Expense per Job Only Sees Expenses Somebody Attached to a Job

Why this matters

Average expense per job is one of the few cost figures a shop can read without a bookkeeper, and it is computed entirely from expenses that a human being remembered to tie to a job. Every receipt that went in without that tie is invisible to it. Some of those belong nowhere near a job and their absence is correct. The rest are real job cost sitting in a pool that no job's margin will ever see, and because the biggest jobs generate the most loose paperwork, that pool concentrates on exactly the work you most need priced right. The figure does not warn you. It reads low, steady and reassuring while the install crew's jobs quietly report margins they never earned.

What the figure is actually counting

The numerator is the sum of approved expense amounts that carry a link to a specific job, dated in your window. Three parts of that sentence each remove something:

• Approved. Anything still sitting unreviewed is outside the numerator, so the figure is also a function of how current your review queue is. That behaviour belongs to a sibling card and is not re-derived here. See related: The Expense Approval Rate and the Queue Behind It.
• Carry a link to a specific job. Overhead has no job to attach to, by design. So does anything bought for stock, anything bought before the job existed, and anything a tech paid for and filed without naming the address.
• Dated in your window. The date on the expense, not the date of the job. A receipt from a job you closed five weeks ago, entered late, lands in this month and is divided across this month's jobs.

The denominator is where shops silently disagree with each other. It is either every job in the window or only the jobs that carry at least one expense, and the two answer different questions. Over every job, you get the average expense burden of doing a job at your shop, dragged down by the jobs that consumed nothing but labour. Over jobs with expenses, you get the average expense burden of a job that actually buys something. Neither is wrong; using one and describing the other is. Find out which one you are reading before you compare it to anything.

Compute the attach rate before you read anything else

The attach rate is the share of approved expenses carrying a job link, and it has to be computed twice, because the two versions do not agree and they mean different things.

• By count, lines with a job link over all approved lines. That measures the habit. It tells you how often the person entering the expense does the extra step.
• By money, attached spend over all approved spend. That measures the materiality. It tells you how much of what you actually spent is visible at the job layer.

They diverge whenever the unattached lines are a different size from the attached ones, which is the normal case. A count rate quoted where a money rate is meant is the standard way this figure gets cleared without anybody noticing the problem.

A month with patchy attachment

Index every amount in units where 1.0 unit is roughly what one hour of technician time costs the shop. A shop closes 60 jobs in a month and approves 140 expense lines totalling 420 units.

• Attached: 96 lines, 252 units. Mean line, 2.63 units.
• Unattached: 44 lines, 168 units. Mean line, 3.82 units.

So the attach rate by count is 96 of 140 lines, 68.6 percent, and by money it is 252 of 420 units, 60.0 percent. Those are 8.6 points apart, and the money figure is the worse one, because the lines that go in without a job link average 3.82 units against the attached lines' 2.63.

Expense per job, as reported, is 252 units over 60 jobs: 4.20 units. Read instead over the 38 jobs that carry at least one expense, the same numerator gives 6.63 units, a figure about 1.6 times the first one with nothing changed but the denominator.

Now open the 44 unattached lines and sort them by hand. The shop finds two populations:

• 26 lines, 71 units, true overhead. Insurance, software, the office internet, a tyre for the shop truck, advertising. These have no job and should never acquire one.
• 18 lines, 97 units, job cost nobody attached. Counter purchases on open jobs, two equipment rentals, a dump run, a permit fee paid at a window, a subcontractor's trip charge. Mean line 5.39 units against the overhead lines' 71 over 26, or 2.73 units - just under double, and that gap is what makes the missed lines worth chasing.

Call that second population the exception pool. It is 97 units, which is 23.1 percent of the month's 420 units of approved spend and 27.8 percent of the 349 units that should have reached a job (252 attached plus 97 missed).

So the true figure over all 60 jobs is 349 units over 60, or 5.82 units, against the 4.20 units reported. The reported number is light by 1.62 units, which is 27.8 percent of the true one - the same share as the exception pool, because it is the same arithmetic read from the other end.

Where the missing spend landed

An average that is 27.8 percent light would be tolerable if the error were spread evenly. It is not, and the reason is structural rather than careless. A one-visit service call generates one purchase, made on the way, usually on the original order. A four-day install generates a running purchase account, two rentals, a dump run, a permit and a sub's trip charge, most of it bought mid-job by whoever was free, and each of those is a separate chance for a receipt to arrive with no job named on it.

The shop traces the exception pool back to the jobs it belonged to. 12 of the 18 lines, 74 of the 97 units, belong to 5 jobs. That is 76.3 percent of the missing spend landing on 5 of 60 jobs, 8.3 percent of the job count, and all five are multi-day work.

Take one of them, a little under the 14.8-unit average of those five. Revenue 62 units. Recorded direct cost 41 units, of which 15 units were attached expenses and the rest labour and travel. Recorded margin is (62 minus 41) over 62, or 33.9 percent. Its share of the exception pool is 13 units, so corrected direct cost is 54 units and corrected margin is (62 minus 54) over 62, or 12.9 percent. That one job lost 21.0 margin points, and it lost them at the job layer only.

This is the part worth carrying away. The job type the shop believed was its strongest earner is the job type that generates the loose paperwork, so it is the job type whose margin is most overstated. Price the next install off that 33.9 percent and you will bid a job type you think clears a third and actually clears an eighth.

The rules that make the number readable

Three lists, and they have to be written down rather than understood.

Must attach, no exceptions. Anything bought at a supplier while a job is open, whether or not it was on the original order. Any rental, dump or disposal fee, permit or inspection fee tied to a specific address. Any subcontractor invoice. Any third-party charge naming a site. The test is not "was it planned", it is "does this charge exist because of one identifiable job".

Never attach to a job. Rent, insurance, software, marketing, office supplies, general fuel, training - and note that two items shops usually add to this list are not simple expenses at all: a financed vehicle payment splits into principal, which is not an expense, and interest, which is, while a tool with a useful life beyond the year is a capital asset your accountant may capitalise and depreciate or elect to expense under 26 USC 179, so put both in front of whoever files your return rather than onto a list the field maintains. Write the list out, because over-attaching is the mirror failure and it is not harmless: moving an overhead charge onto a job shifts it above the gross margin line and raises reported job cost while leaving net profit untouched. A tech who attaches lunch to a job is corrupting the same figure as a tech who attaches nothing.

The undecidable few. Bulk stock, a tool bought for one job and kept, fuel on a day that ran three jobs. Pick a default for each, write the default next to the category, and stop relitigating it monthly. A consistent wrong answer is readable and an inconsistent right one is not - but that trade is only available inside your own management reporting, because the same classification also lands on a tax return, a workers compensation payroll audit and a lender's covenant calculation, none of which accept consistency as a substitute for being right. Set these defaults with your accountant so the readable answer and the filed answer are the same one.

The reconciliation that proves the rules are working

The rules are worthless without a monthly tie-out, because nothing else tells you they are being followed.

Take total approved spend for the window. Subtract attached spend. Every unattached line left over must fall in a named never-attach category. Anything that does not is the exception pool, and the exception pool is a work queue, not a statistic. Somebody opens each line, finds the job from the date and the site on the receipt, and attaches it - or writes it to overhead deliberately and records that they did.

Set the cadence weekly and the stop at 14 days: no line sits unattached and uncategorised past two weekly cycles. That is the same two-cycle clock the approval-queue card sets for deciding an expense, applied to a different act, and both clocks exist for the same reason, which is that a paper trail cools fast. At three weeks nobody remembers which job the counter purchase was for.

Two numbers come out of the tie-out and both are worth keeping monthly: the money attach rate, which should be flat, and the exception pool's share of attachable spend, which is the error bar on expense per job. In the worked month those read 60.0 percent and 27.8 percent. If you cannot state your exception pool's share, you cannot state your figure's error bar either, and an unbounded cost figure is not a cost figure.

Why this survives every check a shop already runs

The reason this defect can run for years is that nothing above the job layer is wrong. The money left the bank. The bookkeeper caught it. The year closed, the return filed, and the profit was real.

Read down that last column. Two of the five rows are right, and they are the two everybody checks: net profit and the bank balance. The three above them are wrong, one of those only when the charge was coded to overhead rather than to cost of sales, and none of the three is reconciled against anything. An owner comparing the P&L to the bank statement will find them agreeing all year while every install job in the shop reports a margin it did not earn. The only instrument that finds it is the tie-out, because the tie-out is the only place the two layers are made to agree with each other.

References

• See related: Cost per Job, and What Is Actually Inside It, for what the wider cost figure includes and excludes
• See related: The Expense Approval Rate and the Queue Behind It, for why the approved subset is not the whole subset
• See related: The Cost Categories Worth Separating, for whether a given split is worth making at all