The Estimate Nobody Grades
A job ships. The truck pulls out, the invoice goes, and the folder closes. What almost never happens next is the one step that would make the next quote sharper: somebody laying the estimate next to what the job actually cost and reading the gap out loud.
Most shops quote well enough to win work and account well enough to file taxes. The part in the middle, where the two numbers meet and someone learns something, is missing. Quoting looks forward at what a job should cost. Job costing looks backward at what it did. The publication TheFabricator draws exactly that line, and it makes a plain point that lands hard in a small shop: when you never compare actual against estimate, a bad quote repeats forever. Not once. Forever, until someone notices, and nobody's looking.
The error you can't feel
Here's what makes this so easy to miss. A single blown quote doesn't announce itself. You bid a bracket assembly, you set the cut and form time a little light, and the job still ships. It might even show a profit, because a fat job somewhere else in the mix covered the thin one. Nothing on your P&L points a finger at the bracket. The money leaked, but it leaked into an average, and averages keep secrets.
Now run that same bracket 400 times a year for 3 years. The little miss you never measured is now a real number, and you've quoted every one of those jobs off the same wrong standard. You didn't make one mistake. You made one mistake and then repeated it 1200 times because the loop that would have caught it was never closed.
The mechanism is boring, which is why it survives. Every job absorbs a slice of your overhead based on the time it's supposed to take. TheFabricator's framing is worth sitting with: a job that runs faster than estimated absorbs less overhead and hands you extra profit, while one that runs long absorbs more and eats the margin you thought you'd priced in. Get the time standard wrong and you get the overhead wrong on every unit that carries it. The quote isn't off by the labor gap alone. It's off by the labor gap plus everything the labor gap was supposed to be dragging along with it.
Close the loop in fifteen minutes
You don't fix this with software. Plenty of shops buy the ERP and still never read the variance report it spits out, which is its own kind of expensive. You fix it by making the comparison a habit somebody owns.
The lightest version I've seen work is a fifteen-minute standup on Friday. Pull the week's finished jobs. For each one, mark it green, yellow, or red against the estimate. Green shipped near the number. Yellow drifted. Red missed badly enough to explain. You don't need to autopsy the greens. You spend the time on the reds and the pattern of yellows, and when a standard is clearly wrong you change it that afternoon, not next quarter. TheFabricator recommends close to this exact rhythm, and the reason it works is that it moves the correction to within days of the job instead of leaving it to a memory nobody actually has.
Watch the small stuff too, because the small stuff is where shops bleed without a wound to point at. Abrasive discs, anti-spatter, contact tips, gas. Each one is nothing per job and thousands per year, and if you're not loading consumables into overhead and pricing them by the machine-hour, they're coming straight out of a margin you already promised the customer. TheFabricator flags those untracked consumables specifically, and they're the kind of cost that never shows up until you go looking.
What you're actually buying
The point of grading estimates is not to catch the estimator being wrong. It's to make the next quote a little less of a guess than the last one. A shop that reads its variances every week is quietly compounding accuracy. A shop that doesn't is bidding in year three off the same blind spots it had in year one, and calling the result experience.
So start Friday. Pull last week's jobs, put the estimate next to the actual, and find the one standard that's been quietly wrong the longest. That's the quote you've repeated the most times, which makes it the most expensive number in the building, and until this week nobody had ever checked it.