Everyone Quotes the 21x. Six Companies in 2007 Produced It.
The lead-response numbers the sales industry runs on trace back to one 2007 study of six companies. Here is what each study actually measured, what has been reproduced since, which famous figures have no source at all, and what an AI voice agent changes about any of it.
The number has a footnote. Almost nobody reads it.
Call a lead within five minutes and you are twenty-one times more likely to qualify it than if you wait thirty. You have seen that line. It is on vendor landing pages, in pitch decks, in the opening slide of most sales-enablement talks given since about 2012.
It is a real finding from a real study. The study looked at six companies. It ran in 2007. It measured something narrower than the slide says, and it states in its own text that it did not measure the thing people usually cite it for.
None of that makes it wrong. It makes it a number with conditions attached, and the conditions are where the useful information lives.
This is a compilation of what is actually published on lead response time, with every figure traced to a page we fetched. Each source is marked vendor-published, industry benchmark or academic/press so you can weigh it yourself. Where a widely-repeated claim has no traceable source, that is listed too, because a claim nobody can source is a claim you should stop repeating.
For the cost side of the same question, the companion piece is the cost per resolved call model.
Correction one: it is not an MIT study
The paper is usually cited as "the MIT study" or "the MIT/InsideSales study". MIT did not publish it. InsideSales.com did. James Oldroyd was a faculty fellow at MIT Sloan when the work was done, which is where the association comes from. The copyright line on the document reads "© 2007 All rights reserved, Dr. James Oldroyd and InsideSales.com, Inc."
The second author is Dave Elkington, then CEO of InsideSales.com, a company selling lead-response software. That does not invalidate the data. It does mean the correct class is vendor-published, not academic.
The third name that appears in reprints, "McClary", does not exist. The third author on the later Harvard Business Review paper is Kristina McElheran, and she is on that paper only.
What the 2007 study actually says
Source: The Lead Response Management Study, Oldroyd and Elkington, presented 16 October 2007 at the MarketingSherpa Business-to-Business Demand Generation Summit. Class: vendor-published. Full text as captured, and a publisher summary that reproduces the numbers without ever stating the year.The sample, in the paper's words: "We examined 3 years of data across six companies that generate and response to web leads, from over fifteen thousand leads and over one hundred thousand call attempts."
Fifteen thousand leads is a decent number. Six companies is not. Every one of the headline figures below rests on those six.
| Finding, verbatim | What it measures | Class |
|---|---|---|
| "The odds of calling to contact a lead decrease by over 10 times in the 1st hour." | Odds of reaching a live person | Vendor-published |
| "The odds of calling to qualify a lead decrease by over 6 times in the 1st hour." | Odds of qualifying | Vendor-published |
| "The odds of contacting a lead if called in 5 minutes versus 30 minutes drop 100 times." | Odds of reaching a live person | Vendor-published |
| "The odds of qualifying a lead if called in 5 minutes versus 30 minutes drop 21 times." | Odds of qualifying | Vendor-published |
| "from just 5 minutes to 10 minutes the odds decrease by 5 times" | Contact | Vendor-published |
| "After 20 hours every additional dial your salespeople make actually hurts your ability to make contact to qualify a lead." | Contact-to-qualify | Vendor-published |
Now the footnotes, which the reprints drop.
"Contact" means a call that connects with a live person and lasts a defined number of seconds. The paper's own note: "this time was different for each company's data ranging from 2 minutes to 6 minutes." So the threshold moved between companies. "Qualify" means a stage where the lead is willing to enter the sales process. The paper's note: "Each company involved in the study had their own way to indicate a qualified lead." Six companies, six definitions. And the line that should end most arguments: "This study did not address close ratios."If you have seen the 21x rendered as "21 times more likely to sign that client", you have seen a number quietly promoted to an outcome it never measured. That exact rewrite is live on at least one legal-marketing reprint.
Correction two: the HBR paper is two studies, not one
Source: "The Short Life of Online Sales Leads", Oldroyd, McElheran and Elkington, Harvard Business Review, March 2011. Class: academic/press, with the caveat that the underlying data is InsideSales platform data and Elkington is its CEO. The hbr.org page returns only the paywalled shell to a fetcher; the full text is readable in the Internet Archive capture.People cite this paper as one finding. It contains two separate pieces of work on two different datasets, and merging them is the most common error in the whole literature.
Study A, the audit. Verbatim: "We audited 2,241 U.S. companies, measuring how long each took to respond to a web-generated test lead. Although 37% responded to their lead within an hour, and 16% responded within one to 24 hours, 24% took more than 24 hours, and 23% of the companies never responded at all. The average response time, among companies that responded within 30 days, was 42 hours."Read the last clause. The 42 hours is conditional on responding within thirty days. The unconditional average, including the 23% who never answered, is not published and is worse.
Study B, the outcome study. Verbatim: "a phenomenon we explored in a separate study, which involved 1.25 million sales leads received by 29 B2C and 13 B2B companies in the U.S. Firms that tried to contact potential customers within an hour of receiving a query were nearly seven times as likely to qualify the lead (which we defined as having a meaningful conversation with a key decision maker) as those that tried to contact the customer even an hour later, and more than 60 times as likely as companies that waited 24 hours or longer."Three things worth pinning down.
The 7x comparator is "even an hour later". Not "a day later". Not "never". One hour versus two.
The 60x figure comes from the 1.25 million-lead sample, not from the fifteen thousand-lead 2007 study. Different data, different years, different companies. They get merged constantly.
And in this paper, "qualify" is defined as "having a meaningful conversation with a key decision maker". In the 2007 paper it meant "willing to enter the sales process", defined separately by each of six companies. Same word, two definitions, four years apart.
What has been reproduced since
The 2007 magnitudes have never been independently replicated. The direction has, repeatedly, by people running the same secret-shopper method on larger samples.
| Study | Sample | Year run | Headline | Class |
|---|---|---|---|---|
| XANT / InsideSales Lead Response Report | 9,538 companies audited of 14,061 attempted | 2013 | Median first phone response 3 hours 8 minutes; mean 61 hours 1 minute; 47% never responded | Vendor-published |
| Drift Lead Response Survey | 433 B2B SaaS companies | 2017 | 7% responded within five minutes; 55% never responded in five business days | Vendor-published |
| Drift Lead Response Report 2018 | 512 B2B companies | April 2018 | 10% responded within five minutes; 58% never responded | Vendor-published |
| RevenueHero B2B lead response | 1,000 B2B SaaS companies | March 2024 | 365 responses received in total; average response among responders 1 day 5 hours 17 minutes | Vendor-published |
| Hennessey Digital Lead Form Response Time Study | 1,333 US law firms | Q1 2025 | Median response 13 minutes; 25% under five minutes, up from 13% in 2021; 26% never respond, down from 40% in 2021 | Industry benchmark |
Two things in that table deserve attention.
The Hennessey series is the only clean trend. Same method, same sector, 2021 against 2025. Response times improved and non-response fell by fourteen points. It measures response time only. It publishes nothing about conversion. The RevenueHero 2024 result reads better than it is. 172 of 1,000 companies responded in under two minutes, which is 17%, far above Drift's five-minute rates from six years earlier. Under-two-minutes at that share is almost certainly automated email acknowledgements rather than a person. The 2007 study's "contact" required a live human on a call for two to six minutes. These are not the same event and should not be plotted on the same axis. And every one of these numbers is a best case. XANT submitted its test leads only between 8am and 5pm local time, stating that this allows "companies to have the best chance at responding quickly." Hennessey submitted only between 10am and noon. Both tested businesses at their most attentive. Hennessey still found a quarter of firms never replying at all.The claims with no source
These circulate as facts. We went looking for the primary document behind each one and did not find it. Treat them as folklore until someone produces the study.
"78% of customers buy from the company that responds first." Attributed universally to a "Lead Connect survey". No report, no methodology, no date, no sample. Every trail ends at an aggregator blog citing another aggregator blog. "Calling within one minute increases conversions by 391%." Attributed to Velocify's The Ultimate Contact Strategy, around 2012. Velocify was absorbed into ICE Mortgage Technology and the document is gone. Only search snippets survive. Every after-hours lead-volume percentage. "65% of form submissions arrive after hours." "52% of calls." "53% of dealer leads arrive outside 9 to 6." None of them resolves to a primary source. If you need an after-hours argument, build it from your own phone logs, which you have and nobody else does. "62% of business calls go unanswered." This one is traceable, to a 2016 post by a local-marketing agency reporting 85 businesses across 58 industries over thirty days. Class: vendor-published. Their split: 37.8% answered, 37.8% to voicemail, 24.3% no response at all. It is real data. It is n=85, and the page contains no after-hours breakdown whatsoever, despite being cited as after-hours evidence constantly. If you use it, use the sample size with it. There is no peer-reviewed academic literature on speed-to-lead. We looked. The only academic-archive record is a catalogue entry for the HBR article itself. The entire evidence base on this topic is vendor-published, which is a finding in its own right.What a voice agent changes
One thing, and it is the thing the studies measure.
An agent that answers on the first ring puts every inbound call in the zero-to-five-minute bucket by construction. Not on average. Every call, including the one at 21:40 on a Saturday, which is the call the Le Marquier support line was built around. There is no queue to be at the back of, no rep to be in a meeting, no Monday morning.
The same applies in reverse. A webhook that fires the moment a form lands and dials inside sixty seconds is a response time you set rather than one you hope for. Our own outbound analysis covers where that works and where it backfires.
That is the whole claim. Speed becomes a configuration rather than a staffing outcome.
What it does not change
It does not make an acknowledgement into a contact. If the agent answers and the caller hangs up at nine seconds, you have a fast non-event. The RevenueHero under-two-minute bucket is the warning: measure conversations that lasted, not responses that fired. Set your own duration threshold the way the 2007 study's six companies each set theirs, and be honest about where you set it. It does not qualify anything the business has not defined. Every source here defines "qualified" differently, because qualification is a commercial decision, not a technical one. An agent applies the definition you give it, precisely and at volume. A vague definition applied precisely is still vague. It does not touch close rates. The 2007 study did not address close ratios, and neither does any study in the table above. Anyone selling you a speed-to-lead number denominated in revenue is extrapolating past the evidence. It does not fix a broken destination. Responding in four seconds into a CRM nobody opens moves the failure downstream. The signs a phone system has been outgrown are usually about what happens after the call connects. And it does not make the 21x yours. That figure belongs to six companies in 2007 with six private definitions of qualification. Your own before-and-after, measured on your own leads with your own threshold, is worth more than any number in this post.How to measure it on your own line
Four counters. None of them need a vendor.
Time from lead creation to first human-grade response, where you define human-grade. Log the distribution, not the average. The averages in every study above are dragged around by long tails, which is why XANT's median was 3 hours 8 minutes against a mean of 61 hours. Share of inbound that never got a response at all. This is the number the industry buries and the one that moves most when you change anything. It sat at 47% in 2013, 58% in 2018 and 26% in 2025 depending on who was measured. Conversation duration, not response count. Pick a threshold, hold it fixed, and count only what clears it. Your own by-hour arrival curve. Since the published after-hours figures do not exist, yours is the only evidence there is. It takes one query against your call log, and it will tell you more about whether you need coverage at 21:00 than any statistic in this article.Then compare. Against yourself, six weeks apart. The published numbers set the direction. Your own numbers decide the spend.
Every figure in this article links to a page we fetched on 19 September 2026. Sources are labelled vendor-published, industry benchmark or academic/press. Where a claim could not be traced to a primary source, we have said so rather than repeat it. If you find a primary document for any of the untraceable claims above, we will add it and say who found it.