Follow-up timing advice in cold email usually arrives as a day count with no denominator behind it. This research post reviews four widely cited follow-up datasets, shows what each one actually measures, and sets out a matched-cohort study design a team can run on its own list to see whether the gap between touches changes qualified replies.

These points apply to B2B cold email sequences sent from a cold email platform, with replies tracked into a CRM and contacts split into cohorts before launch. The study design gives a team a fair answer for its own audience and offer. It does not produce a universal best delay, and small programmes may not reach the sample a reliable answer needs.

What The Published Data Says About Follow-Up Timing

Instantly's report on campaigns from January 1 to December 18, 2025, finds that 58% of replies come from the first step, yet it publishes no positive or qualified reply rate, and neither do the other three datasets reviewed here.

That is the headline finding of this evidence review. Teams running cold email platforms can read where replies land and how fast they arrive. They cannot read, from any of these sources, whether a longer or shorter gap produces more replies that turn into meetings or pipeline.

58%

Share of all cold email replies that came from the first step of a sequence in Instantly's benchmark, with follow-ups contributing the other 42%.

2025 · Campaigns across thousands of active Instantly workspaces · Source

Four widely cited datasets carry much of the published evidence on follow-ups. Each measures something different.

Four widely cited follow-up datasets

Four widely cited follow-up datasets
DatasetSample and periodWhat it measuresQualified replies
Instantly benchmarkBillions of interactions, Jan 1 to Dec 18, 2025Reply share by stepNot published
Smartlead timing benchmark850M+ emails, undatedReply arrival timeNot published
Belkins follow-up study7.5M emails, 2025Replies and meetings by stepMeetings only
BusySeed reply study1,857 replies, Dec 2025 to Aug 2026Reply share and sentiment by stepInterested labels only

Two of the four come from vendors that sell cold email software and publish their own platform data. Instantly aggregates thousands of active workspaces, and Smartlead counts senders with at least 500 emails. Belkins, an outbound agency, draws on its own client campaigns from 2025 and gives no exact start and end dates. BusySeed, an independent agency, publishes the narrowest sample but the most detail, including its own limitations.

None of them publishes a denominator for positive or qualified replies. Instantly's response metric counts every reply. Smartlead does not say whether automatic replies such as out-of-office messages are included. Belkins reports meetings by step, which is the closest any dataset gets to a qualified outcome, but it reports no days-between-follow-ups figures.

What the data does cover:

  • Reply share by step: Instantly, Belkins and BusySeed.
  • Reply arrival time: Smartlead only.
  • Meetings by step: Belkins only.
  • Reply sentiment by step: BusySeed only.

What none of the four covers is timing against qualified replies. No result separates gap length from step number either, because in every sequence the third email is also the one sent latest. That is the gap the study design later in this post is built to close.

Which Follow-Up Gets The Most Replies?

In BusySeed's study of 1,857 cold email replies collected between December 1, 2025 and August 14, 2026, 61.7% attached to the first email, 20.9% to the first follow-up and 4.9% to the third follow-up or later. The second follow-up took 12.5%, so close to four in ten replies still came after the opening email.

Instantly's benchmark points the same way, with 58% of replies from step one and 42% from follow-ups. In both Instantly's and BusySeed's data, the first email earns the largest single share of replies.

Belkins seems to disagree. Across 7,530,489 sent emails, steps two to six together produced 58.6% of all replies. Both findings can be true at once, for two reasons:

  • Per-step versus cumulative share: Belkins found the first email had the highest per-step reply rate, at 0.59%, falling to 0.28% by step six. Five follow-ups that each earn less than the opener can still add up to more than half of all replies.
  • Sequence length: Belkins' sequences ran to six steps, while Smartlead reports a median sequence of three emails. More steps give follow-ups more chances to collect replies.

The spread inside a single dataset is wide too. Across the 14 BusySeed groups with enough data, the first email's share ranged from 46.6% to 89.5%, with a median of 69.4%. One audience and offer can behave very differently from another.

What Step Share Cannot Tell You

Step share shows where a reply attaches in a thread. It does not show the causal effect of sending a follow-up, or of the gap before it. BusySeed states this limitation itself, and adds that it did not test sequences without follow-ups.

The sentiment split is the most interesting signal, with a caution attached. Among interested replies (123 of them), 52.8% attached to the first email, lower than the 60.4% for neutral replies and far below the 87.1% for unsubscribe or hostile replies. Follow-ups may carry a larger share of interested replies than of replies overall. With only 123 interested replies in the sample, that is a hypothesis to test, not a finding to plan around.

How Long Should You Wait Between Follow-Ups?

Smartlead's timing benchmark across more than 850 million emails shows 88.5% of cold email replies arrive within 24 hours and 93.9% within 72 hours, so a gap of three days already clears the window in which nearly all replies to the previous email land. Beyond that window, none of the four datasets shows whether waiting longer helps.

The arrival curve is steep. In the same benchmark, 64.0% of replies arrive within the first hour. A follow-up sent after two or three days rarely overtakes a reply that is still on its way.

Smartlead is explicit about the limit of its own data. It says the benchmark does not show whether a longer gap improves the next email's performance, and it counts hours rather than business days. It does report that the second email earns a 0.73% reply rate, 62% of the first email's rate, but that figure says nothing about the gap before it.

Common advice fills the space with day counts:

  • Instantly's benchmark recommends spacing touches three to four days apart, across four to seven touchpoints.
  • An Instantly blog post states that waiting three days before the first follow-up raises reply rates by 31% versus next-day follow-ups. It attributes the figure only to "industry research" and names no study, so it is an example of an unsourced claim rather than evidence.
  • Woodpecker's defaults set the first follow-up at three days, then six, 12 and 18 days. That is what a vendor ships, not a measured optimum.

None of these numbers is tied to qualified replies. They are conventions, not findings.

What Counts As A Qualified Reply?

A qualified reply comes from a contact who fits the ideal customer profile and, within a fixed window, books a meeting or reaches a sales-qualified stage in the CRM, and the step that wins on meetings can differ from the step that wins on replies, as Belkins found in 2025 that the third step drove 35.6% of email-sourced meetings. Raw reply counts blur that signal with auto-replies and refusals.

Belkins also found that the appointment rate levels off at 0.2% from step three onward. A timing test that only counts replies could pick the arm that wins on volume and still lose on meetings.

The outcome sits at the top of a measurement ladder. Each rung removes noise from the one below:

  1. All replies: Everything that lands in the inbox, including auto-replies and unsubscribes.
  2. Human replies: All replies minus out-of-office messages, bounces and other automatic responses.
  3. Interested replies: Replies tagged positive. Woodpecker uses Interested, Maybe later and Not interested, set by AI or by hand, and lemlist offers AI reply intent detection.
  4. Meetings booked: Replies that become a calendar event, the outcome Belkins tracks by step.
  5. Sales-qualified: In HubSpot, a Sales Qualified Lead is "a contact or company that your sales team has qualified as a potential customer".

Interest tags are useful, but they are not the outcome. They are set by a classifier or a rep, and the count depends on the view. Woodpecker's Inbox counts individual messages, so three Interested replies from one prospect count three times, while its Campaigns view counts that prospect once. For a timing test, use the prospect count.

Reply volume alone is a weak proxy, which is why high cold email reply rates can hide poor pipeline. The study design below uses qualified replies per contact enrolled, as defined above, as its primary outcome and keeps all replies as a secondary measure.

Why The Same Follow-Up Delay Means Different Things In Each Tool

A follow-up set to wait three days can go out on different days in different tools, because Woodpecker and Reply.io count 24-hour days including weekends by default, lemlist counts only scheduled sending days and HubSpot sequences count business days.

That difference breaks any comparison of timing data across platforms. It also breaks a test inside one platform if the arms use different settings. The table shows how each tool handles the parts of a sequence that change elapsed time or the reply count.

How each tool counts delays, handles replies and bills

How each tool counts delays, handles replies and bills
ToolDelay ruleReply handlingBill driver
Woodpecker24-hour steps, weekends countInterest levels, AI or manualContacted prospects and add-ons
lemlistScheduled sending days onlyAI reply intent detectionEmail allowance, users, data add-ons
Reply.io24-hour days, weekends countUnified inbox on Reply UnlimitedUsers, mailboxes and credits
QuickMailOut-of-office adds 14 daysInbox scan every 10 minutesSends, contacts and workspaces
HubSpotBusiness days by defaultUnenrols on reply or meetingPaid Sales or Service seat

How Each Tool Counts A Delay

The same "3 days" setting produces different send times:

  • Woodpecker counts in 24-hour increments from the previous send, with weekends included by default. A working-days-only option keeps the original send time, so a two-day follow-up after a Friday 10 a.m. email goes out on Tuesday at 10 a.m. Woodpecker checks the delay first, then whether sending is allowed that day.
  • lemlist measures delays in scheduled sending days. On a Monday-to-Friday schedule, a one-day delay from Friday sends on Monday.
  • Reply.io treats a three-day delay as 72 hours from the last email of the previous step, weekends included, and sends only if that time falls inside the sequence schedule.
  • HubSpot sequences set delays in business days, up to 90, with "Execute steps on business days only" switched on by default.
  • Instantly, outside this post's tool set, counts all calendar days, including weekends and non-sending days, and holds a step that lands on a non-sending day until the next active window.

How Out-Of-Office Handling Stretches A Gap

Out-of-office handling can add two weeks to a gap without anyone touching the delay. QuickMail marks a lead as out-of-office when it detects an auto-reply and adds 14 more days to the wait step by default, so a three-day wait becomes 17 days. The setting is editable, and on rare occasions QuickMail reads an out-of-office message as a real reply. If one arm happens to hit more holiday auto-replies, its average gap grows and the comparison is no longer clean. Woodpecker filters out-of-office replies from the inbox automatically, and lemlist lists AI out-of-office detection.

What Drives The Bill

Each tool bills on a different meter. Woodpecker bills by contacted prospects, with add-ons on top. QuickMail caps plans by emails sent and uploaded contacts and charges for extra workspaces. lemlist bills by email allowance, users and data add-ons, and Reply.io by users, mailboxes and credits. HubSpot sequences need a paid Sales or Service seat. For a wider view of the options, see the best cold email platforms compared on sending controls and cost.

The decision rule is simple. Convert every arm's delay into elapsed hours, use identical schedule and out-of-office settings in each arm, and record the actual send timestamp for every contact.

How To Test Follow-Up Timing With Matched Cohorts

A fair follow-up timing test splits one contact list into randomised, matched cohorts before launch, changes only the gap before the first follow-up, and holds copy, sender, send window and step count constant in every arm.

What follows is the method statement for that study. A team can run it in any of the tools above and report it the same way.

Inclusion: One ideal customer profile, one offer and one sequence. Mixing segments makes any difference between arms impossible to attribute.

Arms: Two versions of the same sequence that differ only in the gap before the first follow-up, defined in elapsed hours. A short arm and a long arm is the simplest design. Every later gap stays the same in both arms.

Constants: Copy, subject lines, sender mailboxes, send windows, step count, schedule settings and out-of-office settings are identical. In QuickMail, set the same out-of-office delay in both arms. In Woodpecker, set the working-days option the same way in both.

Assignment: Randomise contacts to arms before launch, stratified by company size and email domain, so the cohorts match. Do not rely on a platform's built-in A/B feature for this. Reply.io's documentation says contacts who received variant A in step one may get variant B in step two, and it does not list delay as a variable. lemlist's A/B guidance compares content on reply rates and does not say whether delays can differ between variants. HubSpot's sequence documentation describes a step performance report but no A/B testing. Run each arm as its own campaign instead.

Pre-registration: Write down, before the first send, the primary outcome (qualified replies per contact enrolled), the secondary outcomes (all replies, meetings booked, opportunities), the attribution window and the sample size per arm.

Denominators: Record per arm the contacts enrolled, delivered, bounced, out-of-office, unsubscribed and replied. Report every outcome against contacts enrolled, so a bounce-heavy arm cannot look better by shrinking its base.

Analysis: Compare the primary outcome between arms once the attribution window has closed for every contact, using the test chosen in advance. Report the difference with its confidence interval, not a single winner.

The mechanics of setting up each campaign, from audience selection to authentication and list verification, follow the same steps as a first B2B cold email campaign. The only change is that the list is split before either campaign starts.

  • Choose one ICP, one offer and one sequence
  • Define both arms' first gap in elapsed hours
  • Hold copy, subject lines, mailboxes, send window and step count constant
  • Match schedule, weekend and out-of-office settings across arms
  • Randomise contacts to arms before launch, stratified by company size and domain
  • Run each arm as its own campaign, not a built-in A/B split
  • Pre-register the primary outcome, secondary outcomes, attribution window and sample size
  • Record enrolled, delivered, bounced, out-of-office, unsubscribed and replied per arm
  • Log the actual send timestamp for every email
  • Analyse only after every contact's attribution window closes

How Many Contacts A Follow-Up Timing Test Needs

Checking a follow-up test's dashboard and stopping when one arm looks ahead inflates false positives, because, as Evan Miller shows, repeated significance testing always increases the rate of false positives, so the cohort size has to be fixed before the first email goes out.

Miller's recommendation is direct. Decide on a sample size in advance and wait until the experiment is over. "The more you peek, the more your significance levels will be off."

Qualified replies make this harder than a copy test. Instantly's benchmark puts the average cold email reply rate at 3.43%, with the top quartile at 5.5% or more. Qualified replies are a fraction of that. A cohort sized to detect a difference in all replies is too small to detect a difference in qualified ones.

Vendor guidance sets a lower bar. lemlist's A/B advice is to wait until at least 100 to 200 leads have received each version before choosing a winner, then keep monitoring over 500 or more leads. It also warns that small differences, such as 8% versus 9% reply rates, might be random. Treat that as a floor for reply-rate copy tests, not a target for qualified-reply timing tests.

To size each arm:

  1. Pull the last quarter's qualified-reply rate per contact enrolled from the CRM.
  2. Decide the smallest difference between arms worth acting on.
  3. Use a standard sample-size calculation for two proportions to get contacts per arm.
  4. If the programme cannot reach that size in one send, run the test across several sends with the same settings and pool them.

How To Measure Cold Email Revenue From Follow-Ups

Revenue from a follow-up timing test is measured per contact enrolled, by following each contact's Date entered Sales Qualified Lead and Opportunity properties in HubSpot within a fixed attribution window, because a reply count stops at the inbox.

The primary outcome is the same qualified replies per contact enrolled used in the study design, so a reply counts toward it when the contact books a meeting or enters the Sales Qualified Lead stage within the window.

The sending tool and the CRM do different jobs here. The difference between a cold email platform and a sales engagement platform matters less than the hand-off, where every reply must reach the CRM tagged with the step and arm that earned it.

Tracking a timing test from enrolment to revenue in HubSpot

Tracking a timing test from enrolment to revenue in HubSpot
StageWhat to recordWhy it matters
EnrolledArm and enrolment dateDenominator for every outcome
RepliedStep, arm and reply dateSecondary outcome
Meeting bookedMeeting dateCounts toward primary outcome
Sales Qualified LeadDate entered SQLCounts toward primary outcome
OpportunityDate entered OpportunityPipeline per contact
Closed dealDeal value and close dateRevenue per contact

Three HubSpot behaviours make this work:

  • Stages only move forward automatically. Automatic updates never move a lifecycle stage backward.
  • Date entered stays set. Each stage's Date entered and Date exited properties are not cleared when a record moves backward, so a contact that reached Opportunity and slipped still counts.
  • Sequences unenroll on reply. By default, HubSpot unenrolls a contact who replies to any email in the sequence or books a meeting, which stops later steps and keeps the arms clean.

Some of this sits behind paid tiers. Sequences need Sales Hub or Service Hub Professional or Enterprise and an assigned Sales or Service seat. The Latest time in and Cumulative time in stage properties also need a Professional or Enterprise subscription. A team that runs its follow-ups from HubSpot workflows instead can use business-day delays, which skip weekends and release records at the same clock time they entered.

Report opportunities and closed revenue per contact enrolled, within the same attribution window for both arms. Revenue per reply rewards the arm with fewer, warmer replies and hides the contacts the other arm reached.

Limitations Of This Study Design

A matched-cohort timing test answers the question for one audience, one offer and one sending setup, and because changing the first gap also moves every later send, it measures the whole schedule rather than a single delay.

The design has known limits:

  • One segment: Results do not transfer to other audiences or offers. BusySeed's first-email share ranged from 46.6% to 89.5% between groups, which shows how much segments differ.
  • Linked step and time: Step number and elapsed time move together in any sequence. Holding later gaps constant keeps the arms comparable, but the long arm still finishes later.
  • Drift: Seasonality and mailbox health change during a long test, and holiday auto-replies can stretch gaps unevenly when out-of-office rules add days.
  • Classification: Interest labels set by AI or by hand need an audit sample. BusySeed took its labels as recorded without audit, and Woodpecker's counts change between message and prospect views.
  • Misattribution: Thread errors or skipped follow-up sends can inflate the first email's share, a limitation BusySeed states for its own data.
  • Slow outcomes: Qualified replies and opportunities take weeks to mature in the CRM, so results arrive well after the last send.

Frequently Asked Questions

Do out-of-office replies count as replies in follow-up data?

It depends on the dataset and the tool. Smartlead's benchmark does not say whether automatic replies are included. QuickMail by default marks the lead as out-of-office and adds 14 days to the next wait step, and it occasionally reads an auto-reply as a real reply. Exclude auto-replies from any qualified-reply count.

Should cold email follow-ups go out on weekends?

That depends on how your tool counts the delay and on your send window. Woodpecker and Reply.io count weekends toward a delay by default, lemlist counts only scheduled sending days and HubSpot uses business days. The send window decides the actual send day, so set it the same way in every arm.

Can I trust a vendor's cold email reply-rate benchmark?

Use it for what it measures. Vendor benchmarks are useful for reply arrival time and the share of replies by step, but they aggregate data across a vendor's users and publish no denominator for qualified replies. Instantly's response metric counts every reply, so treat any day-count recommendation as convention, not evidence.

Sources

Sources