The Month Margin Fell and Nothing About the Work Changed

Why this matters

A falling margin figure sends most owners straight at the crew, and it is the one explanation that costs something whether or not it is right. This is a real shape of that problem where every job performed exactly as it always had, no rate moved, nothing was discounted, and the reported margin still fell more than six points. The shop spent two weeks eliminating causes that were never there. The check that answered it takes one grouping, and it is the check to run first.

The signal

A shop running two technicians at the start of the window and three by the end of it watched its gross margin percentage fall for three consecutive months. Suspecting monthly noise, the owner pulled the two most recent full quarters and computed the same figure on each: 47.7 percent, then 41.1 percent. A 6.6-point fall on a quarter of work is not noise.

The figure was the plain average of each completed job's own margin percentage, which is what most shops read. Both quarters were computed the same way, on the same population, with the same cost components, so the comparison itself was sound.

Four candidates, killed in order

Pay rates. The obvious suspect, because labour is the largest cost on most of this shop's work. They pulled the rate history: no raises took effect inside the window, the technician who joined mid-window came in at the existing rate for that level, and nobody moved between pay arrangements. As a cross-check they computed labour cost per logged hour by technician across both quarters, which would also have caught a salaried technician priced on the wrong divisor. Flat. Eliminated.

Part costs. The correct test here is not what the catalogue says today. Part cost is snapshotted onto each usage line at the moment the part is used, so what matters is what the closed jobs actually booked. They compared the snapshotted unit cost of the ten most-used part numbers across both quarters, line by line. Identical. A supplier increase would have shown as a step in the later quarter's lines, and there was none. Eliminated.

Discounting. They compared quoted price against invoiced price on every converted estimate. In the earlier quarter, 2 of 14 converted installs were invoiced below quote. In the later quarter, 1 of 26. Discounting did not hold steady, it fell, so it cannot be the cause of a drop. Eliminated, and more firmly than the other two.

Hours per job. Installs averaged 21.0 crew-hours in the earlier quarter and 21.3 in the later one, a rise of 1.4 percent. Service calls averaged 1.9 hours in both. A margin fall of 6.6 points cannot come out of a 1.4 percent movement in hours on one type. Eliminated.

Two weeks gone, and the number still read 41.1 percent.

The grouping that answered it in one look

The owner grouped both quarters by job type and computed each type's own margin.

Neither type's margin moved by a point. Every job in the later quarter performed exactly as its type had always performed. What changed was how many of each there were:

Installs went from about a quarter of the job count to nearly half. Because the reported figure is a plain average over jobs, each install votes once, and installs have always been the thinner type.

Two cautions before that table is read as proof, both from the ranking card. The counts first: install sits at 14 jobs in the earlier quarter and 26 in the later one, both under its 30-job reliability floor, and service calls only just clear it, so those four cells are not evidence on their own, and that card is explicit that a type cannot be compared against its own prior period across counts like these. Then the definitions: if an install that included a repair could be booked either way depending on who wrote it up, two types showing identical margins can be the same jobs moving between labels. Compare the pair on logged hours per job and on part content. Here they are nowhere near each other, 1.9 hours against 21.0 and almost no parts against a full material list, so the types are real; where two do look alike on those measures, merge them before concluding anything.

What survives both cautions is the line below. Check it against the mix alone. Predicted average for the later quarter, using each type's unchanged margin and the later quarter's shares: 0.464 times 24 plus 0.536 times 56, which is 11.14 plus 30.00, or 41.1 percent. That is the reported figure to the tenth of a point. Mix accounts for the entire fall, leaving nothing for any other cause to explain - which is also why the four eliminations were guaranteed to come back clean before anybody ran them.

The second number, and why it disagreed

The owner also had a weighted figure, total margin earned over total revenue, and it had been nagging at them because the two disagreed about how bad things were. Index a service call's revenue at 1.0 unit and an install's at 12.0 units.

• Earlier quarter: revenue 40.0 plus 168.0, or 208.0 units. Margin earned 22.40 plus 40.32, or 62.72 units. Weighted margin 62.72 over 208.0, or 30.2 percent.
• Later quarter: revenue 30.0 plus 312.0, or 342.0 units. Margin earned 16.80 plus 74.88, or 91.68 units. Weighted margin 91.68 over 342.0, or 26.8 percent.

The weighted figure fell 3.4 points against the reported figure's 6.6. Neither is wrong. The plain average gives a small service call the same vote as a twelve-times-larger install, so when this shop's mix tilts toward installs the plain average swings roughly twice as far as the money does. The multiple is set by the size ratio between the two types, so it is not a constant you can carry to another shop. That disagreement is a mix signature, and recognising it would have pointed at the answer in the first hour.

The number nobody had looked at

Margin earned went from 62.72 units to 91.68 units. That is a rise of 46 percent, on revenue up 64 percent from 208.0 to 342.0 units. The shop had just finished the best quarter in its history, and every percentage on the page was down.

Installs were also now 312.0 of 342.0 units, or 91.2 percent of revenue, against 168.0 of 208.0, or 80.8 percent, the quarter before. A type that had been an occasional add-on was now the business.

One more figure the owner had never computed, and it is the one that put the growth in proportion. Crew hours in the earlier quarter: 40 service calls at 1.9 hours plus 14 installs at 21.0 hours, which is 76.0 plus 294.0, or 370.0 hours. In the later quarter: 30 at 1.9 plus 26 at 21.3, which is 57.0 plus 553.8, or 610.8 hours. Hours rose 65 percent while revenue rose 64 percent, so revenue per crew-hour went from 208.0 over 370.0, or 0.562 units an hour, to 342.0 over 610.8, or 0.560 - flat to within half a percent.

The shop grew by selling more hours at the same rate per hour, which is capacity growth. That is a legitimate way to grow and it is not the same thing as getting better at the work, and no percentage on the margin page distinguishes the two.

Why the mix moved

A mix shift has a cause, and it is worth finding, because the right action depends on it. Two things had happened at once. A property-manager relationship taken on mid-year was feeding install work, which is demand. And with two of the three now committed to install crews most weeks, the dispatch board had been pushing maintenance and service visits out, which is capacity.

That second half matters more than the first. If installs at 24 percent are crowding out service calls at 56 percent, the binding constraint is scheduling, not pricing, and raising install prices addresses the wrong thing first. The test the owner ran: count service requests received against service jobs completed, by month, across the window. Requests held steady and completions fell, which confirms displacement rather than a fall in demand. Had requests fallen too, the service work was simply not there and pricing the installs would have been the whole answer.

What the owner changed

Not a conversation with the crew. Nothing the crew did caused this, and a performance conversation held on these numbers would have been both wrong and expensive.

Two moves, in the order the displacement finding dictates.

First, protect the service capacity. Service requests were holding while completions fell, so the shop was turning away work that earns 0.56 units of margin in 1.9 crew-hours to deliver work that earns 2.88 units in 21.3. That is 0.29 units an hour against 0.14. Per job the install is the bigger earner by five times; per hour of the capacity that is actually scarce it is worth less than half, and hours are what the board is rationing. One of the three was ring-fenced for service and maintenance and kept off the install rotation, which recovers the displaced jobs without touching a price.

Second, price the install type. A margin of 24 percent had been acceptable while installs were a quarter of the count. At nearly half the count and over nine tenths of the revenue, it is the shop's margin. Install pricing went up on the two most common configurations, and the quoted scope was tightened to name the two items the crew had been absorbing.

They also set a standing rule, as a starting point to tune: a job type earns its own margin target, rather than inheriting the shop blend, once it clears both of the ranking card's gates - 30 completed revenue jobs in the period, for reliability, and 5 percent of the period's revenue, for materiality. Installs clear the second overwhelmingly here at 91.2 percent of revenue, and sit just under the first at 26 jobs in a quarter, so this shop sets the install target on two quarters rolled together. A month would never have cleared either gate.

How they confirmed it

The next quarter they re-ran the same grouping. Install margin moved from 24 to 29 percent; the service-call margin stayed at 56. That is the signature a pricing change on one type should leave: its own margin moves and nothing else's does. Had both types moved, the price change was not what did it, and they would have gone looking for what else landed in the same window.

The mix moved too, because that was the other half of the plan: 38 service calls and 22 installs, 60 jobs with revenue. So the reported figure now carries two changes at once, and reading it whole would credit the price change with work the scheduling change did. Separate them by holding one constant.

• Price effect alone. Hold the mix at the problem quarter's 30 and 26 and apply the new install margin: (30 times 56) plus (26 times 29), or 1680 plus 754, which is 2434 over 56 jobs, or 43.5 percent. Against 41.1 percent, the price change is worth 2.4 points.
• Actual reported figure. (38 times 56) plus (22 times 29), or 2128 plus 638, which is 2766 over 60 jobs, or 46.1 percent. The remaining 2.6 points came from the mix moving back.

Both changes worked, and now the owner knows which did what. The weighted figure moved in step: revenue 38.0 plus 264.0, or 302.0 units; margin earned 21.28 plus 76.56, or 97.84 units; 97.84 over 302.0 is 32.4 percent, against 30.2 percent two quarters earlier and 26.8 percent at the bottom.

The plain average, at 46.1 percent, has still not returned to the 47.7 percent it started at, and it should not be expected to. More of this shop's jobs are installs now than were then, and installs carry the lower percentage by nature. Judge a recovery on the figure that tracks the money, and keep the plain average for the thing it is genuinely good at: telling you when a typical job of a given type has drifted.

References

• See related: Gross Margin Percent Is an Average of Averages - why the two figures in this case disagreed, and how to compute both
• See related: The Job-Type Margin Ranking and the Count Beside It
• See related: Job Profitability by Service Type
• See related: Reading Seasonality Out of a Metric Before You Panic - the other common reason a quarter-on-quarter figure moves without the work changing