The 42-hour problem: fifteen years of knowing better, and the median never moved.
In 2011 an audit of 2,241 companies clocked the average web-lead response, among companies that answered, at 42 hours. In 2026 the largest public benchmark reports a median of 42 hours. Same digits, different statistics, one conclusion: the decay of a buyer's attention is the best-documented number in sales development, and most funnels are still built to ignore it.
A 2011 audit of 2,241 companies found a 42-hour average web-lead response, the 2026 benchmark reports a 42-hour median, and the practice change costs one afternoon: measure your own median minutes to first human response and put one name on the number.
Fifteen years ago, three researchers timed how long 2,241 US companies took to answer a web-generated lead. The average, among companies that answered within 30 days at all, was 42 hours [1]. This January, the largest public benchmark of B2B response times closed its one-year window and reported its headline figure: a median of 42 hours [2].
Fifteen years. In between came sales engagement platforms, routing tools, calendar links, enrichment vendors, and now AI agents that can answer in seconds. The number on the board is the same number.
The two 42s are not the same statistic, and this article will not pretend they are. One is a conditional average from an audit published in Harvard Business Review. The other is a vendor median from a page that states its own sample size two different ways. But they point in the same direction, and the direction is this: most companies answer the most expensive signal in their funnel somewhere between tomorrow and never. So we read both studies properly, worked the numbers in front of you, and followed them to Monday morning.
The audit, read properly
What everyone cites as one study is two. In March 2011, Oldroyd, McElheran, and Elkington published an audit: 2,241 US companies each received one web-generated test lead, and the researchers timed the response [1]. The companies did not know they were being measured: no survey optimism, no self-report, just a stopwatch.
The distribution is the finding. 37 percent responded within an hour. 16 percent took between one and 24 hours. 24 percent took more than 24 hours. And 23 percent never responded at all [1].
Add the buckets and the shape gets blunt: 37 plus 16 is 53 percent inside a day, and 24 plus 23 is 47 percent that took longer than a day or went silent forever. Nearly half of the audited market treated an inbound hand-raise as optional.
The famous 42 hours is narrower than its reputation: an average, not a median, computed only among companies that responded within 30 days [1].
The second study is the one the first version of this article never told you about. Alongside the audit, the authors analyzed 1.25 million sales leads received by 42 US companies, 29 B2C and 13 B2B [1]. The outcome was lead qualification, which the paper defines as “having a meaningful conversation with a key decision maker” [1]. Firms that attempted contact within an hour of the form fill were nearly seven times as likely to qualify the lead as firms that tried just an hour later, and more than 60 times as likely as firms that waited 24 hours or longer [1].
From 1.25 million leads across 42 companies. A likelihood ratio, not a percentage-point gap: the paper reports no baseline qualification rate, so 60x cannot be converted into revenue.
Hold on to that last clause. The 7x and the 60x are likelihood ratios, and sixty times a small baseline can still be small. The pair of studies establishes a steep, early decay in the value of a contact attempt, and a market whose median behavior ignores it. It does not establish a dollar figure. And they never mention five minutes: the fastest bucket in the 2011 text is within an hour [1]. Every five-minute claim you have seen attributed to this paper came from somewhere else.
Fifteen years later, the count repeats
You would expect fifteen years of tooling to have moved the distribution. The most recent large benchmark suggests it has not. Between January 2025 and January 2026, Artemis GTM logged 253,817 inbound lead submissions and reported a median response time of 42 hours, with only 7 percent of companies consistently responding within five minutes [2].
The mix is the 2011 shape, recounted at scale: 7 percent of submissions answered inside five minutes, 12 percent in five to 30 minutes, 15 percent in 30 to 60 minutes, 31 percent in one to 24 hours, and 35 percent past 24 hours [2]. The page’s callout that 66 percent take over an hour checks out as the sum of the two slowest buckets: 31 plus 35 is 66 [2].
The benchmark also reports a conversion ladder by response bucket: 21 percent of sub-five-minute leads became opportunities, falling to 2.3 percent for leads answered past 24 hours [2].
Now the caveats, because this is a vendor page and it earns them. The sample is stated as 1,247 B2B SaaS companies in the methodology box and as 127 companies in the citations section, a 10x discrepancy the page never reconciles [2]. The distribution rows are labeled companies, but the counts sum to 17,767 + 30,458 + 38,073 + 78,683 + 88,836 = 253,817, which is exactly the lead-submission total, so we treat them as submissions [2]. The conversion ladder comes with no controls, no confidence intervals, and no operational definition of what counted as an opportunity. And the page sells a $349 product against its own finding, a disclosed commercial interest [2].
So is the 42-to-42 match a finding? No. One number is a conditional average over 2011 audit responders, the other a 2026 vendor median: different statistics, different methods, different populations [1] [2]. Treat the identical digits as a coincidence with good marketing instincts. What is not a coincidence is the shape.
| Study | Year | Sample | Central finding |
|---|---|---|---|
| Oldroyd et al., Harvard Business Review | 2011 | 2,241 US firms audited | Average response time 42 hours; 37% responded within an hour |
| Artemis GTM speed-to-lead benchmark | 2026 | 253,817 lead submissions; 1,247 B2B SaaS companies as stated, though the page elsewhere says 127 | Median response time 42 hours; 7% consistently respond within 5 minutes |
The numbers, worked
What is the slow half of that distribution worth? The honest answer is a range, and the math desk built it from the benchmark’s own table. Take a 1,000-lead month, distribute it across the benchmark’s response-time mix, and apply the reported conversion ladder to each bucket [2].
The load-bearing word in that example is causal. The benchmark asserts that speed drives the conversion gap, but it ran no experiment, and companies that answer in five minutes likely differ in staffing, tooling, and lead quality. Vary that one dial and the gap moves with it. If half the gap is causal: 144.15 x 0.5 = 72 added opportunities per 1,000 leads. A quarter: 144.15 x 0.25 = 36. None, pure selection: zero. And if the true sub-five-minute conversion were half the reported 21 percent, the all-fast scenario yields 1,000 x 0.105 = 105 opportunities and the gap shrinks to 105 - 65.85 = 39.15, call it 39.
| Causal share of the conversion gap | Added opportunities per 1,000 leads | Monthly pipeline at $10,000 per opportunity |
|---|---|---|
| 0% (pure selection) | 0 | $0 |
| 25% | 36 | $360,000 |
| 50% | 72 | $721,000 |
| 100% (vendor's implicit claim) | 144 | $1,442,000 |
We also tried to reproduce the vendor’s own ROI headline, which claims that moving 100 leads a month at $10,000 each from 24-hour to five-minute response adds $1.8 million in annual pipeline [2]. It does not reproduce.
Three traps to carry out of this section. First, likelihood ratios are not percentage-point gaps: the 2011 ratios cannot be converted into revenue without a baseline the paper never gives [1]. Second, framing direction: the 2011 text says fast firms were more than 60 times as likely to qualify, and restating that as slow firms being 60x less likely changes the denominator, so quote the original direction only [1]. Third, author-chosen baselines: the benchmark’s 900 percent increase headline comes from endpoints of 21 and 2.3 percent, and 21 / 2.3 = 9.13, a ratio of 9.13x, which is an 813 percent increase, not 900 [2].
Why the buyer disappears
Neither featured source tests a mechanism. The 2011 authors state plainly that the causes of slow response are unsettled and that they were conducting further research to understand them [1]. The 2026 page asserts causation without any identifying design [2]. So what follows is candidate explanation from adjacent literature, labeled by strength, not a finding of either study. One boring confound sits beside all of it: companies that answer fast may simply be better run in every other way.
The first candidate is memory decay. People forget fast, and most of the forgetting happens early: the decay curve is steep at first and flattens later, a result replicated under controlled conditions in the modern literature [3]. The shape matters, because it matches the qualification cliff, where an hour of delay costs more at minute ten than at hour 30. Established as a memory phenomenon. Plausible, not established, as the driver of lead decay specifically: the bridge from laboratory word lists to web forms is an analogy. By the time you call back two days later, the buyer has half forgotten they asked: you are not returning their call, you are cold calling someone who vaguely recognizes your name.
The second is hot-state decay. The moment of form submission is a hot state. The problem is salient, the frustration or curiosity is live, and the buyer is motivated to act now. Visceral states have outsized influence on behavior while active and fade quickly, after which the same person evaluates the same decision coldly and acts differently [4]. Established as a decision-making principle. Plausible as applied here, since the original work studied drive states and emotion, not B2B purchase intent. The buyer at 10:04 and the buyer at 10:04 plus 42 hours are, behaviorally, two different people. Only the first one wants to talk.
The third runs in the opposite direction, and is worth keeping precisely for that reason. Under information asymmetry, buyers read observable behavior as a signal of unobservable quality [5]. Response speed is the first observable behavior a vendor produces, so a five-minute response signals operational competence before any conversation happens, and a two-day response signals the inverse. Established as an economic theory of signaling. Plausible as an extension here, since the original modeled labor markets. Unlike the decay stories, this mechanism predicts damage even when the buyer still remembers you. Your response time is the first demo the buyer ever sees of how you operate.
None of these has been tested on sales leads in a controlled design we could verify. They are reasons the correlation is believable, not proof that it is causal.
What to do Monday morning
Strip the article to what survives the caveats and three things remain: a steep early decay documented in likelihood ratios [1], a market mix that has not visibly improved in fifteen years [2], and the fact that your own number is measurable in an afternoon. Speed-to-lead looks like a Motion problem, the discipline of the sequence. It is a Measurement problem first: most teams cannot state their median time to first human response from their own data, and a number nobody owns is a number that does not move. One move per rung of the maturity ladder.
Manual. Run the 2011 audit on yourself. The original method was one test lead and a stopwatch [1]. Monday morning, submit one lead through your own web form from a personal email and time the first human response. You will land somewhere in the audit’s distribution, where 37 percent answered within the hour and 23 percent never did [1]. One data point, one hour of effort, and you know which half you are in. Do not buy anything until you have this number.
Assisted. Instrument the timestamp gap. You have a CRM, so the data exists. Pull lead-created time and first-touch time for the last 90 days and compute your median and 90th percentile. Your number is the only one that matters. Report it weekly.
Orchestrated. Rebuild the worked example with your own mix. Bucket your actual response times the benchmark’s way and apply your own conversion by bucket, not the vendor’s endpoints. The benchmark mix puts 31 percent of leads in the one-to-24-hour bucket and 35 percent past 24 hours [2], and those two buckets are where routing changes pay. Fix the single largest slow bucket first. If your orchestration cannot tell you the mix, that is the Monday finding.
Autonomous. Pressure-test the attribution before crediting the agent. If an AI agent now answers your leads in minutes, your vendor report will show a conversion lift. Before you book it, hold out a random slice of leads from the agent for 30 days and measure the real delta against the causal-share table above: 144 added opportunities per 1,000 leads at fully causal, 72 at half, 36 at a quarter, zero at none. A holdout is the only way to know whether your speed or your selection is converting.
Where this lands, industry by industry
Every entry below is a parameterized archetype: neither source names a company, so neither do we. The derived counts scale the worked example, 65.85 expected opportunities per 1,000 leads at the benchmark mix versus 210 all-fast [2], and the audit’s distribution supplies the failure patterns [1]. Each range inherits the causal caveat in full: the floor of every gap is zero.
Limits and caveats
What the 2011 work does not establish: the audit measured response latency to a single test lead per company, the qualification likelihood ratios come from a separate observational study of 42 firms, and neither measures revenue, win rates, or anything past the meaningful conversation [1]. The selection method for the 2,241 companies is not described, the fieldwork date is not stated, and the 42-hour average excludes never-responders by construction [1].
What the 2026 benchmark does not establish: causation, anywhere. It is not peer reviewed, it carries the unresolved 1,247 versus 127 sample contradiction, and its third-party citations mirror its own headline numbers exactly (its Velocify summary repeats both the 42 hours and the 7 percent), so they are not independent confirmation [2]. Its own table does not fully reconcile either: the stated overall conversion of 7.8 percent does not match the submission-weighted bucket average, 0.07 x 21 + 0.12 x 13 + 0.15 x 8 + 0.31 x 5 + 0.35 x 2.3 = 6.59 percent, a gap the page does not explain [2].
Transfer risks: a topline rate is not your rate. The benchmark’s own vertical table shows conversion ranging from 6.1 to 11.3 percent [2], and the mechanisms above have never been tested on sales leads in a controlled design. What would change our read: a randomized speed experiment, or your own holdout data showing the delta when the same lead source gets a faster clock.
None of this tells you the revenue elasticity of your own response time. It tells you the prior is strong, the direction is settled, and the measurement on your own funnel costs one afternoon. The Pressure Test we run on every spotlighted company in this magazine measures precisely this number, with the protocol published on the methods page.