Hi folks. I went looking for the source of a statistic last week and ended up in 1942, which is not where I expected my Tuesday to go.

You have seen the stat. It is on every sales poster ever printed: "80% of sales require five follow-ups, but 44% of salespeople give up after one." I have quoted it myself. I wanted the original study, partly because I am a history major by training and I cannot leave a citation alone.

There is no study. In 2021 the organization credited with the number went looking for its own source and found a member survey from 1942, run by the Long Island chapter of the National Sales Executives Association, with fewer than forty participants. No published methodology, no controls. The tidy cascade that usually follows it, 44% quit after one, 22% after two, and so on, traces back to a marketing blog page that has since been deleted.

So the most repeated number in sales is an unaudited wartime poll of about three dozen door-to-door salesmen on Long Island. I laughed, then I felt a little bad, because I have put that number in a deck.

Here is the part that actually matters. The real data is not kinder to us. It is worse.

What 438,000 contractor quotes actually say

Level, a finance firm that works with trades businesses, went through 438,000 quotes across more than 2,200 contractors. That is a real dataset with a real denominator, and the picture it paints is specific enough to act on.

The median contractor converts 73.9% of the quotes that actually reach a customer and get a decision. Top quartile is 83.2%. The bottom decile sits at 47.1%, which is a coin flip on work you already paid to estimate.

Then there is the number that stopped me: of those 438,000 quotes, 48,000 were created and never sent to the customer at all. Roughly one in nine. Somebody drove to the site, measured, priced it, built the document, and then the document sat there. That is not a sales problem or a pricing problem. That is a nobody-pushed-send problem, and it happens in about eleven percent of cases across two thousand businesses.

The timing data is just as blunt. Quotes that convert close in a median of two days. Quotes that are lost linger for a median of twenty-nine, and about half of them are not marked lost until more than a month after they went out. The customer decided weeks ago. The pipeline report just has not caught up.

Broader sales research says the same thing from a different angle. Velocify's analysis of 3.5 million lead records found 93% of converted leads were reached within six attempts, and that half of inbound leads are never contacted a second time. Not never contacted. Never contacted again. One touch, then silence, on both ends.

Level's own math on what this is worth: a contractor quoting $5M a year who moves from 60% to 85% conversion picks up $1.25M in revenue, roughly $500K in gross profit at 40% margins, with no additional marketing, no additional estimating, and no additional trucks. Same quotes. Just more of them answered.

The reason it does not happen is not laziness

I want to be careful here, because the genre of content that says "contractors are leaving money on the table, just follow up more" is useless and slightly insulting. Everyone already knows they should follow up. Nobody needs that sentence.

The reason it does not happen is that following up is labor, and labor has always had a price.

A landscaper I talk to who works residential jobs around DC has described this tension to me directly, and I think it is the truest thing anyone has told me about running a small shop: he is trying to market the business and run the business at the same time, with the same pair of hands. On Tuesday afternoon he is either on a job site or he is chasing a quote from last week. He cannot be both places, and the job site is the one that pays today.

So the follow-up is not skipped because it is unimportant. It is skipped because at 4:45pm it is competing with payroll, a customer who wants to move Thursday, and a truck that needs an inspection. Every time, forever. For a hundred years the only fix was to hire someone whose entire job was remembering. The median wage for a secretary or administrative assistant was $48,310 in May 2025, which is about $23 an hour before payroll taxes and benefits. If a proper follow-up takes three minutes of a person's attention, it costs a little over a dollar. Not much. But it is a dollar you only spend if you have already hired the person, and most shops under a few million in revenue have not.

That is the whole reason the chore goes undone. It is not a mystery and it is not a character flaw. It is arithmetic.

The arithmetic changed

This is the part I find genuinely interesting, and it is why I keep writing these build-in-public posts instead of tips posts.

A follow-up message that actually knows something, which job it is about, what was quoted, what the customer asked on site, what has already been sent, costs about 3,000 tokens of context and a few hundred tokens of writing. At current pricing for a small production model, $1.00 per million input tokens and $5.00 per million output, that message costs roughly half a cent.

Say you send 50 quotes a month and each one gets three touches. That is 150 messages, and the model cost is about seventy cents. Multiply by ten for a real agent that checks the CRM, handles replies, and retries, and you are still under ten dollars a month for the thing the business has never once done consistently.

The human version of that same work costs about $175 a month in attention you do not have. The machine version costs less than lunch.

I want to be honest about what that number does not include, because the version of this post that stops here is the one that gets people in trouble. The tokens are the cheap part. The expensive parts are connecting it to the systems where your quotes actually live, deciding what the agent is allowed to do on its own, and having someone competent look at what it sent. Those are real costs and they do not fall to zero. What fell to zero was the marginal cost of the chore itself, and the chore was always the bottleneck.

Where these things go wrong

Gartner expects that more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear value, and inadequate risk controls. In the same release they note something sharper: of the thousands of vendors selling agentic AI, they estimate only about 130 are real, the rest being existing chatbots and automation rebranded. They have a name for it, "agent washing," which I admit I enjoyed.

We have hit our own version of this. When we deployed an AI voice agent for Black River Design and Build, the Wisconsin remodeler we work with, the lesson that cost us the most to learn was that it has to say it is an AI up front. Callers who figured it out mid-conversation were not amused, and they were not annoyed at the software. They were annoyed at the contractor. That one applies directly here: a follow-up that pretends to be a person typing is a liability, not a feature, and the moment a customer replies with something a machine should not be answering, a human needs to be holding it.

The version that works is narrower than the demo. It watches the quotes, it drafts and sends the boring touches on a schedule, it flags the ones that come back with a real question, and it never negotiates price. Boring is the point. Boring is what survives contact with a Tuesday.

For anyone looking at these businesses from the outside, and there are now roughly 66 to 76 active private equity backed platforms buying across HVAC, plumbing, roofing and pest control, this is worth more than it looks. A shop where every quote gets three touches on a schedule is not just converting better. It is a shop where the process exists outside the owner's head, which is most of what anyone is actually buying.

The part I keep thinking about

Those 48,000 quotes nobody sent.

Every one of them represents a homeowner who wanted the work done, a trip out to the property, an hour of somebody's evening pricing it, and then nothing. No rejection, no competitor, no lost bid. Just a document that stayed put.

That is the cheapest revenue in the entire business, and it has been sitting there the whole time waiting for someone to have a free Tuesday afternoon. The interesting thing about 2026 is not that AI can write a clever email. It is that the free Tuesday afternoon is no longer the constraint.

If you run a shop, I would like to know the real number: of the quotes you sent in the last ninety days, how many got a second touch? Not what the process says. What actually happened. And if you have this working, I would love to hear what your sequence looks like, because I suspect the good ones are simpler than anything I would build.

As always, tell me if you think I am wrong.

If you want to find out how many of your quotes went out and never got a second touch, and what that is worth in real dollars, start with a Revenue Audit at massivelyuseful.ai.