Change Order Approval Rate and Why a High One Is Not Good News

Why this matters

Almost every shop reads a high change order approval rate as proof the office is writing good changes and customers trust the price. Read the composition and the intuition inverts. The rate is computed only over changes the shop chose to put in writing, and it is decided by customers who have already let a crew into the building. A figure in the mid-nineties is usually telling you either that your scope is systematically short and customers have no real alternative once work has started, or that you only raise the ones you are sure of and eat everything else. Neither is a compliment, and one of the two is costing hours nobody is counting.

What the rate is made of

Numerator: change orders approved. Denominator: change orders raised in the window. Rejections are counted separately, which sounds like bookkeeping and is the whole reason the number is fragile - the denominator is not the population of changes that happened on your jobs, it is the population of changes somebody wrote up.

Four things sit outside that denominator, and each one moves the rate in the flattering direction:

• The change nobody wrote. A crew finds extra work, does it, and the office never raises a change order. It is not a rejection. It is not anything. It leaves the rate untouched and shows up only as hours.
• The change that was talked about and withdrawn before a formal answer. If withdrawals are filed as rejections the rate drops; if they vanish the rate rises. Pick one convention and know which you picked.
• The superseded version. One change revised twice can sit in the denominator three times. One approval against three rows reads as 33 percent on a change that was approved.
• The pending one. A change raised in the last fortnight of the window may have no answer yet. Counting undecided rows as not-approved understates the rate, and dropping them from the denominator is right only if you also drop them from any count of changes raised. This is ordinary cohort truncation, which the estimate conversion card owns; count only changes old enough to have been answered.

The time anchor is the raise date, not the decision date, which means a window that ends today always contains its least-resolved cases at the edge.

One floor before reading anything: do not read an approval rate off fewer than about 15 decided change orders, because at 15 a single rejection moves the figure about 7 points. That floor is stated on decided change orders, which is a different unit of analysis from the phase-instance floor in the phase completion card; do not carry one across to the other.

Why a high rate is not a compliment

Two explanations produce the same high figure, and they call for opposite responses.

Explanation one: the scope was short and the customer had no choice. The estimate missed work that was always going to be needed. The crew is on site, the wall is open, the yard is dug. A customer in that position approves because the alternative is a half-finished job, not because the change was well made. Approval here measures leverage, not agreement, and it usually comes with a quieter cost: the second and third change on the same job get approved with visibly less goodwill, and the review lands in whatever the customer says about you afterwards.

Explanation two: the shop only raises the safe ones. Somebody in the office has learned which changes get an argument and which do not, and quietly stopped raising the second kind. Everything else is absorbed. The rate is high because the population is curated, and a curated rate is not a rate. This one is invisible on the change order records by construction, because the evidence of it is the change orders that were never written.

The approval rate cannot separate these. It is the same number under both.

The cut that separates them

Stop reading the rate on its own and put a second figure beside it: change orders as a share of completed jobs. Not per change order, per job. That converts the question from "how good are the ones we raised" to "how often does the work we quoted turn out not to be the work we do".

• High approval with a high share of jobs carrying a change is explanation one. The changes are real and frequent, and the estimate for that job type is short.
• High approval with a low share of jobs carrying a change is a flag for explanation two, and it needs one more test before you believe it.

That test is the one that closes the case: on the jobs that carried no change order, compare actual hours against estimated hours. If the unchanged jobs came in near their estimates, the low change rate is honest and the shop is simply scoping well. If the unchanged jobs ran materially over, the extra work happened and nobody billed it.

Worked case: 94 percent, and the bad news underneath

A shop closes 60 jobs in a quarter. Eighteen change orders were raised, 17 approved and 1 rejected, so the approval rate is 17 of 18, about 94 percent. All 18 had an answer before the quarter closed, which is what makes the rate readable at all, and 18 decided changes clears the floor of about 15 stated above.

Those 18 changes were raised on 11 of the 60 completed jobs, about 18 percent of them, or a little under one job in five, at roughly 1.6 changes on each affected job. So the shop sits at high approval and a low share of jobs carrying a change, which is the flag for explanation two.

Note what those 11 jobs are before going further. They hold 150 of the quarter's 550 estimated hours, about 27 percent of estimated hours against 18 percent of jobs. Changes land on the bigger work, which is what you would expect and which is also why a share stated in job counts is not the same statement as a share stated in hours. Say which one you are quoting.

Now the test. Split the quarter by whether a change order was raised:

The changed jobs landed within about 2 percent of their revised plan, which is 188 against 184 and is what a working change order process looks like. The 49 unchanged jobs used 472 hours against 400 estimated, 18 percent over, and none of it was billed. Both percentages are totals over totals within their group, not averages of each job's own overrun; those two figures differ whenever job sizes differ, which the gross margin average-of-averages card explains.

So the 94 percent is true and it is the wrong headline. The shop wrote up 34 hours of change work in the quarter and absorbed 72 hours, which is 472 minus 400, spread across 49 of its 60 jobs. The absorbed hours are more than double the billed ones. The approval rate was never going to show that, because every one of those 72 hours is outside its denominator.

The correction is not to raise the approval bar. It is to raise more change orders, accept that the rate will fall, and read the rate falling as the process starting to work.

When the rate is low, it is usually timing

A rate under about 70 percent is a conversation problem far more often than a pricing problem. Three causes, with the tell that separates them:

• Raised too late. The change is presented after the customer has been given a total, sometimes after the work is done. At that point it is a bill, not a choice, and people reject bills they did not agree to. Tell: the gap between the date the work was discovered and the date the change was raised is measured in days.
• Raised too vaguely. The change says what will be done and not what happens if it is not done. Tell: rejected changes cluster on work whose necessity is arguable rather than on price.
• Raised defensively. Every deviation becomes a change order so nobody can be blamed later, including small ones nobody would bill. Tell: the rejected ones are the small ones, in hours, and the customer is rejecting noise.

At a low rate the population is small enough to read one at a time, which is more useful than any ratio. Take the rejections from a quarter and read the reasons in order. Three rejections with three different reasons is normal life. Three with the same reason is a process fault you can name.

The fix for the first cause is a habit rather than a form: raise the change at the moment of discovery, before the added work starts, and say the three things that make it a decision - what was found, what it adds in hours and in schedule, and what happens if it is left. Then ask for in or out. A customer given that at the moment the wall opens usually says yes, and the yes means something different from the one extracted three days later.

Two contract shapes take this whole diagnostic off the table, and it is worth knowing which of your work sits in them. On time-and-materials or cost-plus work the change order is largely ceremonial, since extra work bills itself; the rate will sit near 100 percent and carry no information, and the equivalent read there is billed hours against the hours quoted as a not-to-exceed. And on work under a general contractor, approval runs through their paperwork cycle rather than a customer's judgement, so a low rate can be entirely a submission-deadline problem. Check the rejection dates against the contract's notice window before reading anything into the figure: a change rejected as late is not a change rejected on merit.

Reading the two numbers together

Between about 70 and 90 percent the rate is not saying much on its own, which is the honest answer for most shops most quarters. The share of jobs carrying a change is the figure that keeps working across that whole middle band, and it is the one to trend.

References

• See related: Estimate Conversion Rate and the Cohort Problem - undecided cases at the edge of a window, and what they do to a rate
• See related: Change Value as a Share of the Original Job - what the changes were worth, and the distribution that names the fix
• See related: Gross Margin Percent Is an Average of Averages - why a total-over-total rate and a mean of per-job rates differ
• Trade-standard practice: written, signed change orders authorising the added scope before that work proceeds