Cold outbound returns between 0.5 and 3.4 replies per hundred emails. The spread between those numbers is the entire game.

Two honest numbers, far apart

Belkins analysed 7.5 million cold emails its teams sent between January and December 2025 and reported an average reply rate of 0.45 percent: roughly one reply per 220 emails. Woodpecker, a sending platform, reports an average reply rate of 3.43 percent across more than 20 million emails in its June 2026 update. Both datasets are large and both are credible. The gap between them is not an error; it is the finding.

Belkins' figure comes from agency campaigns run at volume into cold markets. Woodpecker's comes from self-serve users, where small, hand-built campaigns are common, and its own segmentation shows why that matters: campaigns to fewer than 50 contacts averaged a 5.8 percent reply rate, while campaigns to more than 1,000 contacts averaged 2.1 percent. The same dataset attributes 42 percent of all replies to follow-ups rather than the first email. Reply rate is not a property of cold email as a channel. It is a property of how narrow the list is, how well the message fits it, and whether the sequence is finished.

Belkins' segmentation points the same direction from the other side. Replies were highest from founders and owners (0.57 percent) and from companies with fewer than ten employees (0.72 percent), falling to 0.22 percent at companies above 10,000 staff. For an owner-led software company selling to other owner-led companies, the addressable ceiling is meaningfully above the blended average, and the way to reach it is precision, not volume.

The floor rules changed in February 2024

Whatever the message says, it must first arrive, and the terms of arrival were rewritten two years ago. From 1 February 2024, Google's bulk-sender requirements, matched by Yahoo, oblige anyone sending at scale to Gmail addresses to authenticate email with SPF, DKIM and a DMARC policy, to support one-click unsubscribe, and to keep the user-reported spam rate below 0.3 percent, with Google's published guidance to stay under 0.1 percent. Above the threshold, mail faces spam classification regardless of content.

Translate the threshold into campaign arithmetic: 0.3 percent is three complaints per thousand recipients. A stale list, a scraped list, or a message sprayed at people it does not fit will cross that line quickly, and the penalty lands on the sending domain, not just the campaign. This is why competent operators now send from separate warmed domains, verify every address before sending, and treat volume as something earned by engagement rather than a dial to turn. Woodpecker's data puts the average bounce rate at 5.1 percent with good practice below 2 percent; bounces, like complaints, are a list-quality symptom the providers measure.

The practical consequences for an owner-led company are specific. Outbound should never run from the domain the business banks its reputation on; it runs from lookalike domains bought and warmed for the purpose, so a mistake is recoverable. Authentication records are checked before the first send, not after the first blacklist. Every address is verified against a bounce-checking service. And the unsubscribe link is present and honoured, not because a cold prospect is a subscriber in the legal sense in every jurisdiction, but because the person who cannot leave quietly leaves by pressing the spam button, and the spam button is the metric Google enforces.

Why the list decides before the copy is written

Put the two sections together and the order of causes becomes clear. Deliverability failures come from list quality. Spam complaints come from list fit. And the sevenfold spread in reply rates tracks list narrowness more than any copywriting variable in either dataset. The craft in a subject line matters, but it operates inside limits set earlier, when someone decided who exactly should receive the campaign and why those people would recognise the problem in the first sentence.

That decision is positioning work wearing an outbound costume. A list of "UK software companies, 10 to 200 staff" is not a list; it is a directory. A list of companies that share a trigger, a tool, a regulation or a stage, matched to a message about that shared situation, is what produces the 3 to 6 percent campaigns rather than the 0.4 percent ones. In our own delivery this is why the Outbound engine starts from Positioning and ICP rather than from a sending tool, and why reviewing the list and the sender setup is part of our free B2B audit before any sequence is judged.

What the benchmarks do not tell you

Three gaps, stated plainly. Reply rate is not meeting rate: a reply may be a no, and neither dataset above converts cleanly into booked meetings for a given offer; Belkins reports over 1,200 appointments from its 7.5 million emails, which averages one meeting per roughly 6,000 sends across all its campaigns, good and bad. Vendor datasets are drawn from each vendor's own customers, which biases each toward its own best use case. And none of this measures the quiet cost of bad outbound: a burned domain and a poisoned first impression in a market small enough to remember. We have found no reliable published figure for that, and would distrust one.

The arithmetic an owner should run before starting

Decide the number of qualified conversations a month the pipeline needs. Divide by a defensible reply-to-meeting rate for your offer, then by a reply rate you justify from the benchmarks above given how narrow your list really is. If the resulting send volume exceeds what your addressable market can absorb without repetition, outbound alone will not carry the target, and it is better to know that before the domains are warmed. Cold outbound in 2026 rewards exactly one thing: knowing precisely whom you are for, and reaching only them, correctly, repeatedly. Everything else is throughput.