The demo response study, 2023, method and dataset

Full figure set: 62 B2B buyer response statistics.

93.4% of 1,685 B2B companies never put a human in front of a buyer who had asked to see the product, 1,573 of 1,685, 95% confidence interval 90.3% to 96.1%.

Everything below is how that figure was produced, what else the same collection supports, and what it cannot carry. Figures on this page were recomputed from raw source on 14 August 2026.

What we did

2,528 B2B companies were approached once each during 2023, with a demo request or a contact sales request submitted through the company's own website form.

Every company in the draw was receiving more than 10,000 website visitors a month. The constraint was set before collection, so that no finding could be dismissed as describing companies nobody visits.

Every submission came from one buyer identity: a VP of Engineering at a 200 person software company in San Francisco. The persona is described rather than named, so that the company used in the fieldwork is not identifiable. The identity carried a LinkedIn profile and the company had a website, and both matter to the classification: a reply reporting something the sender found on either is reporting a real act performed outside the form, and that is checkable rather than a claim we have to take on trust. That is what makes the test a narrow one rather than a judgment of tone: it asks whether the sender went somewhere the form was not, and names what they found. No form in the study offered a free text message field. That single fact governs the classification rules, because it means every piece of apparent personalization in a reply was reproducing data the buyer had typed into the form.

Every email that followed was captured with its time of arrival.

What bounded collection, stated plainly

Collection captured a maximum of two emails per company. No observation window was applied. There was no 96 hour cut-off, no four day watch, and no elapsed-time boundary of any kind. Collection stopped when a company's second email arrived, or when nothing further came.

Replies in this data arrive as late as 133.6 hours after the form was submitted for a first reply and 149.2 hours for a second email, on a base of 2,386 emails from 1,685 companies. Six first emails and twelve second emails land beyond 96 hours.

The consequence for the headline figure is direct. 93.4% is the share of companies that put no human in front of the buyer within the first two emails they sent. It is not the share that never did so at any point.

Human contact more often arrived at a company's second email than at its first. Of the 17 human verdicts in the hand labeled set, 10 sat at the company's second email. At an observed rate of 6 in 105 for a further email introducing first human contact, and roughly 292 companies plausibly sending a third, the cost of the cap is on the order of 17 companies, 1.0 point. If every one of the 701 two-email companies had sent a third at that rate, the cost would be about 40 companies, 2.4 points.

So the true share is likely about one point lower than 93.4%, and by the widest reading of our own data no more than 2.4 points lower. The lower end of that adjustment, 91.0%, still sits inside the published confidence interval.

Two further consequences of the cap. 3,548 response emails were captured in total, 2,386 of them from the usable base, and both counts are floors rather than totals. And the finding that 984 of 1,685 companies, 58.4%, sent exactly one email and never a second describes behavior inside a capture that stopped at two.

Disposition of the sample

The 798 forms that could not be submitted are a finding rather than attrition. A form a buyer cannot complete is a closed path. They are excluded from response rate arithmetic because no request reached the company, and the 45 errored submissions are excluded for the same reason.

Table labels read "1,685 usable submissions", never "1,685 companies approached". 2,528 were approached. All percentages below are calculated on the usable base of 1,685 unless stated otherwise.

How a first reply is defined

The source workbook stores each company's emails in fixed slots, and slot order is not time order. For 51 of the 701 two-email companies, 7.3%, the email stored second is stamped earlier than the email stored first.

A first reply is therefore the genuinely earliest email a company sent, not the email held in slot 1. Ordering by elapsed minutes and ordering by the RFC 2822 timestamp agree on all 701 two-email companies, so both readings of earliest give the same answer.

Slot order in the source workbook is not time order. For 51 of the 701 two-email companies, the email stored second is stamped earlier than the email stored first, which is why the first reply is defined as the genuinely earliest email rather than whichever sits in the first slot. Six companies enter the fast-reply set on that definition and none leaves.

That last figure needs care. The 192 first replies with no readable body and the 190 hand classified emails are unrelated sets that happen to be close in size. Neither should be printed near the other without naming both.

Content measures and their two denominators

192 of the 1,685 first replies have no readable body, 11.4%. The 284 unreadable emails across the whole usable base are missing rather than truncated: the shortest readable body runs to 120 characters and none is whitespace only.

That leaves 1,493 first replies with a readable body, and 2,102 readable emails across the usable base.

Every content figure therefore has two legitimate denominators. Each one below is published as a share of 1,685 companies, with the readable base given alongside it.

Results

Human reply. 93.4% of companies never put a human in front of the buyer, 1,573 of 1,685, 95% confidence interval 90.3% to 96.1%. An estimated 6.6% did, about 112 of 1,685, 95% confidence interval 3.9% to 9.7%. The 112 is a modeled expectation, not 112 identifiable companies, and it cannot be treated as a list.

What the interval covers, and the sensitivity that sits beside it. The interval is sampling error on three stratum rates and nothing else, with stratum B's rate fixed at the zero we observed across 30 hand reads. Stratum B holds 1,060 of the 2,102 readable emails, so half the corpus enters the estimator carrying no uncertainty at all: at a true stratum B rate of 3% the headline is 91.5%, and at 5% it is 90.3%, which is the lower bound of the published interval.

How the human reply figure is estimated. 190 emails hand classified by Terry Wilson across two independent stratified batches, reweighted to the 1,685 companies. The rules and the stratum table are on the email classification rulebook page. Nothing in the human reply figures rests on a machine classifier.

Speed of first reply. 545 of the 1,685 first replies arrived inside five minutes, 32.3%. The count is exact and depends on no classification rule.

Wait to first reply. Across all 1,685 first replies the median was 19.7 minutes, with a quarter inside 4.1 minutes and a quarter beyond 393.9 minutes. This is the time to a first reply of any kind, including automated acknowledgments, and it is not a Buyer Wait Time.

Wait once automated mail is stripped out. Among the 531 first replies that survive the three mechanical exclusions, the median was 111.7 minutes, a quarter inside 24.5 minutes and a quarter beyond 911.3 minutes. This is a filtered population and is labeled as one. The equivalent figure across all 800 surviving emails, first and second, is a median of 260.2 minutes on a base of 800. The two readings differ by more than a factor of two and only one can appear in any given sentence.

Fast replies and human contact. None of the sixteen fast replies we hand checked was written by a person. Fourteen were automated acknowledgments and two were marketing sends. The sixteen are not a random draw from the 545 that arrived inside five minutes. They are whatever fast replies fell into a sample stratified at three deliberately different fractions, so no interval built on them describes the 545. Their stratum split is B 11, C 4, D 1, and strata C and D were sampled at 18.8% and 14.6% against stratum B at 2.8%, so the sixteen over-represent the strata where human replies live and still returned none. The design-based statement is narrower and holds better: the stratum of the fast first reply is determinable for 387 of the 545, splitting B 342, C 18 and D 27, so at least 63% of the fast replies sit in the stratum where 30 hand reads found no human. Four of the sixteen came from named personal addresses and every one of those four was still an acknowledgment. The three fastest replies in the whole hand labeled set landed at 0 seconds, 3 seconds and 5 seconds, and all three were acknowledgments.

Speed points away from human contact. Among the 190 hand labeled emails the median lag was 765 minutes for the ones a person wrote and 281 minutes for automated acknowledgments. Bases: 17 human verdicts and 162 acknowledgment verdicts. Human-class proportions on that base are directional, because a single reclassification moves any of them by about six points.

Arrival outside working hours. 499 of the 1,685 first replies arrived outside the sender's own working hours, 29.6%: before 08:00, from 18:00, or at a weekend in the sender's local time taken from the timezone offset in the email header. Each company contributes exactly one first reply, so the count of first replies and the count of companies are the same number, but the unit is first replies.

Follow-up behavior. 984 of the 1,685 companies sent exactly one email and never a second, 58.4%. 701 sent two, 41.6%. Inside a capture that stopped at two.

Questions asked. 428 of the 1,685 first replies asked the buyer a question, 25.4%. On the 1,493 with a readable body, 28.7%. A sentence qualifies if it ends in a question mark, contains a second person pronoun, runs to at least three words, and does not match one of fifteen template patterns such as "Questions?" or "Ready to get started?". Hand checking 25 flagged questions found 24 correct.

Further step demanded. 785 of the 1,685 first replies demand a further step from the buyer, 46.6%. On the 1,493 with a readable body, 52.6%. A first reply demands a further step if it contains an imperative call to action with a link, a booking link, or a question directed at the buyer. Nothing happens next unless the buyer does something. Published as a floor, because link tracking hides some of the booking links that feed it. The narrow reading, an explicit action only, gives 461 of 1,685, 27.4%.

Call to action in fast replies. 221 of the 545 fast first replies carried an imperative call to action with a link, 40.6%. On the 515 fast replies with a readable body, 42.9%. The broader further-step measure on the same 545 gives 253, 46.4%. The two quantities are different and are not interchangeable.

Scheduling links, published as a bound. Between 3.9% and 27.8% of first replies sent the buyer a scheduling link, 66 to 468 of 1,685. Link tracking removed scheduler destinations from the record, so the lower figure counts only links still visible and the upper figure adds every reply whose text says a calendar link was supplied. 66 first replies carry a visible scheduler host across nineteen providers. 706 first replies have no URL recorded at all, 41 carry URL lists made entirely of tracking redirects, and 180 have no URL recorded but state in prose that a calendar link was given. Those 180 are the direct evidence of suppression. This corpus cannot narrow the range further.

Scheduling link with no question, lower bound only. 34 of the 1,685 first replies sent a visible scheduling link and asked the buyer nothing at all, 2.0%. This is a floor for the same tracking reason, and it is the observable end of the bounded figure above.

Marketing sends, published as a bound. Between 8% and 26% of companies sent at least one marketing email that never acknowledged the request, best estimate 17%, 289 of 1,685. The blast rates in the three sampling strata are 8 of 30, 0 of 35 and 3 of 125, and applied through the same estimator that produces the 93.4% they give 289.5 companies, 17.2%, with a 20,000 draw bootstrap interval of 8.2% to 26.4%. The quantity counts any email in the capture, not the first response, because the hand labels cannot be resolved to first-email position across the population.

Reply length. First replies run to a median of 136 words, a quarter under 87 and a quarter over 218, range 10 to 1,363, on the 1,493 first replies with a readable body. After stripping quoted replies and bulk footers the core text runs to a median of 115 words.

Structural attributes across the readable corpus. Across all 2,102 readable emails in the usable base: bulk mail infrastructure 996, 47.4%; role sender address 609, 29.0%; imperative call to action with a nearby link 631, 30.0%; asks the buyer a question 661, 31.4%; booking link 90, 4.3%; form echo 84, 4.0%; specific reference beyond merge fields 56, 2.7%; claims a personal action 125, 5.9%. The call to action rule was adjudicated across four independent hand samples during development, reaching 39 of 40 correct on the final independent draw.

Two of those figures are floors and are published as floors. The booking figure is one, because link tracking rewrote some scheduler URLs before we could count them. The specific reference figure is the other, and it is the weaker of the two. Detecting that a sender referred to something outside the form requires reading an email for meaning, and the automated pass does that badly. A hand census of the same corpus found 55 emails in which the sender reported a phone call or voicemail they had personally placed, which is the clearest instance of the feature that exists, and the automated pass flagged none of the 55. Adding those alone takes the count to at least 111 emails, 5.2%, and there is no reason to think a reported call is the only phrasing the pass misses. Read 2.7% as a floor of unknown depth. The hand-classified figure for the same feature, 14 of 17 human replies against 5 of 162 acknowledgments, is the one that carries the finding, and it was produced by reading rather than by detection.

Emails surviving the mechanical exclusions. 800 of the 2,102 readable emails survive, 38.1%, and 536 of those are first replies. An email is removed if bulk mail infrastructure is present, or the sender address is a role address, or it arrived outside the sender's own working hours and describes nothing the sender personally did. Surviving is not evidence a person wrote the email. The rules can prove an email was not personally written. They cannot prove one was.

Form design. 1,192 of the 1,685 forms demanded a phone number, 70.7%, and 493 did not. 292 forms asked for three fields or fewer and 170 asked for eight or more. 52 forms used a CAPTCHA, 3.1%.

Chat software. 766 of the 1,685 companies had chat software installed at the time of submission and 919 did not. The 919 folds in the two companies whose chat value is missing.

What this study cannot support

It cannot resolve differences between site types. The sample was built to measure one overall human reply rate precisely, and it does that. It does not have the resolution to compare subgroups. Estimating each cell honestly gives intervals 6.5 to 10.6 percentage points wide that overlap heavily, eleven of the eighteen cells contain no human verdicts at all, and one cell holds zero hand labels against 124 population emails. Settling the chat installed against no chat comparison would take roughly 880 further hand classifications. The form length comparison cannot be settled at any sample size, because only 73 long-form emails exist in the stratum that decides it. No subgroup never-answered rate is published and no significance test is published on one.

We hold no outcome data of any kind. Nothing in this study records what happened to any buyer after the reply arrived. No revenue, no meeting held, no deal, no churn, no satisfaction. The study measures what a buyer received and nothing about what it was worth to anybody. Any sentence claiming that answering pays, or that not answering costs, is not supported by this data.

Figures affected by link tracking publish as bounds, not points. Scheduling link share and marketing send share are published as ranges for that reason, and the further-step and booking-link figures are published as floors. A point estimate on any of them would be asserting a precision the record does not hold.

Email only. No telephone calls were captured. A company that answered by telephone and never emailed is recorded here as never having put a human in front of the buyer. The size of that undercount is unmeasured, and it has a floor: 50 of the 1,685 companies, 3.0%, wrote to say they had called or left a voicemail. 70.7% of the forms demanded a phone number, which makes the limitation material rather than theoretical, and it is the strongest single argument against the headline figure.

153 companies are counted as never-answered by arithmetic rather than by observation. 153 of the 1,685 contribute no readable email to any stratum, so the estimator, which multiplies one minus the stratum rate over the emails a company sent, assigns each of them a probability of one that no human answered. Nobody read an email from those companies, because there was none to read. They are 9.1% of the base. If they answered at the same rate as the other 1,532, the headline falls from 93.4% to 92.7%, a difference of 0.66 of a point.

One request per company. Each company was approached once. The study measures what happened to one buyer at each company, not the company's average behavior across many buyers.

Form failures were not individually diagnosed. The 798 unsubmittable forms were not investigated case by case. Some share will be bot defenses blocking an automated research script rather than obstacles a human buyer would hit. The 31.6% is published as an upper bound on buyer-facing form failure.

Age of the data. Collection ran during 2023. The figures are a baseline for comparison, not a current reading of the market. Any claim about the present state of demo response needs new collection.

Every unmeasured bias in this study runs in the same direction, toward overstating the never-answered rate. Four of them, on the same base of 1,685. The two-email cap costs about one point and no more than 2.4. The 153 companies with no readable email cost 0.66 of a point. The email-only capture costs an unknown amount and is the one we can put no size on at all. And the governing classification test refuses an email unless it proves a person wrote it, so a genuine human reply that happens to read like a template is counted as an acknowledgment, and nothing is ever counted the other way. Not one of the four moves 93.4% up. So 93.4% is the top of the range this collection supports rather than the middle of it. The direction of the finding does not move on any of that, because three independent lines agree: the hand classification rate, the share of first replies arriving outside the sender's own working hours, and the matched website panel. What moves is the precision.

Prior work this design sits on

Inbound response audits are not new, and this one is not the first to submit forms and time what comes back.

Oldroyd, McElheran and Elkington audited 2,241 US companies for Harvard Business Review in 2011, measuring how quickly firms followed up on online leads. Drift ran a comparable design on 433 B2B companies in 2017. Both counted whether a reply arrived and how fast.

What is new here is the question. Those studies measured arrival. This one classifies, by hand, whether a person wrote what arrived, which is a different variable and the one the Speed to Lead metric has always assumed rather than tested. The matched five year website panel is also new, and it is the only part of this research that can watch the same companies change their minds.

This section is the canonical statement of prior work for the Buyer Wait Time set. The articles point here rather than restating it.

Reproduction sources

How to cite this

GTM Clarity, The demo response study, 2023, method and dataset, 2026. Demo response data collected 2023. Research directed by Terry Wilson, fieldwork by the GTM Clarity research team.

The findings drawn from this study are reported in The Buyer Wait Time Report 2026.