The Close Rate Improved Because They Stopped Quoting

Why this matters

A close rate has two ways to rise and only one of them is good news. This shop's rose more than eight points, held for a quarter, and fell back, with nothing in the selling changing at any point. The owner had already spent money on the explanation by the time anybody checked. The whole diagnosis is in the two counts underneath the rate, and neither of them was on the report.

The signal, and the thing that made it worth chasing

Three quarters, one number.

Quarter B sat between A and C and was a transition quarter for reasons that turn out to matter. A sales training session ran at the start of B, so the jump into C read as the training landing, and the owner bought a second round of it for the whole team.

Then D came in at 32.2 percent, back where it started, with the newly trained team in place. That is what sent the owner back through the record. The interesting part is that the answer was already sitting in the quarters he had, and no experiment had to be run to find it: the shop had already accidentally performed one.

The counts under the rate

The first thing to pull is the two numbers the rate was built from.

Quarter C, the celebrated quarter, booked 7 fewer approvals than quarter A, 23 against 30, a fall of 23.3 percent. Estimates sent fell by 38, from 96 to 58, a fall of 39.6 percent.

That is the entire mechanism in one line: the denominator fell faster than the numerator, so the rate rose. A rate cannot tell you which of its two parts moved, and this one moved because of the part nobody was watching.

Quarter D then restores the denominator to 90 and the rate returns to where it was. Whatever happened, it happened to the count of estimates and not to the selling.

The absence in the record

The record shows who raised each estimate and when. What it does not show, anywhere, is a decision to raise fewer of them, and that absence is the finding rather than a gap in the investigation.

Split the estimates by job type.

Two of the three types barely moved. The counter repair quote went from 41 to 6.

Counter quotes were written by the front-counter person. In quarter B the shop lost a part-timer, the counter person picked up dispatch duties, and the written quote for a small repair was the first thing to go, because for that work the price is already in the price book and the customer is standing right there. Nobody decided to stop. The habit stopped being possible and left no trace except an absence.

Check whether the work disappeared with the quotes, because that is the obvious alternative explanation:

Completed counter repairs fell by 3, from 14 to 11, which is 21.4 percent. Quotes fell by 35, from 41 to 6, which is 85.4 percent. The work was still happening. It was being booked verbally.

Worth noticing in quarter A: 14 completed against 8 approved quotes means 6 of that quarter's counter repairs were already being booked without a written quote. The habit that later became the rule was already half in place, which is usually how this looks.

The computation that settles it

Take quarter A's rate for each job type and apply it to quarter C's mix. If the selling had not changed, this is what quarter C should have reported.

• Counter repair: 6 estimates at A's rate of 19.5 percent, or 1.17 approvals
• Service add-on: 30 estimates at A's rate of 40.6 percent, or 12.19 approvals
• Replacement: 22 estimates at A's rate of 39.1 percent, or 8.61 approvals
• Total predicted: 21.97 approvals over 58 estimates, or 37.9 percent

Against quarter A's actual 31.3 percent, the mix change alone accounts for 6.6 of the 8.4-point rise. The residual, everything that is not mix, is 39.7 minus 37.9, or 1.8 points.

Now size that residual in the unit that matters. 1.8 points across 58 estimates is about one extra approval. Decompose it and it gets thinner still: counter repair contributed plus 0.83 of an approval, service add-on minus 0.19, replacement plus 0.39. So most of the residual comes from the counter type, computed on six estimates.

Six estimates is far under the 30-record floor a rate needs before it can be read as a rank at all. The training's entire measured effect, in the quarter it was credited for, is roughly one approval and most of that one sits in the least readable cell on the table.

The second figure telling the same lie

The owner had also noticed the average estimate value climbing and read it as the shop moving upmarket.

Hold each type's own average value at its quarter A level, which the records confirm within a few percent, and index the smallest counter repair quote at 0.4 units, the service add-on at 1.6 and the replacement at 8.0.

• Quarter A: 41 at 0.4, plus 32 at 1.6, plus 23 at 8.0, is 16.4 plus 51.2 plus 184.0, or 251.6 units over 96 estimates, an average of 2.62 units
• Quarter C: 6 at 0.4, plus 30 at 1.6, plus 22 at 8.0, is 2.4 plus 48.0 plus 176.0, or 226.4 units over 58 estimates, an average of 3.90 units

The average estimate value rose 48.9 percent and not one job type's typical value moved by anything. Dropping the smallest type out of the denominator raised the mean of what remained. Same mechanism, second figure, and it was being read as a second piece of evidence for the same wrong conclusion.

What the training was actually worth

Since the question was live, they answered it properly, per job type, on rolled windows where a single quarter is too thin.

The counter row is where the whole story was, and its middle cell is unreadable: 33.3 percent of six estimates is two approvals, and the quarters either side of it sit at 19.5 and 18.4 percent on 41 and 38 estimates. Nothing about counter selling changed.

Replacement sits at 23, 22 and 21 estimates per quarter, all under the 30-record floor, so it has to be rolled before it can be read. Rolling adjacent pairs: quarters A and C together give 18 approvals on 45 estimates, 40.0 percent; quarters C and D give 18 on 43, 41.9 percent. A 1.9-point rise on 43 estimates is well inside what this many records produce by chance.

So the honest verdict after three quarters is that the training has not yet proved itself, and the shop will know in another two. That is a genuinely better position than the one they were in, which was certainty in both directions inside six months.

What they changed

The denominator became a policy, not a staffing outcome. Either a job type is always quoted in writing or it is never quoted, and the decision is made once per type rather than per shift. This shop chose to keep writing the counter quote, because it wanted the paper trail on parts and the customer's agreement to the price. A shop that chooses the other way is fine too, as long as it then excludes that type from the rate entirely rather than reporting a rate over a type it half-quotes. Half-quoting a type turns the close rate into a measurement of who was on the counter.

The rate is read per job type, on a population that clears 30 records, rolled across quarters where it does not. The blended rate stays on the report, because it is the one an owner glances at, but nothing is decided on it while the mix is moving.

Every rate is printed with both of its counts. Estimates sent, approvals, then the rate. This is the change that would have caught the whole thing in quarter C, in about four seconds, because 96 and 30 against 58 and 23 is not a subtle picture once both pairs are on the page.

The rule that came out of it

A ratio is unreadable without the quantities underneath it, so print the components. A rate compresses two counts into one figure, and the compression loses the thing you most need: the same movement can come from the numerator, from the denominator, or from both, and the rate itself carries no record of which. Every other version of it names one particular pair. Compliance quoted beside coverage. An approval rate beside the queue that fed it. A recurring share beside the absolute one-off figure. A collections figure beside the invoices issued that set its denominator. Different pairs, same rule, and the fix never changes: print both counts and let the reader do the division.

So when a rate moves, read the counts before you read the rate. There are three cases and only one of them is what everybody assumes.

A sibling card covers the same inversion one stage earlier, at the lead rather than the estimate, where the denominator structurally excludes leads nobody touched. The difference is worth holding onto: there the exclusion is built into how the number is defined, so it is always there. Here the denominator moved because a habit stopped, which means it can move back, and it did.

References

• See related: Lead Conversion Rate Excludes the Leads You Never Touched - the same inversion at the lead stage, built into the definition rather than caused by a habit
• See related: Estimate Conversion Rate and the Cohort Problem - the other reason this denominator misbehaves
• See related: Average Estimate Value Counts Drafts and Conversion Does Not - the second figure that moved here, and why
• See related: Gross Margin Percent Is an Average of Averages - the same mix arithmetic applied to margin rather than to a close rate