Estimate versus actuals: compare per operation

What does the order total hide?

A worked example. A subcontractor with twelve people estimates a batch at 90 hours: 40 hours of turning, 35 of milling and 15 of deburring. After delivery, 100 hours are booked, just over 11 per cent more. The cost engineer raises the standards for this part by 11 per cent and moves on.

Operation by operation, the picture changes. Turning was spot on. Milling ran from 35 to 52 hours, almost half as much again. Deburring was estimated at 15 hours and took 8, so nearly half of that line was slack. In the total, the two deviations largely cancel each other out.

That sends all three standards the wrong way. Turning goes up although it was right, deburring goes up although it was already too generous, and milling, which was nearly half too tight, moves by only 11 per cent. The next estimate is no better for it, only different.

How do pre-calculation and post-calculation differ?

The pre-calculation works out in advance what a part should cost: material, setup time, programming time, cycle time and finishing, based on the drawing and the route you expect. It is the basis for the quote, not the selling price itself. Many shops simply call it the estimate.

The post-calculation sets reality beside it after delivery: booked hours, material used, scrap and outside costs. Accountants call it job costing or actual costing. An estimate without actuals repeats the same assumption order after order; actuals without an estimate are an archive with nothing to measure against.

The value lies in the difference and in where it comes from. Management accountants call this variance analysis: checking, line by line, whether the time, the hourly rate or the material was off.

Cost accounting splits a labour variance in two, and the split works for machine hours as well. The efficiency variance is the hours the estimate allowed for the parts actually made, minus the hours booked, times the estimate’s rate. The rate variance is the estimate’s rate minus the actual rate, times the hours booked. Material splits the same way, into price and usage. This article is about the efficiency variance: the time per operation.

How do you set booked hours beside the estimate?

It starts with the records. In many shops actual times live on job cards and production sheets, written by hand and often without the machine named. ‘About an hour and a half’ is no basis for correcting a standard. Production actuals only become usable when every time belongs to one operation, one machine and one date.

After that, the comparison itself is simple. Every estimate line gets its booked counterpart: turning beside turning, milling beside milling, deburring beside deburring. The difference per line is the number that says something about the standard. Keep the order total alongside to check the sum, not as the result.

Book the time of an operation in elements, each with its own frequency. Setup and first article inspection come once per batch; loading, unloading and run time with every part; in-process inspection with every part or every few parts. Divide a per-batch element by the quantity and multiply a per-part element by it. Then you can trace which element was off.

Be clear about what you are comparing. A booked time also holds waiting, measuring and changeovers; the run time the control records holds only the NC program. Why those times drift apart is the subject of four cycle times for the same part.

1. Book hours per operation, with machine and date.
2. Close each record as soon as the operation is finished, not the next morning.
3. Book setup time separately from the run time of the batch.
4. Set the booked hours of each operation beside its line in the estimate you quoted.
5. Note what happened with every large deviation, even in a few words.

Is every deviation an estimating error?

No. Actual is not the same as representative. A booked time can contain a broken drill, a breakdown, an operator stepping in, or a record that was only closed the next morning. Treat every actual as the truth and the next estimate gets worse, not better.

A 2025 study in Procedia CIRP on anomalous cycle times on CNC machines starts by filtering: server issues, maintenance and shift changes come out of the measured program durations before anything is calculated. The same discipline belongs in every post-calculation, even one kept in a spreadsheet.

Subduxion distinguishes three kinds of deviation. Process variation is ordinary scatter: the operation took a little longer or shorter, as it does on every batch. An explained exception has a cause you can point to: tool breakage, a breakdown, an unclosed record. With an unexplained deviation, nobody can say where the time went. Process variation belongs in the standard, an explained exception is marked and left out, and an unexplained deviation is investigated before it counts.

Throwing away every outlier is not right either. If the same tool breaks on every batch, the problem lies in the process choice, and the time belongs in the estimate after all. That means every record has to show which tool, material and program the time was made with.

When should you adjust a standard?

One order running over is not yet a reason to change a standard. Subduxion’s guide treats this as the representative signal: three consecutive orders with the same milling overrun. Then you adjust the standard of that one operation, milling in this example, and leave the rest alone.

Fix that rule before the figures come in: per operation, a threshold for the deviation and the number of consecutive batches that must exceed it. Decide case by case and you end up adjusting after whichever order stood out most. There is no need to wait when the process changes. If the operation gets a different program, tool, fixture, machine or material, revise the standard before the next quote, because the old time belongs to a different process.

Turning was right, so the turning standard stays. Deburring was too generous, and that standard too changes only once the pattern returns. Adjusting per operation like this is calibration: the estimating method moves only where measured times show it repeatedly.

A repeat part is where you notice the difference first. The next request arrives and the question is whether last time’s price still holds. With a post-calculation per operation you know which line to change and why; without that comparison, you adjust on instinct. If you carry the last calculation over, recalculate it at today’s rates: a correct time at an old rate still gives an outdated cost price.

What you adjust is the cost price. The selling price is a separate decision, with margin and risk allowance on top, and does not follow automatically. An operation that turns out structurally dearer calls for a higher price or a different route. You decide that part by part.

What does the estimate need to contain?

This only works when the estimate itself consists of lines per operation, and that is exactly where the ERP and the spreadsheet fall short. The ERP keeps the hours booked per order, not the estimate line the price was decided on; the spreadsheet keeps one cost, or a piece time that folds turning, milling and deburring into one figure. An owner with ten years of actuals in a filing cabinet then has the hours, but not the line they belong beside.

That costs money on the next quote. Raise the part from the worked example by 11 per cent across the board and the milling line stays well short of the 52 hours milling took, while the deburring line moves further above the 8 hours deburring took. On every new batch you give margin away on milling, and on deburring you are dearer than you need to be, in a quote the customer sets beside others.

Each line needs at least three things: time, rate and quantity. Keep setup time apart from cycle time, because setting up happens once per batch and cycle time counts per part. How to count setups, and what each one costs in setup time, is covered in counting setups.

Finally, compare against the quote version that went out, not against an estimate you updated later. Only the version you sent records what you promised, against which drawing revision and with which inputs. Start with the next estimate: one line per operation, and the same breakdown on the job card.

Five habits that make a post-calculation useless

A post-calculation can add up and still tell you nothing. These five habits are why.

Avoid them and you need no new system to begin. A spreadsheet with one line per operation and honestly booked hours is enough for the first comparison.

• Comparing only the order total, so that a generous and a tight operation cancel out.
• Booking hours to the order instead of to the operation.
• Treating every actual as the truth, including the one with a broken drill in it.
• Adjusting the whole part after one overrun, instead of one standard after a repeated signal.
• Letting the selling price move automatically with the cost price.