The Most-Quoted Statistic in B2B Sales Comes From a 1942 Survey of 40 People

A ghost stat is a number that circulates as a benchmark with no retrievable primary source, or with a source that measured something else entirely.

B2B sales runs on a shared deck of them. The most-quoted follow-up statistic in the industry, the one saying 80% of sales require five follow-ups, traces to a survey run in 1942 by the Long Island chapter of an association that no longer exists. Its successor organisation, Sales & Marketing Executives International, says the sample size was fewer than 40 people.

We picked ten numbers that circulate widely in B2B sales and went looking for the primary source of each. Five hold up. Five do not. Below is the fact check, one statistic at a time, and every answer is written out underneath it, so nothing here is locked inside the game.

Holding up means the number survives tracing. It does not mean the sentence is safe to paste on its own: a figure that survives still has to travel with its sample and its date, which is most of what the rest of this article is about.

Where the follow-up statistic actually comes from

The number usually appears as a pair: 80% of sales need five follow-ups, and 44% or 48% of salespeople give up after one. It is credited to the "National Sales Executive Association".

That organisation cannot be found. A VentureBeat investigation searched Google, the Council of Better Business Bureaus, IRS records and international company registers and found no trace of it. Of more than 300 citations online, two questioned the figure. Salesforce and Microsoft were among those repeating it.

The closest real ancestor is Sales & Marketing Executives International, which publishes the origin itself: a 1942 survey of members of its own Long Island chapter, sample size under 40, no published methodology. SMEI uses the number as a teaching example about vetting statistics rather than as a finding.

A wartime survey of fewer than 40 door-to-door salesmen is still setting cadence policy in 2026.

The test we used

Three questions. A number has to answer all three before it belongs in a board deck.

1. Is there a retrievable primary source? Not a blog citing a blog. The report, with a stated sample size.
2. Is the measurement date stated, and is it recent enough to still describe your world? A real study from 2011 is still a real study. It is not a 2026 benchmark.
3. Is the definition stated? Reply rate against what denominator. Selling time counting which activities. Most disagreements between two credible numbers are definition disagreements.
4. What would this number sell if it were true? Not a test of truth, a test of how hard to look. Numbers that suit an argument circulate further than numbers that do not, so the ones that flatter the case being made are the ones worth checking first.

Statistics that fail the first question are ghost stats. Statistics that pass the first and fail the second are real research being quoted out of its own decade, which is the more common and more forgivable error. The fourth question does not tell you whether a number is wrong. It tells you which ones to check first.

What each number turned out to be

The full answer key, including the five that hold up.

Rows five and ten are the same research, from the same edition, with the same sample. One is quotable and one is not. The only difference is whether the breakdown travels with the number.

The failures are also not fabrication from nothing. Three of the five are real research wearing the wrong label: a genuine 2011 study about consumer real estate, a genuine 2017 survey, and a genuine Salesforce report from a superseded edition.

Why the bad ones survive

Tracing a number tells you whether it is true. It does not tell you why a false one lasted eighty years, and that is the more useful question, because the answer predicts which numbers to check next.

The tempting explanation is that people repeat what suits them. The duller explanation is the right one most of the time: checking is work, the number sounds plausible, and nobody is paid to be the person who checked. Never reach for malice where inattention will do.

You can watch it happen without attributing a motive to anyone. Two examples, both live, both checkable:

• RAIN Group's own blog page on how many touchpoints it takes, last updated in June 2026, states the eight-touch figure and tells you the sample: "we surveyed 489 sellers who outbound prospect." It does not say when. The sample survived the retelling and the date did not, and the date is in their own report: June and July 2017.
• Apollo's page on sales follow-up email carries both halves of the 1942 pair. "80% of sales require five to twelve follow-up attempts after the initial contact" is credited to Peak Sales Recruiting. "44% of salespeople give up after just one follow-up attempt" is credited to Yesware. Neither carries a date or a sample size, and neither names the association the figure is usually attributed to. On one page, one statistic has two different parents.

Neither page is dishonest. Both are doing the ordinary thing: quoting a figure that is already everywhere, from a source that seemed fine.

Repetition is not evidence, but it is indistinguishable from evidence at a glance.

What the incentives explain is not why anyone repeats a number, but which numbers survive being repeated. "It takes eight touches to book a meeting" is a useful sentence for anything that automates touches. "209 cold calls per appointment" makes dialling sound hopeless, which reads well to anyone selling an alternative to dialling. 3x pipeline coverage tells a team that whatever pipeline it has, it needs three times more.

None of that requires a single person to act in bad faith. It is a filter, not a conspiracy: every failure in this set overstates the difficulty of sales, and difficulty is what sales tooling is sold against, so those are the numbers that get picked up and the others quietly are not.

Almost none of these were fabricated, which is the part worth sitting with. The 209 figure is a real study, carefully done, about the wrong industry in the wrong decade. The eight-touch figure is real research whose fielding date stopped being mentioned. Nobody lied. The date fell off, because the date was the part that made the number less quotable.

Why we reach for someone else's numbers at all

That explains the supply. It does not explain why an industry full of people with CRMs is quoting strangers about its own job.

Mostly because it cannot answer the question itself. Validity surveyed 602 CRM users and administrators across the US, UK and Australia for its 2025 report on CRM data management: 76% said less than half of their organisation's CRM data is accurate and complete, and 37% said staff regularly fabricate data to tell leaders what they want to hear. Note what that source does and does not give you, since this article is about exactly that: a named publisher, a stated sample and a stated territory, but no published fielding window. Sample yes, date no. Use it accordingly.

If three quarters of teams distrust more than half of their own records, then "what is our reply rate against sends" has no answer anyone in the room will accept, and a borrowed number fills the hole. It also arrives pre-argued, which is the second reason: a benchmark is usually not being used to learn something, it is being used to settle something. Internal data gets contested by whoever it embarrasses. Nobody argues with a number from a report.

The third reason is arithmetic. A rep closes a few dozen deals a year and a manager watches a few hundred, and nothing in a sample that size separates a real pattern from a run of luck. The instinct to reach for a big-N figure is correct even when the figure is not.

The last reason is that nothing removes a bad number once it is in. A journal flags a retracted paper. A blog post from 2019 with a broken citation simply keeps ranking. There is no retraction mechanism in this field, no replication, and no cost to being wrong, which is why a survey of under 40 people from 1942 is still in circulation while the people who ran it are dead.

None of these numbers survived because they were convincing. They survived because nothing was built to kill them.

What to quote instead

The intent behind "b2b sales statistics" is usually just: give me a defensible number. Here are the ones from this set that survive the test, with what they actually measure.

That last one is worth showing in full, because it is the clearest example of why definitions decide arguments. The 7th edition of Salesforce's State of Sales breaks an average workweek like this:

Read that chart counting prospecting as selling and you get 40% selling. Read it as time actually spent with a customer and you get 22%. Both are honest. Neither is quotable without the breakdown attached, which is why we publish the breakdown every time rather than the headline.

Note which direction that error runs. Every version of this statistic in circulation is more alarming than the current edition, and a more alarming number is the one that sells sales software. The current figure is the less useful one to quote, which is exactly why it is the one worth quoting.

Why this outlives the trivia

Pipeline math compounds. A coverage target of 3x, set because someone in the 1990s won a third of their deals, quietly sets your hiring plan, your quota, and what counts as a healthy quarter. A cadence policy built on a 1942 survey decides how many times your reps call before they stop.

So the fix is not a better list of benchmarks, and this article is not one. The fix is being able to answer the question about your own team, which is the only number that was ever going to describe your market.

That is the part we work on. Spiich keeps the record current as the work happens, so the figure in the pipeline review came from the meeting rather than from somebody's recollection of it on a Friday. It will not tell you what the industry's reply rate is. It does mean you can stop asking.

Ten statistics, selected by hand because they are widely repeated rather than drawn at random, traced on 27 August 2026. The even split is how the fact check was built, not a finding about the category, and nothing here supports a claim about what share of sales statistics are unsourced. Found one we got wrong? [Tell us](/contact) and this page changes.