Quoted Against Incurred
Comparing the estimate with the outturn is the point of job costing. What to compare, what the comparison is worth, and the three ways it is usually spoiled.
One job quoted at 48 hours, read four ways at close
What the estimate said, at the rate then in force.
What the job system holds.
Defect work and handover questions, charged nowhere.
Grossed up by the period's unattributed rate.
The job was 27 per cent over on the recorded figure and 48 per cent over on the honest one. Only the first two numbers existed in any system; the other two took an afternoon to construct and changed what the next quote looked like. One employer's own job, not a benchmark.
Comparing what a job was quoted at with what it incurred is the point of job costing. Everything else in the discipline exists so that this comparison can be made honestly. Done properly it improves the next estimate; done carelessly it produces an argument about whose fault the overrun was.
The commercial issue in “Quoted Against Incurred” becomes easier to audit when estimated work, recorded hours and approved changes remain distinct. Teams considering a transparent approach to employee monitoring at tech companies for employee monitoring at tech companies can strengthen the project-time evidence without treating a software report as a substitute for the quote, contract or accepted invoice.
Three things spoil it, and all three are fixable before the comparison is run.
Teams documenting “Quoted Against Incurred” may use IRS business guidance as an independent reference. A useful source should prompt a specific check rather than replace case-by-case judgement.
What to compare
Hours against hours, before money. Money confuses the comparison with rate changes, mix and discounts; hours are the thing the estimate was actually about.
Money comes second, and separately.
Spoiler one: the scope moved
The job that was quoted and the job that was delivered are different jobs, and the difference was agreed informally.
Without a record of scope changes the overrun is attributed to estimating, and the estimator correctly resents it.
The fix
A change log on the job: date, what changed, estimated hours impact, who agreed. Four columns, usually fewer than five rows.
Most of what looks like estimating error is scope with no paperwork.
Spoiler two: the incurred figure is incomplete
Hours after invoicing, rework booked elsewhere, and the unattributed share. All three understate the outturn, and all three are the subject of their own notes here.
Comparing a complete estimate to an incomplete actual flatters estimating and hides the real variance.
Spoiler three: the comparison is used for appraisal
The moment estimates are a performance measure, estimates rise and the comparison becomes a test of padding rather than of accuracy.
An operation that has done this has usually lost two or three years of usable estimating data.
What a good comparison produces
A correction to a class of work, not to a person. "Installations of this type run 20 per cent over; the estimate now includes it."
That sentence is the whole output and it is worth having.
Running it
At close, for every job above a size threshold. Quoted hours, incurred hours, scope changes, difference, one line on the cause.
Five minutes a job and it is the only moment the estimate can still be learned from.
The threshold
Below a certain size the noise exceeds the signal. In most operations jobs under about twenty hours are not worth individually reviewing; they are worth reviewing in aggregate by type.
Aggregate review
For small jobs: average quoted against average incurred by job type, by quarter. The individual comparison is noise; the aggregate is reliable.
This is where most estimating improvement actually comes from, and it takes one query rather than fifty reviews.
The direction, again
Whether jobs run over or under matters more than by how much. A class of work that runs over on four jobs out of five has an estimating problem with a direction and therefore a fix.
Mixed directions of similar size are noise, and chasing them produces nothing.
Feeding it back
Into the estimating basis, with a date. "Type C now estimated at 1.2 times the base figure, from twelve jobs, March."
An adjustment with a count and a date behind it is adopted. An instruction to "be more realistic" is not.
The estimate that was never written down
A surprising share of jobs are quoted at a price with no hours estimate behind it. The comparison is then impossible and the job teaches nothing.
Requiring an hours figure at quote — even a rough one, even a range — is what makes the entire comparison possible, and it costs the estimator a minute.
Estimates by range
Better than a point for anything uncertain, and it changes the review: the question becomes whether the outturn fell inside the range, which is answerable without anybody being wrong.
A class of work where the outturn routinely falls above the range has an estimating problem. One where it falls inside has a usable estimating process, whatever the individual variances look like.
What to do with a job that was wildly over
Review it separately and in detail, because the large ones carry most of the money and most of the learning. The aggregate analysis will average it away.
One detailed review a quarter, on the worst job, usually produces more improvement than the whole routine comparison.
The estimating table
Per job type: number of jobs, average quoted, average incurred, the ratio, and the direction split. Five columns.
It is the table that makes estimating improvable, and it is the reason the rest of this section exists.