The Estimate That Was Right for the Wrong Reasons
Why this matters
The most dangerous job type in a shop is not the one that misses badly. It is the one that hits the total every quarter while both halves of it are wrong in opposite directions. Total variance near zero buys a job type permanent immunity from review, because every threshold, every exception report, and every instinct is pointed at the outliers. Meanwhile the offset is holding together on a condition nobody chose and nobody is watching, and the day that condition changes the margin drops without a single new thing going wrong. This is a walkthrough of exactly that failure on an equipment replacement type, from the first signal to the confirmed cause, including the three explanations that were tested and discarded on the way.
The first signal
A shop noticed that realized margin on its equipment replacement type fell in the fifth quarter of tracking. Against a quoted margin of 45.9% of price, the type had realized 45.9% for four straight quarters, then landed at 39.5%. That is a drop of 6.4 percentage points of margin, on the type carrying about half the shop's estimated labor hours.
One thing had visibly changed that quarter: the veteran install crew had been split up to seed a second crew, and roughly half the replacements in the quarter were run by the new pairing.
The owner's conclusion took about four seconds, and it was wrong.
What the report said, and why the report was not lying
Every quarter this type had been reviewed by total variance against an investigate-above-10% threshold, and every quarter it came in at a median of about 0% total variance. It never once appeared on an exception list. Four quarters of clean data.
Indexing estimated total cost to 1.00, the type's estimate broke down as labor 0.60 and parts 0.40, and price was set at 1.85 times estimated cost, giving the 45.9% quoted margin from 0.85 / 1.85.
The report was accurate. It was also aggregated at exactly the level that made the problem invisible, which is the actual lesson of this case: a threshold applied only to the total cannot see two component errors that cancel.
Wrong diagnosis 1: the new crew is slow
The obvious one, and the one the timing supported. Test it by comparing the new crew's hours against the estimate, not against the old crew.
The template estimated 12.0 labor hours per replacement. The new crew's jobs came in at a median of 12.1 hours. Measured against the shop's own published standard, the new crew was performing to spec, within a rounding difference.
Then the veteran crew's history was pulled for the same type over the previous four quarters: a median of 9.6 actual hours against the same 12.0 estimated, so actual labor ran 20% below the estimate, or put the other way, the estimate sat 25% above what that crew actually consumed.
That reframes the whole picture. The new crew did not cause anything. They simply stopped doing something the shop did not know was being done.
Wrong diagnosis 2: supplier prices went up
The second explanation is always available and usually partly true, so it needs a real test rather than agreement. Pull the actual paid invoices for the major components on this type and compare paid cost against the same quarter a year earlier.
Paid parts cost was up about 4% year over year. Real, but nowhere near large enough to matter here, because the gap being explained is roughly 30% between what the template estimated for parts and what the jobs actually consumed. A 4% market movement cannot produce a 30% gap. Back the market movement out multiplicatively rather than subtracting it, since these compound: 1.30 divided by 1.04 is about 1.25, so roughly 5 points are market and about 25 are something else.
The important consequence: the gap predates the price increases. It was there in quarter one.
Wrong diagnosis 3: somebody is discounting
Cheap to test and worth testing, because a quiet discounting habit produces exactly this margin shape. Compare quoted amount to invoiced amount across the quarter's jobs of this type.
Invoiced landed within 1% of quoted on average. No meaningful concession pattern. Discounting is out.
Opening the estimate line by line
With all three easy explanations discarded, the only remaining move is to stop looking at the type and start looking at a job. Take five recently completed replacements and rebuild each one from source records: the supplier invoices, the parts pulled from the truck and the shelf, the time entries, the disposal receipts.
The template's parts line covered the major component plus six line items. The teardown of the five jobs found eleven line items actually consumed on every one of them. Five items were being used on essentially every job of this type and appeared nowhere in the estimate: a component that became mandatory at a code revision after the template was written, two categories of fittings and consumables that had never been rebuilt since the template was created, disposal of the removed equipment, and a small quantity of material that had always been treated as shop stock.
None of these are exotic. Each one on its own is small enough that no tech would flag it. Together they account for most of the roughly 30% by which actual parts cost exceeded the estimated parts cost.
The offset, quantified
Run the two component errors through the index and the reason four quarters looked perfect falls out.
Estimated: labor 0.60, parts 0.40, total 1.00.
Veteran crew actual: labor 0.60 x 0.80 = 0.48, because labor ran 20% below the estimate. Parts 0.40 x 1.30 = 0.52, because parts ran 30% above the estimate. Actual total cost: 0.48 + 0.52 = 1.00.
Zero total variance, produced by a 20% labor underrun paying for a 30% parts overrun. Realized margin against the price of 1.85 was (1.85 - 1.00) / 1.85 = 45.9%, identical to quoted, for four consecutive quarters.
New crew actual: labor 0.60 x 1.00 = 0.60, because they ran to the estimate. Parts unchanged at 0.52. Actual total cost: 1.12, or 12% over estimate. Realized margin: (1.85 - 1.12) / 1.85 = 0.73 / 1.85 = 39.5%.
The 6.4 percentage points of margin that disappeared were never earned by the estimating template. They were being donated by one crew's speed, and the donation ended when the crew was split.
Why it held for four quarters and then stopped
The offset was never a stable state. It required one crew to beat one line by an amount that happened to match another line's shortfall, on a mix that stayed constant. Any of four things would have broken it: splitting the crew, which is what happened; a larger supplier increase, which would have pushed parts past what the labor underrun could cover; a shift toward jobs with a heavier parts share; or the veteran crew getting faster still, which would have made the type look better than quoted and probably triggered a price cut on a job type that was already thin.
That last branch is worth sitting with. The same masked error, broken in the other direction, would have produced a confident decision to bid the type more aggressively. A mask does not fail safely. It fails in whichever direction the underlying conditions move.
The correction, and why it had to be two of them
Parts. Rebuild the materials list from the teardown, adding the five missing items, and raise the template's parts line by the measured 1.30. That moves estimated total cost from 1.00 to 0.60 + 0.52 = 1.12 on the index.
Price. Holding the 45.9% target margin against the corrected cost means pricing at 1.12 x 1.85 = 2.07 on the index, which is a 12% increase on this job type. That number is uncomfortable, and it is the honest one. The type had been priced 12% below what its true cost basis required to hold the quoted margin, since the template was written. Be precise about what that does and does not mean: the work was still profitable, earning a realized 39.5% against the 45.9% sold. It was not being sold below cost. It was being sold at a margin nobody had chosen, which is a quieter problem and the reason it survived so long.
Labor, deliberately left alone. The tempting third move is to cut the labor line to 9.6 hours to match the veteran crew. Do not. Twelve hours is what the shop's current crew mix actually consumes, which is exactly what an estimate is supposed to hold. Cutting to a number only the best crew achieves converts a solved problem into a structural underbid on every job the second crew touches. Re-measure it quarterly instead, because if the new crew trains up to 10 hours the line genuinely should come down then, on evidence, rather than now, on nostalgia.
The process fix. Set the variance threshold on labor and parts as separate tests rather than on the total. A type may pass the total and still have both components outside their own limits, and that combination is precisely the one worth investigating, because it is the only one that hides.
What would have changed the conclusion
If the new crew had come in at 15.0 hours against the 12.0 estimate, 25% over, there would be two genuine problems rather than one, and the training explanation would have been partly correct. The test that separates them is always the same: compare against the estimate, never against the previous crew.
If paid parts cost had been up 25% year over year rather than about 4%, the market explanation would have survived and the correction would be a repricing cycle plus a supplier conversation, not a rebuilt materials list.
If this type were sold time and materials rather than fixed price, the parts miss would have billed straight through to the customer and never appeared in margin at all. It would have surfaced as invoices landing well above quotes and customers pushing back, which is the same defect wearing a completely different symptom.
If the shop had been reviewing labor hours against the estimate per job all along, as a standing habit rather than as an exception triggered by total variance, this would have been caught in quarter one, when the veteran crew's 9.6 hours against a 12.0 estimate should have raised the question of why an estimate that generous was still producing exactly the quoted margin.
References
- U.S. Small Business Administration (SBA), financial analysis and cost control guidance for small business
- Trade-standard practice for component-level estimate variance analysis
- See related: The Variance Threshold Worth Investigating, How to Read Your Own Job Costing Data, How to Separate Estimating Error From Execution Error, How to Adjust an Estimate Template From Real Data