Lead Response Time: The 5-Minute Rule Came From 2007 Phone Dials. Does It Survive a 2am Telegram DM?

A lead messages your Instagram at 02:14. Your first human reply goes out at 09:40, seven and a half hours later. Every article about lead response time will tell you that deal is already gone, and nearly all of them trace back to a study published in 2007 that measured outbound phone dials and said, in its own text, that it did not look at close rates.

That study is not wrong. It is being quoted about a situation it never observed.

This post does two things. It goes back to the primary sources behind the five minute rule and reads what they actually say, including the parts that never survive the trip to an infographic. Then it asks the question those researchers could not have asked in 2007: what happens to the decay curve when the lead does not arrive as a web form at 2pm on a Wednesday, but as a Telegram message at 2am from someone eight timezones away?

Where the five minute rule actually comes from

The source is the Lead Response Management report, run by Dr. James Oldroyd and published by InsideSales.com. It was presented at MarketingSherpa's Business-to-Business Demand Generation Summit on October 16, 2007. It is the origin of the two numbers you have seen a thousand times.

Here is the finding, quoted exactly:

The odds of contacting a lead if called in 5 minutes versus 30 minutes drop 100 times. The odds of qualifying a lead if called in 5 minutes versus 30 minutes drop 21 times.

That is where 100x and 21x come from. Note what the sentence actually says: called. Not emailed, not messaged. Called. The unit of analysis is a phone dial placed by a sales rep to a person who filled in a web form.

The report is more specific than its reputation. It also states that "from 5 minutes to 10 minutes the dial to qualify odds decrease 4 times," which is a striking claim: in this data, five extra minutes cost you three quarters of your odds. If that is true of your channel, nothing else in your sales process matters nearly as much. It is worth asking whether it is true of your channel.

The dataset: three years of data across six companies, over fifteen thousand leads and over one hundred thousand call attempts. Six companies is not a large sample of companies. It is a large sample of dials from a small sample of businesses, which is a different thing, and it matters for how far you should generalize.

The study is candid about this in a line that never gets quoted. Oldroyd, per the report, "emphasizes that he finds these clear patterns in the data only when data from several companies is combined together." The effect is visible in the pooled data. It was not reliably visible inside any single company. If you have ever run this analysis on your own pipeline and found nothing, that is consistent with the original research rather than a contradiction of it.

The sentence that should have ended the infographic industry

Buried in the same document, describing the design of the study, is this:

This study did not address close ratios.

The most-cited speed-to-lead research in existence did not measure whether speed makes you money. It measured two things: whether a dial reached a human (contact), and whether that call turned into a qualifying conversation (qualification). Revenue was never in the model.

This is not a gotcha. Contact and qualification are perfectly reasonable things to study, and they are upstream of revenue in an obvious way. But there is a large gap between "you are more likely to reach someone if you call while they are still at their desk" and "responding in five minutes makes you 21 times more likely to win the deal." The second claim is the one that ends up in board decks. The first is the one the data supports.

It is also worth knowing who published it. InsideSales.com sold a web-form callback dialer, a product whose entire value proposition is calling leads within seconds of form submission. The document says so directly: "This study caused a significant shift in our corporate positioning. Our patent-pending web-form callback dialer telephony opens new frontiers in web-marketing, lead generation and sales." The same report closes by stating that customers "typically see a 2-4x increase in contact ratios and lead qualification rates using the InsideSales.com technology."

A vendor funding research that validates the vendor's product is not automatically bad research. Plenty of good science is industry-funded, and Oldroyd was a real academic doing real analysis. But you should hold the finding at the confidence level the design supports, not the confidence level the marketing implies. The honest summary is: in pooled data from six companies, calling fast dramatically improved your odds of reaching a person, and reaching a person is how sales happen.

There is one more part of that study nobody cites. Part 1 was a survey of 495 companies asking sales and marketing leaders when the best time to call back was. The result, in the authors' words: "we couldn't find ANY statistically significant answers to our question of WHEN." The survey found nothing. That is why they went and got the call data. The famous study exists because the obvious method failed first.

The Harvard numbers are not the numbers you were given

The other pillar is a 2011 Harvard Business Review piece, The Short Life of Online Sales Leads, by Oldroyd again, with Kristina McElheran of Harvard Business School and David Elkington of InsideSales.com. This is where the 42 hours figure comes from, and it is routinely conflated with the 2007 work.

The authors audited 2,241 U.S. companies by submitting a test lead to each and timing the reply. The distribution:

The average response time, among companies that responded within 30 days, was 42 hours. Read that qualifier again: among companies that responded within 30 days. The mean is computed on a truncated sample with the worst quarter of the distribution partly excluded. The real average, if you could include the 23% who never replied, is undefined. This is the first hint that the mean is the wrong statistic for this metric, a point worth holding onto for the measurement section below.

Now the finding that cuts against this post's own skepticism, which should be said plainly rather than buried. The famous 7x number does not come from that audit. The authors attach it to what they call a separate study, and its scale is the one thing in this category that is not small: 1.25 million sales leads received by 29 B2C and 13 B2B companies in the U.S. Set that against the six companies behind the five minute rule. This is real evidence, and it deserves more weight than the rest of this section might lead you to give it. Quoted exactly:

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.

Two things are load-bearing here. First, the 7x comparison is one hour versus two hours, not one hour versus "later." That is a much narrower and much more interesting claim than the version in circulation, and it says the curve is brutally steep at the front. Second, look at the definition they supply in their own parentheses: qualify means "having a meaningful conversation with a key decision maker."

So the dependent variable, again, is a conversation. Both landmark studies measure whether you got to talk to a human being. Neither measures whether you sold anything. Every speed-to-lead statistic you have ever been shown is, underneath, a measurement of conversation attainment via telephone.

That is the fact that determines whether any of this transfers to a DM.

The mechanism was presence, and presence is exactly what changed

The 2007 study did something unusual for a vendor white paper: it admitted it did not know why speed worked, then guessed anyway. The guesses are the most useful part of the whole document, because they describe a mechanism you can test against a new channel.

Their first explanation, quoted:

When a person submits a lead in a web form, you know where they are at that exact moment: they are at their computer desk, probably right near their phone. We call this "presence". If you call them immediately, they answer. If you wait, they move on to something else, often away from their phone.

This is the whole thing. The five minute rule is not a law about human attention or buying psychology. It is a law about physical co-location with a ringing telephone. A web form submission is a location ping. It tells you a specific human is sitting at a specific desk right now. The five minute window is how long that ping stays accurate.

Once you see it that way, the 100x contact multiplier stops being mysterious. Phone calls are synchronous. The connection either happens in real time or it does not happen at all. A missed call is not a delayed call, it is a null event. So the contact rate is governed almost entirely by whether the person is next to the phone, and the probability that they are still next to the phone decays fast. Thirty minutes is enough time to go to a meeting.

Their second explanation was interest decay: "Interest and need wane quickly. A few days later they often don't even remember they submitted a lead." That one does transfer to DMs. The third was the "Wow effect," the impression made when a callback lands almost immediately. Hold that thought, because it inverts on text, and the inversion is the most important thing in this post.

Here is the problem. On a DM channel, the presence mechanism does not exist. Not "is weaker." Does not exist.

A Telegram message does not require the recipient to be anywhere. It sits in an inbox. It generates a notification that persists on a lock screen. The person who messaged you at 02:14 was not sitting by a phone waiting for it to ring, and they will not be "away from their desk" at 09:40. They will be exactly as reachable at 09:40 as they were at 02:14, because reachability on an asynchronous channel is not a function of time.

The 100x number measures a variable that a DM channel does not have. There is no contact event to miss. Delivery is guaranteed and deferred. That single structural fact means the steepest part of the classic curve, the part that generates the most dramatic multiplier, simply does not apply to the channel most of your inbound now arrives on.

There is exactly one modern channel where the 2007 mechanism survives intact, and it is worth naming because it is the exception that proves the rule. A live chat widget is presence-gated in precisely the way a phone call was. The person is on your site right now. They will close the tab. If you do not answer while they are there, you have not sent a late reply, you have sent nothing, because there is often no identity to reply to. Live chat is the channel where five minutes is genuinely too slow, and it is the one place the original research transfers without modification.

Everywhere else, the tab does not close. That is the whole difference.

So what does decay on a DM channel, and how fast?

Something still decays. It is just not reachability, and being precise about what it is changes what you should do about it.

Three things decay on an asynchronous channel, and they run on different clocks:

Intent. The reason they messaged. This is the mechanism the 2007 authors correctly identified and the only one that transfers cleanly. Someone who messages at 2am about a product is in a state that will not exist at 2pm. This decays on a scale of hours to days, depending on how urgent the underlying need was.

Competitive displacement. They messaged five vendors, not one. This is the real driver behind the widely repeated claim that most buyers purchase from whoever answers first. Note that this decays on a scale set by your competitors' response times, not by any property of the buyer. If every vendor in your category answers in 12 hours, a 6 hour response is fast. If one of them runs an AI agent that answers in 90 seconds, your 6 hours is last place. Your speed target is relative, and nobody publishing a universal benchmark can know it.

Context. The conversation thread itself goes stale. At 02:14 they were looking at your pricing page with a specific question. By 09:40 they have to reconstruct their own mental state to engage with your answer. This is a real cost and it is invisible in every study, because phone-era research had no thread to go stale.

Notice none of these produce a five minute cliff. They produce a slope. On a DM channel, the difference between 90 seconds and 10 minutes is probably close to nothing, because the person is not going anywhere and 10 minutes does not meaningfully change their intent. The difference between 10 minutes and 14 hours is large. The difference between 14 hours and four days is probably decisive.

That is a fundamentally different management problem than "call within five minutes." It says: the hard deadline is not minutes, it is before their intent expires and before someone else answers. For most businesses, most of the time, that is a window measured in hours, not seconds. Which sounds like good news, and would be, except that the platforms went and invented a brand new cliff that the phone era never had.

Meta gives you exactly 24 hours, and it is not a guideline

This is the part email-era research could not have modeled, because it is not a behavioral finding. It is a business rule enforced in code by the platform your lead is messaging you on.

On Meta's Messenger and Instagram messaging APIs there is a standard messaging window. Per Meta's own platform policy, businesses have up to 24 hours to respond to a user, and messages sent inside that window may contain promotional content. Once the window closes, you cannot send a free-form message. You are restricted to a narrow set of message tags for specific approved purposes, and the workarounds that do exist, like one-time notifications and sponsored messages, are documented as Messenger-only and not available on the Instagram messaging API.

WhatsApp works the same way and is even more explicit. Meta's WhatsApp Cloud API documentation describes a customer service window: a 24-hour timer starts when a user messages or calls the business, and it resets to 24 hours if the user messages again before it expires. While the window is open you can send service messages freely. When it closes, in Meta's words, "you can only send pre-approved template messages."

Read that as a sales constraint rather than a technical one. On these channels, if you do not reply within 24 hours, you do not get to reply at all. Not "your reply is less effective." You lose the legal right to send the sentence you wanted to send, and you are downgraded to a pre-approved template that had to be submitted and reviewed before you knew what this conversation was about.

No such rule has ever applied to email. You can reply to an email from 2019. That is why every piece of speed-to-lead advice written for the email era treats response time as a soft optimization with diminishing returns. On Meta channels it is a step function with a wall at hour 24.

And the wall is not the same height everywhere:

This table is the actual 2026 answer to "does the five minute rule still hold." It does not hold, and it has been replaced by something both looser and harsher: you have far more than five minutes, and far less than forever, and the exact number depends on which app the message came from. If you are running one inbox across several channels, your response time policy cannot be one number. A 20 hour reply is fine on Telegram and a near-miss on Instagram.

This asymmetry has a strategic consequence people miss. The channels with no reply window are the ones where a slow human can still win, and the channels with a 24 hour wall are the ones where you either automate or accept structural losses. If most of your inbound is on Meta properties, the decision about overnight coverage has already been made for you by Meta. If most of it is on Telegram or email, you have room to be deliberate. Knowing your channel mix is therefore a prerequisite to setting any response target at all, which is what the omnichannel messaging benchmarks piece is for.

Meta also publishes your responsiveness back to your prospects. Facebook Pages can display a badge indicating the business answers messages quickly, computed from your response rate and response time, which turns your internal SLA into a public storefront signal. The platform is not neutral on this question. It has an opinion, and it shows that opinion to your buyers.

Why 24/7 human coverage does not survive contact with arithmetic

Every article that tells you to answer leads faster stops right before the part where you work out who does it at 3am on a Sunday. So let us do that part, because the numbers are not close.

A week contains 168 hours. A full-time employee is nominally 40 hours a week, but nobody delivers 40 coverage-hours for 52 weeks. Subtract annual leave, public holidays, sick days and training and a realistic figure is somewhere near 36 coverage-hours per week averaged across the year. Divide:

168 / 36 = 4.7

You need roughly five people to keep one chair occupied continuously. Not five people to handle your lead volume. Five people to make sure that at any random moment, one person exists. That is the floor before you have considered whether one person is enough during your busy hours, before redundancy, before anyone quits.

Now attach it to actual volume. Take a small team getting 40 inbound conversations a week, and assume 35% of them land outside your working hours, which is conservative if you sell to more than one continent. That is 14 conversations. Your working week is Monday to Friday, 9 to 6, which is 45 hours. The uncovered remainder is 123 hours.

To staff those 123 hours you need 123 / 36 = 3.4 additional full-time people. So the trade is: hire between three and four people, to answer fourteen messages.

Run the utilization. Fourteen conversations at a generous eight minutes of real handling each is 112 minutes of work. Spread across 123 hours of paid availability:

112 minutes / 7,380 minutes = 1.5%

Your night shift is idle 98.5% of the time. This is the actual reason small teams do not have 24/7 coverage, and it has nothing to do with discipline or caring enough about lead response time. On an asynchronous channel, cost scales with hours of availability while value scales with number of conversations, and those two quantities have come completely unglued from each other. The phone era hid this problem because inbound calls only arrive when someone is awake to dial. Messages do not have that courtesy.

There are only four honest responses to this arithmetic, and it is worth naming all of them rather than pretending the fourth is the only one:

• Accept the delay. Answer at 09:40, lose whatever you lose. For some businesses this is genuinely correct and we will get to which ones.
• Follow the sun. Hire in other timezones. Works, but it is a real org with real management overhead, and it is a solution available to companies of a certain size and not below it.
• Restrict the channel. Turn off DMs outside business hours, publish your hours, set expectations honestly. Underrated, and much better than silence.
• Make the marginal cost of availability approach zero. Which is the actual argument for an AI agent, and it is an argument about cost structure, not about intelligence.

That last point deserves emphasis because it is usually made badly. The case for automating the 2am reply is not "AI is as good as your best rep." It is that 98.5% idle is an impossible thing to pay a human for, and something has to occupy that shift or the shift stays empty.

What an AI agent realistically closes, and what it does not

Here is where most vendor content lies, so let us be specific about the mechanism instead.

AI Agents in CRM Solid read incoming DMs and reply in your voice across Telegram, X, email and the social inbox, using a persona and a knowledge base you define, with a rules engine, rate limits, human handoff, and per-contact pause. Thumbs up and thumbs down on a reply teaches the agent in place. If you want the setup mechanics rather than the argument, that is in the deploy AI agents guide.

What an agent genuinely does at 2am, in descending order of how confident you should be:

It keeps the conversation alive. This is the big one and it is nearly certain. Referring back to the platform windows above: a reply inside 24 hours preserves your right to have a free-form conversation on Instagram and WhatsApp at all. An agent that does nothing but answer inside the window has already prevented a category of loss that no amount of excellent human selling at 09:40 can recover.

It answers the answerable. A large share of inbound DMs are questions with correct answers that exist in your documentation. Do you integrate with X. Do you ship to Y. Is there a free plan. These do not need judgment, they need retrieval, and an agent with a decent knowledge base does them at least as well as a tired human, arguably better, because it does not skim.

It qualifies and routes. Asking what the person is trying to do, capturing it against the contact record, and putting them on the right board is mechanical work. Combined with lead scoring and pipeline routing, this means your 09:40 human opens a qualified conversation rather than a cold "hi." That is a real transfer of value even if the agent never persuades anyone of anything. The deeper piece on AI lead qualification covers where this goes wrong.

It denies your competitor the first-response slot. If displacement is the real decay mechanism on DM channels, and it probably is, then being present in the thread at all is most of the defense.

Now the other list, which matters more.

An agent does not close a considered purchase. If your product requires trust, a custom quote, a negotiation, or a decision by more than one person, the agent is not going to get there and you should not configure it to try. The failure mode is not that it fails to close. It is that it produces a plausible, confident, slightly wrong answer about something consequential, and now your 09:40 human starts the relationship by correcting their own company.

An agent does not know what it does not know. A knowledge base has edges. The most valuable thing you can configure is not a better persona, it is a sharper handoff trigger. An agent that says "that is a good question and I want to get you the exact answer, someone will confirm this morning" is worth more than one that guesses. Handoff is not the agent failing. Handoff is the agent working.

An agent does not fix a bad offer or a dead lead. Speed is a multiplier on something. If the something is zero, faster produces zero sooner.

The honest frame is this: the agent's job at 2am is not to close the deal. It is to make sure a live, qualified, correctly-routed conversation still exists at 09:40, on a channel that has not locked you out. That is a modest claim. It is also, given the arithmetic above, worth several full-time salaries you were never going to spend. If you want the category distinctions between an autoresponder, a rule-based bot, an LLM chatbot and an actual agent, the AI agents vs chatbots breakdown is the sibling piece to this one.

The moment you automate, your response time metric starts lying

This is the part that will actually hurt you, and almost nobody writes it down.

The instant you put any automation on a channel, time-to-first-response becomes worthless. It goes to four seconds and it stays there forever, no matter how badly you are serving people. You have built a metric that structurally cannot report failure. Every dashboard turns green and every dashboard is lying.

Worse, this is the exact metric most teams report to leadership, because it is the one that is easy to compute and the one the 2007 study appears to endorse. So you get the following pathology: the team ships an autoresponder, average response time drops from 14 hours to 4 seconds, the number goes in the QBR deck, everyone is congratulated, and conversion does not move at all, because nothing about the customer's experience changed. They still waited until 09:40 to get an answer. They just got a receipt first.

You need at least four separate clocks. Here is a single 2am conversation measured properly:

Six numbers, and they tell six different stories. The 4 seconds is noise. The 52 seconds is the number that plausibly maps to the classic research, because it is the first moment the customer received actual information. The 7h 26m is the number your competitor is beating you on if they have humans in that timezone. And the last row is the one that decides whether you had a business at all.

Some rules that follow from this:

Report time to first meaningful response, not time to first response. Define meaningful as: contains information specific to what the person asked. A greeting is not a response. "Thanks, someone will be with you shortly" is not a response, it is a hold message with good manners. If your tooling cannot distinguish these, that is a tooling problem, not a definitional one.

Always report first-human alongside it. Not because human is better, but because the gap between the two is the single most diagnostic number you have. A 52 second AI response and a 7 hour human response is a healthy pattern. A 52 second AI response and a never human response means your agent is quietly absorbing conversations that needed escalation, and your handoff triggers are wrong.

Use the median and the 90th percentile. Never the mean. Response time distributions are viciously long-tailed. One lead answered after nine days moves your mean and tells you nothing about typical experience. Recall that even the HBR authors had to write "among companies that responded within 30 days" to make their average computable, and that they had a 23% never-responded group sitting outside it. If the researchers had to truncate the distribution to get a mean, the mean is the wrong statistic. Your p90 is where your reputation lives.

Measure the no-response rate as its own number. The most important finding in the HBR audit was not 42 hours. It was that 23% of companies never replied at all. That is not a slow response, it is a different failure, and averaging it into a response time metric erases it. Count it separately or you will never see it.

Measure on the lead's clock, not yours. A 14 hour response looks catastrophic until you notice every one of those hours was overnight, and then it looks like the cost of not employing five people. Segment by whether the message landed inside or outside your working hours before you draw any conclusion, because those are two different operational problems with two different solutions.

Once these are separated you can put them somewhere they get looked at. Cross-module reporting is where response distributions stop being a spreadsheet exercise, and hot-visitor alerts are the other half of the same problem: knowing someone is on your pricing page right now is a presence signal, which is the closest thing a website gives you to the 2007 study's original mechanism.

The counter-argument: instant replies can cost you the deal

Everything above argues for speed. Now the case against, because it is real, it is evidenced, and it is the reason "reply in 4 seconds" is bad advice on some products.

Start with a finding that has nothing to do with sales. In The Labor Illusion: How Operational Transparency Increases Perceived Value (Buell and Norton, Management Science, 2011), the authors ran five experiments on simulated travel and dating sites. Participants chose between a service that returned results instantly and one that made them wait, with identical results.

When the waiting service showed its work, displaying which airlines it was searching rather than a blank progress bar, 62% of participants preferred waiting 30 seconds over instant results, and 63% preferred waiting a full 60 seconds. When the wait was shown without that transparency, preference for waiting collapsed to 42% at 30 seconds and 23% at 60 seconds.

Read the first pair of numbers again. Given identical output, most people chose to wait a minute rather than be served instantly, provided they could see effort being expended. The instant service was the less valuable one. The authors' explanation is reciprocity: perceived effort by the provider triggers a felt obligation, and that mediates the increase in valuation.

Now apply it, and the 2007 report hands us the perfect case study. It describes a "Wow effect" produced by InsideSales.com's own callback technology, which dialled leads in under three seconds. Prospects reacted with "wow, that was fast! You are impressive," and reported feeling that the rep "must be really on top of things."

Look carefully at what generated that reaction, because it is not what it appears to be. The dialing was automated and effectively instant, so the speed itself was cheap. But what arrived on the other end of that call was a human being, available, immediately, for you. The speed was free. The thing the speed delivered was expensive, and the prospect correctly inferred the expensive part.

A DM reply in 0.8 seconds delivers no human. It is proof that no person read the message, because no person can read and answer in 0.8 seconds. So the same signal inverts: in 2007, near-instant response proved a human was standing by for you, and in 2026, near-instant response proves that one is not. Identical variable, opposite meaning, because what is expensive changed.

There is a beautiful confirmation of this in unrelated research. Fast response times signal social connection in conversation (Templeton et al., PNAS, 2022) found that in live conversation, response gaps under about 250 milliseconds are an honest signal of connection precisely because they are too fast to be consciously controlled. You cannot fake them, so they mean something.

Hold those two side by side. In spoken conversation, too-fast-to-fake proves sincerity. In text, too-fast-to-type proves automation. Speed is an honest signal in both cases. It is just honestly signalling different things, and the sign flips depending on whether producing the speed is hard. This is the single most important thing to understand about response time in 2026, and no study from the phone era could have told you, because in the phone era speed was always expensive.

Buell and Norton also found the boundary, and it is the sharpest warning in the paper. In their fifth experiment they varied whether the outcome was good or bad. Transparency about effort increased value for favourable and average outcomes, but participants valued the transparent service less than the instant one when the outcome was unfavourable. Visible effort that produces a bad answer is worse than no visible effort at all.

Translated into your inbox: an agent that spends 40 seconds "thinking," announces that it has checked your knowledge base, and then returns an answer that does not help is strictly worse than a blunt instant "I will get someone to answer this properly in the morning." If you are not confident in the answer, do not dress up the delivery. Take the handoff.

Where speed to lead is cargo cult

Some businesses should stop optimizing this metric entirely. Naming them is more useful than another paragraph about urgency.

Long-cycle, high-consideration, committee purchases. If your deal takes four months and involves a procurement review, the marginal value of answering in 90 seconds instead of four hours is approximately nothing. The buyer is not making a decision today. They are assembling a shortlist over weeks. Both the 2007 and 2011 studies drew heavily on categories like insurance, lending, automotive and education, where a web form means a person actively shopping right now with intent to transact soon. That is not your market. Applying their multipliers to a six figure annual contract is a category error, and the studies never claimed otherwise.

Products where instant availability reads as desperation. This is the labor illusion running in reverse at the level of the firm rather than the interaction. If you sell a scarce, premium or expert service, a reply at 2am on a Sunday can carry an unintended message: that you had nothing better to do. Consultancies and specialist agencies discover this repeatedly. There are markets where a considered reply on Monday morning outperforms an eager reply on Saturday night, and the mechanism is not mysterious.

Research-mode contacts. Not everyone who messages you is a lead. A student, a competitor doing diligence, and a person comparing options for a purchase in Q4 all look identical in the inbox at 02:14. Speed spent on them is a real cost with no return, which is why qualification before escalation matters more than raw response time.

Anywhere your answer quality is variable. Covered above, but it generalizes: if the fast answer has a meaningful chance of being wrong, and the topic is consequential, slow and correct beats fast and confident. Speed is only free when accuracy is not the binding constraint.

The steel-manned version of the whole speed argument is narrower than the popular version, and it is this: response time matters enormously when the buyer has an active, transactional, comparison-shopping intent and low switching cost between vendors. It matters much less otherwise. The five minute rule was measured in exactly the first case and gets applied indiscriminately to the second.

A response time policy you can actually run

Concrete version. Adjust the thresholds, keep the structure.

Tier the response, do not uniform it. Three distinct events, deliberately separated in time:

Do not send a bare acknowledgement. There is no tier for "we got your message." It is the metric-poisoning move from the section above and it consumes the buyer's attention without giving them anything. Either answer the question or say something true about when a human will.

Add deliberate latency to the agent. If your agent can reply in 800 milliseconds, make it wait. Somewhere between 20 and 60 seconds is a defensible band: fast enough that intent has not decayed, slow enough that the reply does not announce itself as machinery before the first word is read. This feels wrong to engineers and is correct.

Write the 2am message so it does not overclaim. Here is copy that works, for an inbound Instagram DM asking whether you support a particular workflow:

Yes, that works. You would set it up under Pipelines, and the routing part is automatic once the channel is connected. Two things I would want to check before promising it fits your case: how many accounts you are running, and whether you need the handoff to a specific person or just to a team. It is 2:15am here so I am the AI agent covering nights. I have flagged this for the team and someone who has actually built this setup will pick it up first thing. If you want to leave the two answers here, they will have them before they reply.

Look at what that does. It answers the question. It names the limits of its own answer. It discloses what it is without apologising for it. It sets a real expectation with a real time attached. And it asks for the two pieces of information that make the human's 09:40 reply better, which converts dead overnight hours into pipeline work. It does not say "our team will reach out shortly," because shortly is not a time.

Disclose the agent. Partly for regulatory reasons, which are real and getting stricter. Mostly because the alternative is worse for you. An undisclosed agent that gets caught converts a speed advantage into a trust deficit, and the catch rate is high, because people ask bots whether they are bots.

Set the handoff triggers before the persona. Pricing negotiation, anything about contracts or legal terms, a second consecutive question the knowledge base cannot answer, any detectable frustration, any explicit request for a human. All of these should stop the agent and page a person. On CRM Solid this is what per-contact pause is for: the moment a human takes over, the agent stops touching that thread rather than talking over its own colleague.

Set rate limits and honour them. An agent that fires three messages in a row because the prospect sent three lines is not being responsive, it is being a problem. Coalescing consecutive messages into one considered reply is the correct behaviour.

Instrument the gap, not the speed. Review the first-human minus first-meaningful delta weekly, by segment. That is where the operational truth is.

Most speed-to-lead statistics are laundered, including ones in this article's competitors

A note on epistemics, because researching this piece was instructive in a depressing way.

The single most-quoted claim in this category is that roughly 78% of customers buy from the company that responds first. It is attributed almost universally to a company called Lead Connect. We tried to find the original. Every citation leads to another blog post citing another blog post, and the trail terminates without ever reaching a study, a methodology, a sample size, or a date. We could not verify it, so it does not appear as a fact in this article. Treat it the same way.

The 2026 vintage is worse. Searching for current benchmarks returns a wall of pages offering extremely precise figures: a study of 253,817 inbound leads across 1,247 companies, a benchmark drawn from 939 B2B companies, 47 data points on lead response. The precision is the tell. These pages are frequently on domains with no research operation, no named authors, no methodology section, and no way for anyone to check anything. Numbers with four significant figures and no author are decoration, not evidence.

Meanwhile the research that is real, and that everyone is nominally citing, dates from 2007 and 2011. The 2007 study measured phone dials at six companies and told you in writing that it did not look at close rates. The 2011 piece audited whether a test lead got a reply, and took its qualification finding from a separate, much larger pool of 1.25 million leads. That larger pool is the strongest evidence anyone in this category has, and it still only measured whether somebody got a conversation. Both are nearly two decades old. Neither saw a single Instagram DM, because Instagram did not exist when the first was published.

So the state of the evidence is: the foundational work is old, narrow, honest about its limits, and about a channel most of your leads no longer use. The modern work is mostly invented. That is not a comfortable thing for a company that sells software in this category to write, but you should know it before you set a target off someone's infographic.

The practical rule: if you cannot open the primary source and read the methodology, do not put the number in a plan. Three verifiable statistics beat fifteen laundered ones. Every external number in this post links to the document it came from, and you should hold anyone writing about response time to that standard, including us.

Frequently asked questions

What is a good lead response time in 2026?

It depends on the channel, which is the honest answer nobody gives. On Instagram, Messenger and WhatsApp, treat Meta's 24 hour window as a hard deadline, because after it you cannot send a free-form reply at all. On Telegram and email there is no platform limit, so your target is set by competitors and intent decay: hours, not seconds. Inside working hours, under 30 minutes is a defensible human target.

Is the five minute rule still true?

It was true for outbound phone dials to web form leads, which is what the 2007 InsideSales and MIT study measured. Its mechanism was presence: the lead was sitting near a phone and would soon walk away. DM channels have no presence requirement, so the sharpest part of that curve does not transfer. The underlying point, that intent decays quickly, still holds.

Should I measure first response time or first human response time?

Both, plus the gap between them. Time to first response becomes meaningless the moment you automate anything, since it pins at a few seconds forever. Measure time to first meaningful response, meaning the first message containing information specific to the question asked, and report first human response next to it. Use the median and the 90th percentile, never the mean.

Can an AI agent replace 24/7 human sales coverage?

It replaces the coverage, not the selling. Staffing 168 hours a week takes roughly five people to keep one chair filled, and a night shift handling fourteen conversations runs at about 1.5% utilization, which no small team can justify. An agent removes that cost. It should qualify, answer documented questions, and hand off. It will not close a considered purchase.

Can replying too quickly hurt conversion?

Yes, on high-consideration purchases. Buell and Norton's labor illusion research found people preferred waiting 30 to 60 seconds over instant identical results when they could see effort being spent, by roughly 62% to 63%. A sub-second reply proves no human read the message. Add 20 to 60 seconds of deliberate latency and never dress up a low-confidence answer as hard work.

Why did my response time improve but conversion stay flat?

Almost always because you shipped an acknowledgement rather than an answer. Dropping from 14 hours to 4 seconds changes your dashboard and nothing about the customer's experience, since they still wait until morning for real information. Check whether your first message contains anything specific to what was asked. If not, you improved a metric, not a business.

Where to start

Pick one channel and one number. Find the median and the 90th percentile of your time to first meaningful response on the channel that brings you the most inbound, split by whether the message arrived inside or outside your working hours. Almost nobody has this number, and the out-of-hours half of it usually settles the argument about what to automate on its own.

If that split shows what it usually shows, AI Agents is the part of CRM Solid built for the overnight half, and there is a free plan to test it on real traffic before you commit to anything. Point it at one channel, set the handoff triggers hard, and watch the gap between first meaningful and first human. That gap is the whole game.