's margin really leaks: the numbers to watch

A former CRA auditor on the handful of numbers that decide whether an owner-run shop makes money: quoted versus realized margin, throughput at the constraint, first-pass yield, and more.

---

Here is something I see in almost every owner-run shop I walk into. The bank balance is fine, the backlog is full, the machines are busy, and the owner still cannot tell me which jobs made money last year and which ones quietly did not. That is not carelessness. A light manufacturing business is a project business wearing a production costume. Every job is quoted, scheduled, built, and shipped as its own small project, and the year-end statement is just the sum of hundreds of those projects blended together. The blend hides everything.

I spent years as an auditor at the Canada Revenue Agency, reading how businesses actually work versus how their owners described them. Now I run an operations firm and hold a Lean Six Sigma Black Belt. I do not claim to have run a shop floor. What I do is make a shop's own numbers tell the owner the truth, and then help fix what they reveal. In fabrication, machining, cabinetry, and food processing, that truth almost always lives in the same handful of numbers.

The job that looked profitable and wasn't

Picture a job quoted at a healthy thirty percent gross margin. By the time it ships, the shop realized twelve. Nothing dramatic happened. The estimate under-counted the finishing hours, so the welding took longer than the sheet said. The constraint was overloaded that week, so the job ran late and went to overtime. And a few parts came back wrong and got fixed on a time card nobody flagged as rework. Three small leaks, and the owner never sees them: the invoice was paid and the customer was happy.

Multiply that across a year and you get the single most useful idea in this whole business: quoted margin versus realized margin. The gap between what you priced and what you kept is where your profit actually lives or dies. Most shops never measure it, because measuring it means closing the loop between the quote and the actual hours, and that loop is exactly what a spreadsheet-and-QuickBooks setup leaves open.

Estimating accuracy is the number under all the others

If you only start watching one number, make it this one: actual hours divided by quoted hours, broken out by operation. Not the blended job total, the operation. Cutting, welding, finishing, setup, install. This is the master variable, because a shop can do everything else well and still lose money if it quotes wrong. And most gut-priced shops quote wrong in the same predictable place. They under-count the unglamorous back end: the finishing, the deburring, the setup, the punch list. On stainless work, finishing can take nearly as long as the welding itself, and it is the line nobody remembers to price.

The fix is not a fancier quoting tool. It is a feedback habit. Every finished job updates the estimating model, so the standard hours you quote next month are calibrated to what the last job actually consumed. In a shop that has priced from the owner's head for twenty years, closing that loop alone can recover several points of margin.

Protect the constraint, and ignore the rest

Every shop has one machine or one skilled person that every job has to pass through. The welding booth. The CNC spindle. The nesting router. The one certified operator. That is your constraint, and it sets the ceiling on what the whole business can produce. An hour lost at the constraint is an hour lost to the entire company. An hour saved anywhere else is a mirage.

So the utilization number that matters is not the average across all your machines. Keeping every machine busy looks efficient and usually just builds a pile of inventory in front of the bottleneck. What matters is throughput at the constraint alone, and the honest way to read it is OEE on that one asset: availability times performance times quality. One number, one machine. If you cannot tell me how many hours of backlog are sitting in front of your constraint right now, you are not scheduling it, you are reacting to it.

One caution on that number. The world-class OEE figure of eighty-five percent was defined for discrete, repetitive lines making the same part all day. A high-mix job shop with constant changeovers legitimately runs forty to fifty-five percent, and sixty-five is world-class for that model, so track your own trend rather than someone else's benchmark.

First-pass yield, and the cost of rework nobody counts

First-pass yield is the percentage of parts you build right the first time, with no rework and no scrap. It is one of the most honest numbers in the building, and in some shops it is literally the margin number. In food processing, a defect usually cannot be reworked because rework would violate food safety, so a bad batch is scrapped outright, which means first-pass yield and gross margin move together almost point for point. Everywhere else, low first-pass yield is a hidden second payment: you pay once to build the part, then again to fix it.

Owners rarely track this because the second payment is invisible. The welder who fixes his own mistake just keeps welding on the same time card, so the cost never shows up as a line item and never gets managed. Ask a shop for its scrap and rework rate and you usually get a blank look, then a guess. The fix is to convert scrap plus rework to dollars, because simply putting a number on it changes behaviour before any improvement project starts. In a year where steel, aluminum, and resin prices swing hard, every scrapped part is margin you paid full price for and threw in the bin.

On-time delivery, measured against the date you first promised

On-time delivery sounds simple until you ask which date it is measured against. Many shops quietly measure against the revised date, the one they slipped to when the job ran late. Measuring against the date you slipped to is measuring nothing. The number that means something is delivery against the original promised date, because that is the promise the customer remembers. When it is soft, the cause is usually upstream: promise dates set by sales to win the order rather than to match the hours the constraint actually has. Every over-promise turns into overtime, an expedite, or a late shipment that costs you a customer.

WIP aging and setup time, the two quiet reads

Two more numbers round out the picture, and both are easy to watch. Work-in-process aging is how long jobs sit on the floor between operations. WIP that ages is cash tied up in half-built work, and it hides problems, because a job stalled for three weeks has something wrong that nobody has surfaced. Watch how long the average job sits, and the oldest ones especially.

Setup time is the other, and on low-volume work it is where the hours actually go. A shop that measures cycle time but not setup is optimizing the wrong forty percent. If you run twenty setups a week at forty-five minutes each, cutting that in half through simple staging and quick-change habits gives you back a large block of constraint capacity every year without buying a single machine. In a year of expensive financing, capacity you unlock from equipment you already own is the highest-return move on the table.

Where to start

You do not need all of this at once, and you certainly do not need new software to begin. Start by closing one loop: pick your last ten jobs, compare actual hours to quoted hours by operation, and see where the estimate broke. That single exercise usually tells an owner more than a year of financial statements, because it turns the blended average back into the individual projects it hid. From there, protect the constraint and make scrap visible.

If you want a straight, outside read on where your margin is actually leaking, our operations health check traces it through your own numbers and shows you where to start. You are always welcome to reach out and talk it through, with no obligation.

Frequently asked questions

My backlog is full. Doesn't that mean the shop is healthy? Not on its own. A full backlog of underpriced work just accelerates the losses. Backlog is only good news at a known margin, so the real question is not how much work you have booked, but what margin you will realize on it.

Why measure OEE only on the constraint and not every machine? Because the constraint sets the ceiling for the whole shop. An hour gained anywhere else just builds inventory in front of the bottleneck. Watching one number on the one machine every job passes through keeps you focused on the throughput that actually pays.

We are a small shop with no ERP. Can we still track these numbers? Yes. A short, honest look at your last ten jobs on a spreadsheet, comparing quoted hours to actual hours by operation, will surface more than most software does. The barrier is the habit of closing the loop, not the tool.