Last month the sales assistant at a fit-out contractor told 118 people it could not help them. Nobody on the sales team saw one of those conversations. Then the owner read thirty of them on a Thursday afternoon, and found nine she would have called back herself. The contractor is an illustrative composite, not a client, and every number in this article is invented for the example: illustrative, not a statistic.
Here is the month. 410 enquiries reached the website assistant. It passed 292 to the sales team and declined 118 (410 minus 118 is 292). Its rule looked sensible: a project under a minimum size gets the reply "we focus on larger projects", and the conversation closes. The owner was pleased with it. The sales desk was quieter, fewer calls were about small jobs, and the weekly report showed qualified enquiries rising. Nothing in the report described the 118.
She read thirty of them, picked evenly across the month. Twenty-one were right declines: a one-room repaint, a request from another country. Nine were not. That is 9 of 30, or 30%. I will come back to what that is worth. First, why nobody saw it.
Why a wrong "no" is invisible
A wrong yes announces itself. A salesperson spends an hour with someone who was never going to buy, and the hour sits in a calendar and in a complaint at the Monday meeting.
A wrong no leaves nothing behind. The enquirer never becomes a lead record, so there is no lost-deal reason. They do not complain. They phone the next contractor. Every sales dashboard I see reports on the people who got through the door: qualified, booked, won. It is a cousin of the survivorship problem. You study the ones who passed, and conclude the filter is fine.
I am not against the filter. An assistant that politely turns away a one-room repaint has saved someone an afternoon. My point is narrower. A decision that nobody reviews gets no feedback, and a filter without feedback drifts. I wrote earlier about leads that went quiet after they were in, in four rules for following up a quiet lead. This article is about the ones that never got in.
Read the pile blind: the Declined-Pile Audit
In my work I call this the Declined-Pile Audit. It takes one afternoon and needs no new software, only the conversation records you already have.
- Define "declined" widely. Count any enquiry that received no human contact: auto-closed, redirected to the FAQ, told "not a fit", out of area, below minimum, asked to resubmit, or simply never answered. Pull the last 60 days.
- Sample, do not browse. If the pile is under 40, read all of it. Otherwise read 40 picked evenly across the period (every third or fourth one, by date). Not the first 40, and not the ones that look interesting. In the composite, the owner read 30 of 118, which is roughly every fourth.
- Have someone who sells do a blind read. For each conversation, mark Right decline, Wrong decline, or Cannot tell. Judge only what was visible in the conversation. For every Wrong, write one line saying why.
- Sort each Wrong into a cause, then price it. The four causes are in the next section. The pricing comes after that.
- Change one thing and set the date. Alter one rule or add one question, not five. Write down the re-audit date (30 days out) and the number that should move: the wrong-decline share. Keep a "second look" lane: declines carrying a referral or repeat-client flag get a human glance before the conversation closes.
Thirty is a thin sample. Treat the 30% in the composite as a rough size, not a measurement, and let the next audit tighten it. A rough size is still far better than the nothing the report was showing.
Four ways a "no" goes wrong
When I sort the Wrong pile, the reasons tend to fall into four groups. The counts below are from the composite: nine wrong declines, 4 + 2 + 2 + 1 = 9.
- Phase hidden (4). The first job is small and the account is not. "Just the reception first", from a client planning a whole floor. The assistant read the headline and ignored the plan behind it.
- Source unseen (2). Both were referred by an existing client. The assistant treated a warm introduction as a cold enquiry, and closed it with the same line.
- Field missing (2). The assistant never asked the one question that mattered, such as whether the job was the first of several sites.
- Rule too blunt (1). A size threshold applied to the headline, with context in the same message that pointed the other way.
The useful thing about this sort is that it shows the repair. Four of nine come down to one missing question about the larger plan. Two of nine come down to a flag the assistant never looked at. Neither fix needs a looser filter.
Price both mistakes before you change anything
Here is the calculation, again with invented numbers (illustrative composite, not a statistic). Use your own figures when you run it.
Wrong-decline cost per month = wrong declines in sample / sample size x pile size x close rate x average margin.
- 9 / 30 = 30% wrong in the sample.
- 30% of 118 is 35.4, so about 35 wrong declines a month.
- At a 10% close rate, that is about 3.5 jobs a month.
- At $4,000 average margin, 3.5 x $4,000 = $14,000 a month, about.
| Step | Illustrative figure |
|---|---|
| Sample | 9 wrong of 30 read = 30% |
| Pile | 118 declined, about 35 wrong declines |
| Wrong-decline cost | 35 x 10% close rate x $4,000 margin = about $14,000 a month |
| Wrong-accept cost | 100 dead enquiries x 20 min x $30/h = about 33 h = about $1,000 a month |
| Per case | $400 vs $10, about 40 to 1 |
Wrong-accept cost per month = salesperson minutes x accepted-but-dead enquiries x hourly cost.
- Suppose 100 of the 292 passed enquiries went nowhere.
- At 20 minutes each, 100 x 20 = 2,000 minutes, about 33 hours.
- At $30 an hour, 33 hours is about $1,000 a month.
The asymmetry is the finding. One wrong decline is worth, on average, 10% x $4,000 = $400. One wrong accept costs 20 minutes at $30 an hour, which is $10. It takes about 40 wrong accepts to equal one wrong decline. Because the sample is thin, test the result: at half the observed rate (15%), 118 x 15% is about 17 wrong declines, which is still about $7,000 a month against $1,000.
An honest caveat. Some wrong declines are only wrong in hindsight. That is why step 3 judges only what was visible in the conversation, the same test I described in asking what was knowable then. Mark those Cannot tell and leave them out of the count.
The decision in the composite was not to loosen the filter. Sending every small job to the sales desk would have wrecked the 292 and the time it saves. She added one question ("Is this the first phase of a larger project?") and a second-look lane for referral and repeat-client flags. If the assistant also makes promises at the front door, the commitment ladder covers what it may say.
She now reads thirty declined conversations on the first Thursday of every month, and the wrong-decline share is the one number she watches. A filter you never audit is not a filter. It is a policy you have not read.
Related evidence: A qualifier tuned for precision (fewer wasted sales hours) can quietly lose recall (good buyers declined); reading a sample of the declines is how you measure recall. (recall, the share of real buyers the assistant actually let through)
The weekly report only shows the enquiries that passed the filter, so the declined ones never enter anyone's judgement of whether the rule works. (survivorship bias, judging a filter only by what passed it)
