Days From Estimate to Job and What a Long Tail Means

Why this matters

This is the one number in the quoting chain where the standard statistic is simply the wrong one. The mean number of days from estimate to job lands, in most shops, in a gap where almost no customer sits, and it then sets the follow-up schedule and the quote validity period for every job the shop sells. A shop reading a mean of roughly four weeks calls its customers back in the fourth week, by which time more than half of the customers who were ever going to say yes already have.

What it measures, and over which pairs

Elapsed days from an estimate being raised to the linked job being created, computed over the pairs that exist.

That last clause is doing a lot. An estimate that never converted has no job to pair with, so it contributes nothing. This distribution therefore describes how long the wins took, not how long to keep pursuing, and the difference between those two questions is the subject of a sibling card on survivor bias. Keep the clocks straight too: this one runs from estimate raised to job created. The sales cycle is a different clock, lead created to lead won, measured between different events on a different population.

The mean lands where nobody is

A quarter's 48 conversions, banded by lag:

Forty-eight conversions. Taking each band at its midpoint, the mean is about 27 days. The median falls between the 24th and 25th values, both of which sit in the 4-to-7 band, so the median is around day 6.

Three weeks apart, same data. Now look at where the mean landed. The 22-to-30 band holds 3 of the 48 conversions, 6.3 percent of them, and only two bands on this table are thinner. The mean is describing a region the shop's customers mostly skip.

That is what a mean does to a two-humped distribution. It is a balance point, and the balance point of two clusters separated by a gap is the gap.

Read it as density, not as counts

Bands of different widths cannot be compared by their counts. Convert to conversions per day and the shape appears:

• Days 0 to 7: 25 conversions across 8 days, about 3.1 a day. That is 52.1 percent of the quarter's conversions in the first week.
• Days 8 to 30: 9 conversions across 23 days, about 0.4 a day, roughly one eighth the density of the first week. This is 18.8 percent of conversions spread over nearly three times the elapsed time.
• Days 31 to 180: 14 conversions across 150 days, under 0.1 a day.

The 27-day mean sits in the 8-to-30 stretch, running at about an eighth the density of the first week. It is not a typical wait, and it is not the quiet end either: the tail past day 30 runs at under a thirtieth of the first week's density. The mean has landed in the valley between the two places customers actually are.

One more line from the same table, and it is the one that reframes the quarter: as many conversions landed past day 30 as landed in the first four days, 14 of 48 each, 29.2 percent. A shop that closes its file at 30 days is walking away from as much converted work as it collects in its best four days.

A two-number test for the shape

You do not need to band anything to find out whether your mean is safe to quote. Compute the mean and the median and divide one by the other.

• Ratio near 1 to 1.2. The distribution is close to symmetric. The mean is fine and the median adds nothing.
• Ratio above about 2 to 1. A tail is pulling the mean, and the mean should not be quoted on its own. Band the data before you decide anything with it.
• In between. Look at the table. The mean is drifting but may still be readable.

This shop reads a mean of about 27 days against a median of about 6, a ratio near 4.5 to 1, which is a tail so heavy that the mean has left the data entirely. Treat the 2-to-1 figure as a starting point to tune: it is a trigger to go and look, not a finding.

The same test works on any elapsed-time figure in the shop, because every one of them is a duration with a floor at zero and no ceiling, which is the exact shape that generates right tails.

The two humps are two different customers

The shape has a cause, and naming it is what turns the distribution into a decision.

The fast hump is a customer who had already decided before the quote existed. Something failed, or they had committed to the work and only needed a number to proceed. The quote is a formality and the competition is speed.

The tail is discretionary or gated work. The customer is waiting on a season, a budget, a second quote, a landlord, a permit, or the money. Nothing you say on day 4 changes their timing, because their timing is not about you.

Those are two populations, not one population with a variable mood, which is exactly why averaging them produces a number that describes neither.

What the fast hump sets: follow-up cadence

Cadence is when you touch. It comes off the first hump, because that is where the decisions are being made.

• First follow-up at 48 hours. Inside the 0-to-3 band, which holds 29.2 percent of conversions on its own.
• Second at day 5 or 6, before the first week closes and takes 52.1 percent of the quarter's conversions with it.
• Third at day 14, where the valley has started. The conversation changes here from "are you ready to book" to "is this still live, and what is it waiting on". That second question is the one that produces a marked outcome instead of a permanent maybe.
• Monthly after that, for as long as the quote is still open. The wins tail is 150 days wide, so weekly contact across it is harassment and it teaches the customer to stop answering. How long to keep the file open at all is not a question this table can answer, because it holds no losses.

Tune those to your own table rather than copying them. The rule that travels is the shape of the rule: put the first two touches inside your own first hump, and put the third where your own density collapses.

What the tail sets: how long a quote stays honoured

This is the other decision the distribution owns, and most shops set it by habit at 30 days.

On this table a 30-day validity window expires on 29.2 percent of the conversions this shop actually gets. Some of that work still lands, after an awkward conversation about a number that is no longer good. Some of it does not.

Set the window at the band holding your own 90th percentile of conversion lag. Here the 43rd and 44th of 48 ordered values both sit in the 61-to-120 band, so the honest window is around 90 days, with a stated materials clause rather than an open-ended promise. A materials clause is what lets you honour the date without honouring a price your supplier has since moved.

That number is specific to a shop's mix. Read it again whenever the mix moves, because a shop that adds a discretionary or budgeted job type gains a tail it did not have.

The tail also arrives as a scheduling problem, which is the part nobody plans for. A conversion on day 95 lands on a crew whose calendar was built without it, from a customer who has been quiet for three months and now expects the work soon. Two arrangements take the sting out: hold the tail estimates on a list that gets reviewed before the schedule is set for the following month, so a long-dormant quote becomes a phone call rather than a surprise, and write the lead time into the quote as plainly as the price. A quote that says what it costs and nothing about when work would start has promised the customer's preferred answer by default.

What changes the answer

The bimodal shape is normal, not universal, and two conditions replace it.

Third-party gated work. Insurance-funded, permit-gated, or landlord-approved jobs cannot convert quickly no matter how ready the customer is. That shop's distribution has no fast hump at all: it is a single cluster sitting weeks out. The tell is the missing first band rather than a small one, and for that shop the mean and median move back into agreement and the mean becomes usable again.

A single dominant job type. A shop that does one thing has one customer type and one decision pattern, so its distribution is single-humped and the mean is honest. The moment a second type with a different buying pattern reaches a meaningful share, the mean starts drifting into the gap. Reading the distribution per job type is what catches this, and it is worth doing once a year even in a shop that thinks it does one thing.

The count floor is different here from the one a rate needs. Ranking two job types against each other on a rate needs about 30 records per type. A distribution needs enough records that the shape is stable, which in practice means a few dozen conversions plus a check that last quarter's shape looks like this quarter's. If the two shapes disagree, you have a mix change to explain, not a statistic to report.

The three regions and their tells

A shop that reads one mean across all three regions gets a follow-up schedule tuned to the valley, which is the only region where nobody is deciding anything.

References

• See related: Estimate Conversion Rate and the Cohort Problem - this distribution is where that card's cycle allowance comes from
• See related: Average Sales Cycle Describes Winners Only - the loss timing that sets how long to keep pursuing, which this distribution cannot tell you
• See related: Average Estimate Value Counts Drafts and Conversion Does Not - reading the same estimates by size rather than by timing
• See related: First Response Time and the Window That Decides the Job - the clock that runs before this one starts