Is cold email still worth it for agencies in 2026?

Three months in, the agency has sent 18,000 emails, collected 61 replies and booked four meetings, two of which no-showed. The founder does the division, gets 0.34 percent, reads a blog post saying the average is 3.43 percent, and concludes the channel is finished. That conclusion is wrong, and so is the comparison that produced it. The published average does not say what it divides by or whether an out-of-office counts, and the agencies that publish their formula land between 0.45 and 4.5 percent for campaigns that all work. Put the two numbers on one definition and the gap shrinks to something a list change could close. Cold email in 2026 is a measurement problem wearing the costume of a market problem, and the founders who quit are mostly quitting over arithmetic.

Why does everyone say cold email reply rates collapsed?

The decline everyone quotes compares a 2019 study of link building outreach with a 2025 study of B2B sales email. The 8.5 percent baseline comes from Backlinko's analysis of 12 million outreach emails, published in April 2019 with data supplied by Pitchbox. Read the study and the categories are listed plainly: link building, guest posting, sponsorships, infographics, resources, reviews, mentions, roundups. Sales prospecting is not among them and is never mentioned.

So the sentence that has been copied across hundreds of pages, that reply rates fell from 8.5 percent to 3.43 percent, puts a bloggers-asking-bloggers-for-backlinks number next to a cold-sales-email number and calls the difference a collapse. They were never the same metric. One of them was people in the same industry emailing each other about content, which is roughly warm outreach with extra steps.

This matters beyond pedantry. If you believe the channel lost 60 percent of its effectiveness, the rational move is to leave it. If you believe nobody actually measured that, the rational move is to find out what your own number is and what it is divided by. Those lead to opposite decisions, and the second one is supportable.

The Backlinko study does hold up on the part people ignore: one extra follow-up lifted replies by 65.8 percent, and sending three or more messages produced the best overall rate. That finding is seven years old and still the cheapest available improvement to most agency campaigns.

What do the 2026 cold email benchmarks actually say?

Four publishers with first-party platform data put the average somewhere between 0.45 percent and 5.1 percent, and all four are reporting honestly. The numbers disagree because the measurements do.

Source	Average reply rate	Sample	What it divides by	Auto-replies
Belkins, 2026 study	0.45%	7,530,489 emails, 2025	Emails sent	Excluded
Instantly, 2026 report	3.43%	Platform-wide, Jan to Dec 2025	Not stated	Not stated
Saleshandy, Oct 2026	4.5%	30.31M emails, 40,006 sequences	Emails delivered	Counted
Overloop, Jun 2026	3.4 to 5.1%	Compiled across named studies	Varies by study	Varies
Backlinko, 2019	8.5%	12M emails, link building	Emails sent	Not stated

lemlist put a number on the disagreement: the spread between published sources reaches 7.6x. A 7.6x spread is not a benchmark. It is a warning that the word "reply" has at least four meanings in this market, and that any page telling you the average without telling you the denominator is selling you something.

Belkins is the most useful of the four precisely because it is the lowest. They switched their own methodology mid-stream and said why. Previous years divided replies by unique recipients who opened an email; the 2025 study divides by total emails sent, which they call "a stricter and, frankly, more honest denominator". Then they state the consequence outright: a 5 percent rate against openers and a 0.45 percent rate against total sends can describe the same campaign.

Read that twice before you judge your own campaign. The same emails, the same replies, the same booked meetings, reported two ways, produce numbers that differ by a factor of eleven.

What is a reply rate actually measuring?

A reply rate is a fraction, and the whole argument about cold email lives in its denominator. Three choices sit underneath every published figure, and each one moves the result more than any subject line ever will.

The first choice is the denominator itself. Divide by emails sent and you include everything that bounced, got filtered and never arrived. Divide by emails delivered, as Saleshandy does, and you quietly remove your own bounce problem from the measurement. Divide by recipients who opened, as Belkins once did and much of the market still does, and you are measuring replies among people your tracking pixel happened to register, which is a shrinking and self-selected group.

The second choice is what counts as a reply. Saleshandy classifies every response as interested, not interested, auto-reply, out of office, or do not contact, and their headline 4.5 percent includes the auto-replies and the out-of-office messages. Belkins excludes auto-replies and bounce notifications. If a fifth of your responses are holiday autoresponders, that single decision moves your reported rate by a fifth.

The third choice is whether open rate is measured at all. Belkins disabled open tracking across every campaign in the study period, and their reason should end the argument: the tracking pixel was hurting deliverability across the industry in 2024. Apple's Mail Privacy Protection inflates the rest. An open rate above 50 percent now suggests machine opens rather than human interest, and lemlist treats anything under 15 percent as evidence the mail is landing in spam. The metric has become a deliverability signal rather than a measure of attention.

Which leaves one defensible way to run this. Pick a definition, write it down, and never change it: replies divided by emails sent, auto-replies excluded, positive replies counted separately. Your number will look worse than the blogs. It will also be the only number on your dashboard that means anything, and the only one you can compare to last quarter.

Did cold email get harder, or did the measurement get honest?

Both, and the only clean evidence of real decline sits inside a single dataset. Comparing publishers tells you nothing, because they measure differently. Comparing one publisher against itself, on one definition, across one year, tells you everything.

Belkins ran 7.5 million emails through the same methodology from January to December 2025. The first half of the year averaged 0.50 percent. The second half averaged 0.40 percent. That is a 20 percent fall in twelve months, same denominator, same client base, same definition of a reply. February peaked at 0.54 percent and December bottomed at 0.35 percent. September and October settled back to 0.45 percent, which suggests seasonality on top of the trend rather than pure decay.

A 20 percent annual decline is a real headwind and a manageable one. It means a campaign that worked last year needs the list tightened or a channel added this year, which is ordinary maintenance. It does not mean the channel is finished. The stories claiming a 60 percent collapse since 2019 are describing the 8.5 percent comparison that was never valid in the first place.

The honest reading is that two things happened at once. Reply rates drifted down as inboxes filled and filters tightened. At the same time the industry stopped being able to measure optimistically, because open tracking became a liability and the serious vendors moved to denominators that flatter nobody. Some of the apparent crash is the market finally reporting what was always true.

What changed in the inbox between 2024 and 2026?

Two mailbox providers wrote rules, and both drew the line at 5,000 messages a day. That number is now the single most important threshold in outbound, and most founders running a campaign have never heard it.

Google's sender guidelines split senders in two. Under 5,000 messages a day to Gmail accounts, you need SPF or DKIM. Over it, from 1 February 2024, you need SPF and DKIM and DMARC, the From header has to align with either the SPF or the DKIM domain, and marketing messages have to support one-click unsubscribe. The spam threshold is where it gets unforgiving: keep complaints reported in Postmaster Tools below 0.30 percent, and Google's own monitoring guidance asks you to stay under 0.10 percent and "avoid ever reaching a spam rate of 0.30% or higher". Three complaints in a thousand is the ceiling. One in a thousand is the target.

Microsoft followed with the same threshold. Proofpoint's write-up of the April 2025 announcement sets out the terms: domains sending over 5,000 emails a day to outlook.com, hotmail.com and live.com need SPF and DKIM passing and DMARC at minimum p=none aligned with one of them, and from 5 May 2025 non-compliant mail goes to the Junk folder with eventual outright rejection. Functional reply-to addresses, a working unsubscribe and list hygiene are on the same list.

Nothing in either rulebook bans cold email. What both rulebooks ban is unauthenticated volume from a domain nobody maintains, which was the dominant way agencies ran outbound in 2021. The founder who sent 18,000 emails from two badly configured domains did not meet a dead channel. They met a compliance threshold and lost.

What changed	When	The actual rule	What it breaks
Gmail bulk sender rules	1 February 2024	SPF, DKIM and DMARC over 5,000/day, spam under 0.30%	Shared or unauthenticated sending domains
Outlook high-volume rules	5 May 2025	SPF and DKIM pass, DMARC p=none minimum, or Junk	The same, on Microsoft consumer inboxes
Open tracking usefulness	Through 2024	Pixels hurt deliverability, Apple inflates opens	Open rate as a performance metric
Published benchmarks	2025 to 2026	Vendors moved to sends-based denominators	Any comparison against an older number

Who does cold email still reach?

Founders and owners at small companies, and the reply rate falls away sharply as the company gets larger. Belkins breaks their 7.5 million emails down by seniority and headcount, and these two cuts move results more than any copy change on offer.

By seniority, founders and owners replied at 0.57 percent, C-level at 0.42 percent, and VPs lowest at 0.32 percent. By company size the spread is wider: companies with 0 to 10 employees replied at 0.72 percent, 11 to 50 employees at 0.49 percent, and companies with 10,000 or more at 0.22 percent. A founder at a ten-person business is more than three times as likely to answer you as somebody at an enterprise.

For an agency this is the most useful finding in the entire 2026 benchmark literature, and almost nobody publishes it. If your clients are SMBs and your buyer is the owner, cold email is still aimed at the easiest audience it has. If you sell to enterprise procurement, the same effort produces a third of the replies and a sales cycle that outlasts the campaign budget.

Geography moves it too, in the same dataset. Poland returned 1.43 percent across 46,000 emails, Ireland 0.74 percent, Denmark 0.73 percent, Canada 0.63 percent across 266,000, the US 0.51 percent across more than 5.6 million, and the UK 0.48 percent. The pattern is saturation: the more cold email a market already absorbs, the lower the return, and the US absorbs the most of anyone.

What separates a 0.4 percent campaign from a 5 percent one?

Four things, in the order they matter: who is on the list, how many are on it, whether a second channel runs alongside, and whether a person answers fast. Copy is fifth, and it is the one most founders rewrite first.

List size has the clearest first-party evidence. Saleshandy's analysis of 30.31 million emails found campaigns targeting fewer than 200 prospects replying at close to twice the rate of campaigns targeting over 500, and Cleverly cites mailforge data putting lists of 50 recipients or fewer at 5.8 percent against 2.1 percent for lists over a thousand. Every incentive in the tooling points the other way, because uploading 10,000 rows takes the same ten minutes as uploading 200.

The second channel is where the biggest verified gap sits, and it is the finding agencies act on least. Saleshandy measured positive reply rates by channel mix across the same dataset: email only returned 0.11 percent, email plus LinkedIn 0.21 percent, email plus a call 0.28 percent, and email plus LinkedIn plus a custom touch 0.88 percent. That is an eightfold difference on one consistent denominator. In the same breath they report that 99 percent of the sequences in their data were still email-only. The single highest-return change available to most agency campaigns is the one almost nobody has made.

Follow-up discipline is third and it is free. Instantly's 2026 report puts 58 percent of replies on the first email and 42 percent across the follow-ups, recommends 4 to 7 touches spaced three to four days apart, and finds elite performers keeping first-touch emails under 80 words. Backlinko's 2019 figure of a 65.8 percent lift from one extra follow-up is the same finding from a different decade.

Then the offer, which decides whether a reply is positive or merely polite. A stranger who cannot price what you do or picture who it is for replies with a question at best, and the diagnosis for that is in our guide on why an agency offer does not convert on cold outreach. The method for building a list from live signals rather than a firmographic filter is in our guide on building a lead list from buying signals.

How do you work out whether cold email is worth it for your agency?

Work backwards from one closed client, because that is the only number that settles the argument. The channel is worth it when the cost of the meetings that produce one client is comfortably below what that client is worth, and that comparison is specific to your deal size and close rate rather than to any published average.

The pressure to get this right is well documented. Promethean Research's 2026 State of Digital Services, built on 119 agency leaders surveyed in February 2026, reports average after-tax net margin at 13 percent for 2025 and average revenue growth at 7.5 percent, and names sales, profit margins and lead generation as the biggest concerns heading into 2026. Agencies that cut their service list grew 13 percent on average and posted 30 percent net margins, which is the same lesson the list data teaches: narrower converts better.

Here is the chain, with an illustrative agency rather than a client. Say you are at $120,000 a month, your average client pays $4,000 a month and stays nine months, so one client is worth about $36,000. You want three new clients a month. At a 25 percent close rate on meetings held and an 80 percent show rate, three clients means 15 meetings booked a month.

Step	How to get it	Illustrative figure
Clients needed a month	Revenue target divided by average deal	3
Meetings booked	Clients divided by close rate and show rate	15
Contacts written to	Meetings divided by 0.5 to 2.5 percent	600 to 3,000
Sends a month	Contacts times 4 to 7 touches	2,400 to 21,000
Mailboxes	Sends divided by 20 to 30 a day each	4 to 35

The 0.5 to 2.5 percent meeting rate is Cleverly's published range, which they describe as one or two meetings per hundred emails sent. Treat the wide end of that chain as the real answer: 600 contacts and 3,000 contacts are both defensible estimates for the same 15 meetings, and which one applies to you depends entirely on list quality.

For the honest floor, look at what a specialist publishes about itself. Belkins reports booking over 1,200 appointments across those 7.5 million emails in 2025, which works out at roughly one appointment per 6,000 sends. Set that against Cleverly's one or two per hundred and the gap is two orders of magnitude. Nobody is lying. One figure describes a tightly targeted campaign and the other describes an annual average across a large book of client work, including the campaigns that did not land. Your own number lives somewhere in between, and sixty days of your own data is the only way to find it. Turning the sends figure into mailboxes and domains without tripping a spam threshold is covered in our guide on how many domains and inboxes cold email needs.

What does cold email cost to run properly in 2026?

The software is cheap and the operating cost is not, which is why the channel looks like a bargain and then quietly is not. List prices on the sending layer have barely moved in three years.

Layer	Tool and tier	Published price, October 2026	What it covers
Email sending	Instantly Growth	$47 a month, or $37.60 on annual billing	5,000 emails a month, 1,000 contacts, unlimited mailboxes
Email sending	Smartlead Base	$39 a month, or $32 on annual billing	6,000 sends, 2,000 contacts, unlimited mailboxes
Mailboxes	Google Workspace Business Starter	EUR 6.80 per user a month, standard rate	One mailbox per user, 300 user maximum
LinkedIn	HeyReach Growth	$79 per sender a month, or $64 on annual billing	One LinkedIn sender seat, unified inbox

Four mailboxes, a sender and one LinkedIn seat lands under $160 a month at list prices. That is the number people quote when they say outbound is cheap, and it is accurate as far as it goes.

What it leaves out is the part that decides the outcome. Somebody has to build the list from signals, verify the addresses before they touch a mailbox, write and approve the drafts, watch the spam rate against Google's 0.30 percent ceiling, and answer every reply the same day. That is a real job rather than a subscription, and the choice between a hire, an outsourced provider and a built system is worked through in our guides on hiring an SDR or building the system first and outsourcing lead generation versus bringing it in-house. The full tool stack with costs at three volumes is in our guide on building an AI outbound system.

If you would rather not assemble and run it, Ourix builds the outbound system inside the business and runs it: verified lists of the buyers you want, outreach on email and LinkedIn, every reply in one inbox, and a pipeline view of what got booked, on your own domains and accounts. Research, list building, first drafts and follow-up run on AI, and a person owns the judgement and every reply. Campaigns go live in about 30 days, everything is yours to take over at any time, and prices are not published: the build is a one-time fee scoped to your channels and your market, and you get the exact number on the call.

Is cold email legal for an agency in 2026?

Yes, on both sides of the Atlantic, and the requirements are narrower than the panic suggests. Neither regime demands prior opt-in for business-to-business mail, and both demand things most campaigns are missing.

In the United States, the FTC's CAN-SPAM compliance guide sets out accurate From and Reply-To headers, a subject line that reflects the content, clear disclosure that the message is an advertisement, a valid physical postal address, and a working opt-out. The deadlines are specific: the opt-out mechanism has to work for at least 30 days after sending, and requests must be honoured within 10 business days. Penalties run to $53,088 per email, and the guide says plainly that the law makes no exception for business-to-business email. Hiring a provider does not move the liability, because both the company promoting the product and the company sending the mail can be held responsible.

In the European Union, the relevant text is Recital 47 of the GDPR:

"The processing of personal data for direct marketing purposes may be regarded as carried out for a legitimate interest."

That is the basis B2B cold email rests on, and the wording is permissive rather than absolute. The same recital conditions it on whether the recipient's interests override yours, "taking into consideration the reasonable expectations of data subjects based on their relationship with the controller". In practice that means writing to a work address about something relevant to that person's job, keeping the volume to one person modest, and honouring an opt-out at once. A recital is interpretive context rather than a binding article, national rules on top of the GDPR differ, and none of this is legal advice.

Where is cold email the wrong channel for an agency?

Four situations, and in each the money is better spent elsewhere. The first is an enterprise buyer. At 0.22 percent for companies of 10,000 or more in Belkins' data, against 0.72 percent at the small end, you are paying three times as much per reply for a committee that will outlast the campaign.

The second is a low-ticket offer. At a $1,000 monthly deal and a nine-month life, one client is worth $9,000, and the chain above needs 600 to 3,000 contacts worked properly per three clients. The arithmetic can clear, but the margin for a bad list is gone.

Third is a business still working out who the buyer is. Outbound at volume industrialises whichever guess you hold this month, and the list has to be rebuilt each time the guess changes.

Fourth, and most common, is having nobody to answer. A reply that waits until Thursday is a reply you paid for and discarded, and the fix for that gap is reply handling rather than another tool. If meetings are needed inside two weeks, mailbox warmup alone rules this out, and paid traffic or the phone is the honest answer. Where referrals fit alongside all of this is covered in our guide on getting agency clients without relying on referrals.

What to do next

1. Write down one definition of a reply and keep it: replies divided by emails sent, auto-replies excluded, positive replies counted separately. Recalculate your last campaign on it before you judge the channel.
2. Check the two thresholds. If any sending domain is over 5,000 messages a day, confirm SPF, DKIM and DMARC are passing and aligned, and read your Gmail spam rate against the 0.30 percent ceiling.
3. Cut the next list to 200 contacts chosen on a live signal rather than a firmographic filter, and verify every address before it touches a mailbox.
4. Add the second channel. Email plus LinkedIn against the same list is the one change with an eightfold gap behind it, and a reply on either channel should stop the sequence on both.
5. Set the sequence at 4 to 7 touches, first email under 80 words, and name the person who answers every positive reply the same day.
6. Run 60 days, then compare positive replies and meetings held against your own baseline. Never against a vendor's average.
7. If you would rather have the system built and run for you, take the free growth audit and go through your own numbers first, whoever ends up building it.