Speed to Lead is timing a robot

Every B2B sales team runs on a stopwatch. The clock starts when a buyer raises a hand and stops when the company responds.

Faster is better. Boards look at it. Vendors sell against it. Once a quarter somebody presents the chart showing the number coming down.

The clock is honest. The event it stops on is not the one anybody thinks it is.

When a buyer fills in a demo form, two things can happen next. Software fires, or a person reads the request and answers it.

Both put an email in the buyer's inbox. A response-time dashboard cannot tell them apart, because it records the arrival of a message rather than the arrival of a person.

Optimizing that number has a predictable effect. The cheapest way to make a response arrive sooner is to take the human out of the path.

So every gain a team books against its response time makes it likelier that nobody wrote what the buyer received. The metric quietly rewards the thing it was supposed to catch.

So we measured it.

We approached B2B companies through their own websites the way a buyer would. We submitted a demo or contact sales request, then read every email that came back with its arrival time attached.

Four readings of that data point the same way. One of them is a test any team can run this afternoon, on messages it has already sent.

The findings

All figures come from the 2023 demo response study, on a usable base of 1,685 B2B companies. Treat them as a 2023 baseline rather than a current reading. Rows counted in first replies rather than in companies say so in the cell, and the median-lag row names the 17 and 162 hand verdicts it rests on.

1. An answer in seconds, and nobody there

545 of the 1,685 first replies arrived inside five minutes, 32.3%. We read sixteen of them one at a time, and not one had been written by a person.

That count is a timestamp subtraction. It rests on no judgment about who wrote anything.

What we read were automated acknowledgments and marketing sends. Several came from a named personal address. That changed nothing about what was inside them.

The buyer had asked to be shown the product. What came back inside five minutes confirmed that a form had gone through. Nothing in it referred to anything the buyer had asked for.

Read that as a sighting rather than a rate. The replies we read were not a random draw from the 545.

The design supports something firmer. Nearly two thirds of the 545 come out of the part of the corpus where bulk mail infrastructure had already settled the question. Thirty hand reads in that part of the corpus found no person at all.

That is enough to break the metric.

A response-time report treats the 545 as the good quarter. We looked inside it and found nobody. This data does not say what the rest were.

Where a dashboard would have recorded a run of wins, there was no one there.

2. The one a person wrote arrived last

Across the emails classified by hand, the median lag was 765 minutes for the ones a person wrote and 281 minutes for automated acknowledgments. Base: 17 human verdicts and 162 acknowledgment verdicts.

Speed does not merely fail as a signal here. It runs backwards.

Human replies took nearly three times as long as the machines. Proportions built on 17 emails are directional. The direction is not in doubt.

For the buyer the order comes out upside down. The message that arrives first is the one nobody wrote.

A message somebody did write, where one comes at all, lands the better part of a day later, on top of an answer the buyer has already had.

The tail makes the same point without any arithmetic. The three fastest replies in the whole hand labeled set landed at 0 seconds, 3 seconds and 5 seconds after submission. All three were acknowledgments.

Before the hand classification we built an automated classifier to do the job at scale. It weighted speed and a named personal sender address as evidence of a person.

Measured against the hand verdicts, it called 116 emails human where the correct answer was 17.

We threw it away. The features it trusted are the features that run backwards.

One thing does separate a human reply from an acknowledgment. The sender reports an act they personally performed on this submission, and names the fact they found. That was present in 14 of 17 human emails and 5 of 162 acknowledgments. The signals that separated nothing, a named sender and a booking link among them, are set out in Speed and personalization did not separate the humans from the robots.

Nothing else came close. Asking the buyer a question separated nothing, 71% in both classes. A named personal sender address separated nothing either.

So the two signals a sales leader instinctively trusts are a fast reply and a reply from a named person. Neither carries much information about whether anybody read the request.

3. Nobody was awake when it was sent

499 of the 1,685 first replies, 29.6%, arrived before eight in the morning, from six in the evening, or at a weekend, in the sender's own time zone.

Close to a third of B2B first responses arrive while the person who supposedly sent them is asleep.

This one needs no classification at all.

Autoresponders fire when the form is submitted. People reply when they are at work. So the send timestamp alone tells you which one you are looking at, without opening a single email.

We recovered the sender's own local time from the timezone offset carried in the email header, and checked it against normal working hours.

A reply signed by a named person that lands while that person is asleep reads as attention when the buyer opens it. In this data it is a reliable sign that nobody was there.

Both of the out-of-hours replies a person had written mentioned something only a person could know. That is the same feature that separated the two classes by hand, on a base small enough to read as direction.

To run it on your own data:

1. Pull your last hundred outbound first responses to inbound inquiries.
2. Take the send timestamp of each one, in the local time of the person on the From line.
3. Mark every one that landed before eight in the morning, from six in the evening, or at a weekend.
4. Remove those from your response-time metric and recalculate.

Whatever is left is closer to the truth. The gap between what you reported and what survives that filter is the share of your performance that nobody was awake to produce.

4. The reply came back with homework

At least 46.6% of first replies demand a further step from the buyer, 785 of 1,685.

Book it yourself. Click through. Answer a question before anything else happens.

That is a floor. Link tracking stripped some booking destinations out of what we captured, so the true share is higher, and this corpus cannot say by how much.

The buyer has already done the work. They found you, they decided to talk, they filled in the form and handed over their details. What comes back asks them to do the next piece as well.

Speed does not soften it. Among the 545 first replies that arrived inside five minutes, 221 carried an imperative call to action with a link, 40.6%.

Only 34 first replies of the 1,685 sent a visible scheduling link and asked the buyer nothing at all, 2.0%. That too is a floor, for the same tracking reason.

This is the part the stopwatch hides.

A fast reply that hands the buyer a task has not reduced the wait. It has moved the wait onto the buyer's calendar, and stopped the clock on the way past.

428 first replies of the 1,685 did ask the buyer a question, 25.4%. The hand labeling shows what that is worth. Questions appear at the same rate in human and automated emails, so a question mark is a template feature rather than a sign of attention.

5. Where Speed to Lead came from

Read the phrase and notice who is in it. The company is the subject. The buyer appears as an object called a lead. The thing being timed is the seller's internal handling of its own paperwork, and the buyer's experience is assumed to follow from it.

The evidence underneath is thinner than the confidence around it.

The claims everybody repeats, about buyers going with whoever answers first, get recycled from vendor blog to vendor blog with no reachable primary source. The anchor everyone is actually citing is a 2011 Harvard Business Review article by James Oldroyd, Kristina McElheran and David Elkington on how quickly firms follow up on online leads.

It is a practitioner article rather than peer-reviewed research, and it is fifteen years old. What it measures is the seller's process.

None of that makes it wrong. It makes it a claim about internal timing, carried for fifteen years as though it described what buyers do, with nobody going back to check whether companies were doing the thing.

We checked. Better than nine in ten never put a person in front of the buyer at all.

6. What to report instead

You do not need a new dashboard.

Publish the share of buyers who never reached a person, and publish it every time you publish the median wait. A median on its own describes the best-behaved slice of your pipeline and deletes everyone else.

Ours: 93.4% of 1,685 companies never put a human in front of the buyer, against a median first reply of 19.7 minutes across all 1,685.

That 19.7 minutes is exactly the number a stopwatch would celebrate. It is sitting on top of a population where more than nine in ten buyers never met anybody.

Stop the clock on a person. Not an acknowledgment, not a bot, not a calendar link.

None of which makes an acknowledgment a bad thing. Telling a buyer you have their request, and when somebody will come back, is worth doing, and a receipt that sets an expectation and keeps it beats silence. The objection is to counting it as the answer. A receipt is not a reply.

Strip bulk sends, role addresses and unattended out-of-hours mail out of our own data, and the median across that different and much smaller population of 531 first replies is 111.7 minutes, nearly six times the all-replies figure.

Count the channel the buyer used, not the one you prefer. A company with no chat that answers an email in four minutes with a person is doing better than one running staffed chat that ignores its own form.

That is Buyer Wait Time, and the full specification is published for anyone who wants to adopt it properly.

7. Method

2,528 B2B companies were approached during 2023 with a demo or contact sales request through their own website form, from an address they did not recognize. Every one received more than 10,000 monthly website visitors. 798 forms could not be submitted and 45 errored, leaving the usable base of 1,685.

Every submission came from one buyer identity, and no form offered a free text message field. That is why merge fields carrying a first name or a company name prove nothing about who wrote a reply.

Whether a reply was written by a person was established by hand. 190 emails were classified individually, in two independent batches drawn group by group, then scaled back up to the 1,685 companies in proportion to how many emails each group held.

The deciding group was drawn twice and returned 12.3% human on the first draw and 11.7% on the second, 8 of 65 and then 7 of 60. The full method sits on the demo response study page. The classification rules, including the working-hours rule, sit on the classification rulebook page. Every figure quoted here appears with its base and its interval in the full figure set. Both studies are reported together in The Buyer Wait Time Report 2026.

The 2011 audit this article takes apart in section 5 is also the design this study borrows from, along with Drift's 2017 version of it. Both counted whether a reply arrived. This one asks who wrote it, which is set out with the rest of the lineage on the demo response study page.

Three limits matter.

Capture stored the first two emails from each company and no more. So every count of what a company sent is a floor, not a total. The never-answered share could be about a point lower than 93.4% for that reason, and no more than 2.4 points lower.

There was no fixed observation window. Collection stopped at the two-email cap rather than at a deadline, and some replies land as late as 133.6 hours after submission.

And we captured email only. A company that telephoned and never wrote is recorded here as never having answered. With 70.7% of forms demanding a phone number, that undercounts. 50 companies did write to say they had called, 3.0% of the 1,685, which puts a floor under the phone without measuring it.

Notes and bounds

The 93.4% rests on 1,573 companies out of 1,685. Its 95% confidence interval runs from 90.3% to 96.1%.

The sixteen hand read fast replies are not a random draw from the 545. Eleven of them sit in the group where bulk mail infrastructure already decides the answer, and one sits in the group that decides the study.

What the design does support is narrower, and it holds. We can place the fast first reply in a group for 387 of the 545. That splits 342 into the bulk mail group, 18 into group C and 27 into group D. So at least 63% of the fast replies come from the group where thirty hand reads found no human at all.

Proportions built on the 17 human verdicts move by about six points on a single reclassification, which is why they are reported as direction rather than as rates.

The discarded classifier, measured against the hand verdicts across all 190 emails, agreed 70 times, 36.8%. A named personal sender address appeared on 17 of 17 human emails and on 152 of 162 acknowledgments, which is why the classifier trusted it and why it carries nothing.

The working-hours rule was validated by hand before it was used. Across a first batch of 36 replies that arrived outside hours, 34 were automated. Both exceptions described something the sender had personally done, one having telephoned and left a voicemail, the other explaining that they only ran demonstrations in a language other than English.

The median first reply of 19.7 minutes has its quartiles at 4.1 minutes and 393.9 minutes.

Cut the 1,685 into cells by site type and the honest interval on each cell runs 6.5 to 10.6 points wide, so the cells overlap heavily and cannot be compared.

8. Questions

If none of the fast replies you checked was human, what should a good response time look like?

Ours is a benchmark rather than a target, and the useful question is a different one. Not how fast your first response goes out, but whether the hundredth buyer who asked for a demo this quarter ever spoke to anybody. We cannot point at a company in this study that got a person there inside five minutes, because not one of the fast replies we read by hand had been written by a person.

The fast replies you read by hand are a thin base for a headline. Why should I believe it?

Believe exactly what it supports and nothing more. Every fast reply we read was a piece of software, and nothing in the data tells us the rest were different. They are not a random draw either, so no rate of any kind can be built on them. The precise figure here is the overall one: 93.4% of 1,685 companies never put a human in front of the buyer, built from 190 hand classified emails. The fast-reply observation is a direct sighting that agrees with the median lag, with the 0, 3 and 5 second replies, and with the out-of-hours count.

Our autoresponder sets expectations. Is that not better than nothing?

It is better the way a sign reading back in five minutes is better than a locked door with no sign.

Neither one is a person. Buyers price the difference correctly, because an automated reply costs nothing to send and everybody knows it.

Would a phone call not change these numbers?

For some companies, yes. We watched email, so a company that telephoned and never wrote is recorded here as never having answered.

What the email shows is that some of them did phone. 50 of the 1,685 companies, 3.0%, sent an email in which the sender said they had called or left a voicemail. One said he had tried again just now and left a message. It is a floor rather than a rate, because a call nobody mentions afterwards is invisible to us, and so is a call to a company whose email we could not read.

So the phone was in use at a handful of these companies, and we can see it only because somebody wrote to say so.

What we cannot do is tell you whether the companies demanding a phone number behaved differently from the ones that did not. This study was built to measure one overall rate precisely, and it cannot resolve differences between site types. Anyone publishing a subgroup comparison off a base like this one, us included, is reporting the shape of their sample rather than a difference in behavior.

Is 2023 data still relevant?

Treat the response times as a baseline rather than a live reading. The direction is supported independently and more recently. We classified a matched panel of 4,265 B2B websites in 2021, again in 2025 and again in 2026. The share with no chat at all rose from 64.0% to 72.3%, which is 2,731 sites to 3,083. In 2026 it read 74.1%, with a 95% confidence interval of 69.1% to 79.1%. That interval is too wide to say whether anything moved after 2025.

The 2026 wave also went further than counting widgets. It opened each one, asked a buyer's question and waited three minutes. Of the sites whose bot had offered to bring a person in 2025, 4 of 88 judged still delivered one. The conversations behind that count are in Ask a B2B chatbot for a person and it asks for your details first.