When cold email works for an expert firm, and when it wastes your name

Cold email works for an expert firm in one shape. The list is short and every name on it is chosen. The message proves you did the reading. And your presence holds up when the recipient searches you.

Cold email fails in the shape most people picture when they hear the phrase. Buy a list, load a sequence, send thousands, wait.

Those two are different activities that happen to use the same tool.

The size of the list predicts the result

Hunter, which sells outreach software, publishes a yearly report on how cold email performs. The 2026 edition draws on 31 million emails sent by its users in 2025.

The average sequence in that data went to 449 recipients and drew a reply rate of about 5%.

Sequences sent to 21 to 50 recipients replied at about 6%. Sequences sent to more than 500 replied at about 2%.

Cutting the list did more for the reply rate than any change to the writing. Relevance is hard to keep at scale. A recipient can tell within a line whether the sender knew anything about them.

The same report finds two more patterns. Emailing one or two people at a company beats emailing three or more. And sending from your own domain roughly doubles the reply rate against a free address.

Read those findings with the source in mind. Hunter sells a tool that sends sequences, so it has a commercial interest in cold email working. What it does not have is an interest in telling customers to send less. Its own data says exactly that.

Treat every published reply rate as a range

Reply-rate figures vary by a factor of ten across the industry, and the disagreement is about measurement.

Belkins, an outbound agency, reports an average reply rate of 0.45% across its 2025 campaigns. That is one reply for every 222 sends. The figure counts strictly cold contacts and excludes the open-rate inflation that flattered older studies.

Platform-wide averages that include warmer lists, marketing campaigns, and link-building outreach land three to eight times higher.

Both figures can be accurate. The two are measuring different things.

So the useful move is to hold the published numbers loosely and measure your own. A practice sending 30 emails a month has no statistical population anyway. What it has is a list of named people and a record of who replied.

What the people receiving these say

Hunter also surveys decision makers. Their answers are more instructive than the send data.

The top complaint is no longer irrelevance. It is that the email feels too eager to sell, named by 65% of respondents. Irrelevance sits close behind at 61%. Personalization drawn from public information moves in the other direction: 67% say it makes them more likely to reply.

About a quarter of decision makers say they never receive a cold email worth their attention. Roughly six in ten receive something useful at least once a month.

The bar is on the floor, and it is still rarely cleared. For a practice with 20 years of specific expertise, that is an opening. The thing in short supply is exactly what you have.

What the precision version looks like

Start from accounts rather than volume. Name 20 to 40 organizations you could serve well and would be glad to have as clients. The constraint is the point: a list that size forces you to know who you are writing to.

Choose one person per organization. Find the person whose problem you solve, by name.

Then do the reading. Look at public filings, conference talks, hiring pages, and changes in their market. You are trying to see what this year looks like from their chair. The research is most of the work, and the writing gets easy once it is done.

The message itself is short. Say what you noticed. Say what it usually means in your experience. Then ask a low-pressure question. Asking whether you can send something useful outperforms asking for a meeting.

Send three messages in total, spaced over two or three weeks. One message leaves replies on the table. A fourth and fifth start costing you the goodwill the first three earned.

Send from your own domain, in the volumes a person could plausibly write by hand. The mailbox providers' bulk-sender rules apply to senders in the thousands per day. A practice sending a few dozen sits outside them. What still applies is your complaint rate. Enough people marking you as spam will follow your domain for a long time.

Where cold email fails

Bought lists fail first. The addresses are stale. The bounces damage your domain. And nobody on the list asked to hear from you.

Secondary domains and rotating inboxes fail next, and the reason is worth naming. That infrastructure exists to spread the damage of sending more mail than one domain can survive. A practice that needs it has already decided to be a volume sender.

Cold email also fails when there is nothing behind it. Anyone interested enough to consider replying will search your name first. A thin presence ends the conversation before it starts. This is the whole reason the lookup does the persuading. Outreach can open the door. It cannot build the credibility the recipient checks for.

And several licensed professions place limits on approaching prospective clients directly. The rules differ by state and by board, and they change. If you hold a license, check yours before the first send.

What AI changed, and what it did not

The research is where the machine earns its place. Reading the public material for 40 organizations takes hours. Pulling out what each is dealing with, and drafting a first pass, now takes one.

The words are a different matter. In the same survey, 69% of decision makers were bothered by machine-written email. And the send data shows manually edited emails outperforming fully automated ones. Both point the same way. A model that studied the recipient carefully will beat a person who studied nobody. A model producing fluent copy about a company it knows nothing about is the noise everyone is already deleting.

So the division holds. Point the machine at the research and the qualifying. Write or edit every message yourself. And send them yourself, because every one of them goes out under your name.

There is a second thing AI changed, and it works against you. The cost of sending plausible personalized mail fell to nearly zero. Volume rose. Filters and recipients both got harder. The inbox you are writing into is more defended than it was two years ago.

The advantage nobody advertises

The competent version of this is slow. Twenty emails at 30 minutes each is a working day. The people selling outreach software cannot sell a working day.

The time cost is why the approach still works. Almost nobody spends it, so the inbox fills with the cheap version. A message that clearly took an hour reads as an unusual event.

Paid channels price access to demand that already exists, and cold email prices attention in your own hours instead. For a practice with a short list of accounts worth having, those hours buy more than the auction does.