A cold outreach strategy that finds the emails already working and keeps the replies you earn.

A company that sells employee training was drawing 1 positive reply per 100 contacts in week one from a five-email sequence to HR directors and training managers, and the owner was ready to rewrite all of it. The cold outreach strategy I used instead changed one email at a time. Every two weeks the team counted interested replies per email in the sending tool, and I rewrote the weakest, with the list, the offer and the pace of about 500 new contacts a week unchanged.

By week twelve the sequence was drawing 3 positive replies per 100. No single rewrite doubled anything, and two moved it backward and were swapped back. The first count had already shown the first and fourth emails drawing most of the interested replies, which a full rewrite would have thrown out.

Why a cold outreach strategy starts with a count before a rewrite.

Instantly, which sells a cold email platform (and which I use), analyzed sends across thousands of workspaces from January 1 to December 18, 2025: 58% of replies came from the first step and 42% from follow-ups. Smartlead's step data from the first half of 2026 shows the same shape.

Each follow-up earns fewer replies than the one before and still earns some, and in the cases I've seen, many of the late replies are about timing. Judge each follow-up by its own count of replies and unsubscribes, and keep it until the count says otherwise.

John Dawes of the Ehrenberg-Bass Institute estimated in 2021 that about 5% of business clients are in the market in a given quarter, a heuristic drawn from firms switching banks or law firms about once every five years. An HR director who ignores an email in March may need a program in June.

Claude Hopkins built Scientific Advertising (1923) on "keyed advertising, by traced returns, largely by the use of coupons" (chapter 1), and he judged ads by the replies worth having: "One ad may bring too many worthless replies, another replies that are valuable." A step report is the same coded coupon.

Step 1. Set up the count in the tool you already have.

Define an interested reply first: it asks about the offer, agrees to a call or names a better time. Out-of-office replies, unsubscribes and "not interested" stay out of the count.

• Instantly: the campaign's step analytics show replies and opportunities per step and variant.
• HubSpot: Sales > Sequences, open the sequence, then the Step performance report on the Performance tab (Sales Hub or Service Hub, Professional or Enterprise).
• Salesforce: Sales Engagement cadence reports show which step drew the most engagement.
• No CRM: a spreadsheet row per interested reply, using the quoted text to see which email it answered.

Divide by the people who received each email, because in a rolling sequence the newest contacts haven't reached the late emails yet.

At a larger company, count per rep and per group, because a step that works for operations directors can fail for CFOs.

Step 2. Rewrite the lowest email, one every two weeks.

Rewrite the email with the fewest interested replies per 100 recipients, changing one thing (the body or the subject line). In the training sequence the replacements went in this order:

1. The "just bumping this" follow-up became two lines about new managers in their first 90 days.
2. The customer-story email became one sentence about a company that trained new supervisors before its busy season.
3. The third email's subject line now names the problem, not the service.
4. The last email's subject line, which repeated the first, got its own.

With counts this small, a gap of one or two replies is a tie.

Step 3. Give every email a reason it's arriving now.

Of 23 "personalized" cold emails that reached me in one week, one congratulated "Carlos Calle" on its growth as if my name were a company. A reason names something the prospect is working on and why the email arrives now, like "Your demo requests go to a shared inbox, and you just posted two SDR roles."

In a 1978 copier experiment, Ellen Langer's team asked 120 people to let a stranger go first. For a small favor any "because" worked, even "because I have to make copies." For a big favor the empty reason did no better than none, and only an actual reason raised the yes rate. The study tested strangers, not prospects.

A meeting request is a big favor. In a Gartner survey of 632 people involved in B2B purchases (fielded in 2024), 73% said they actively avoid suppliers who send irrelevant outreach, which puts the account at risk along with the reply. Smartlead's median sender in 2026 put a custom variable beyond the first name or company in 75% of emails (Smartlead can't judge if those were relevant). When three in four emails carry a merge field, the reason is what sets yours apart.

Step 4. Put the phone call after the prospect has seen your name.

More dials from an unknown number usually get the same result faster. Answers go up when who's calling and why changes.

1. Test warm names first: people who replied, accepted a connection request or visited the pricing page.
2. Open on context: something they did in public or in reply, like a post or the email they answered. In Gong Labs and 30 Minutes to President's Club data from more than 300 million cold calls (2024), an opener that leads with peers you've worked with succeeded 11.24% of the time, against 2.15% for "Did I catch you at a bad time?" Gong sells call software and doesn't define "success", which makes the gap the useful part.
3. Test the order and gaps between email, connection request and call.

Register outbound numbers at the Free Caller Registry, which passes them to the providers that label US calls.

Skip pre-recorded drops, because the FCC ruled in 2022 that ringless voicemail to a mobile phone is a call that needs consent, and check mobile numbers against the National Do Not Call Registry, because the FCC's do-not-call rules cover mobile numbers too. Check state and country rules before calling, and at a larger company, have legal review them.

Step 5. Treat a wrong-person reply as an introduction.

"Not me, try our operations director" often gets marked as a no. In my own campaigns, those replies are tagged wrong person in Instantly, and a webhook passes each one to Claude, which pulls out the name or title the prospect suggested.

In one week, 27 replies came in, 6 were marked wrong person, and 4 of those named a person or a role. Clay found a verified email for all 4, each got a short email opening with who suggested them, and I read every draft first. Three of the four replied within two weeks, and one booked a call.

One mistake to avoid is taking names from the signature, like an assistant's, instead of names written in the message. Without an AI step, a person adds the name to a "2026-11-wrong-person-replies" list and sends the email in the asset below. At a larger company, check the CRM first, because the suggested person may already belong to another rep's account or an open deal.

Step 6. Protect the inbox before you add volume.

Gmail requires senders to personal Gmail accounts to keep the Postmaster Tools spam rate below 0.3% (ideally below 0.1%). Senders of 5,000 or more a day, counted across every mailbox on one primary domain, also need SPF, DKIM, DMARC and one-click unsubscribe.

From November 2025, Gmail has been stepping up enforcement, including rejections of mail that misses these rules. Yahoo has matching rules, and Microsoft has rejected non-compliant high-volume mail to Outlook.com, Hotmail.com and Live.com from May 5, 2025, and all three cover personal accounts.

• Send from a separate, warmed domain with SPF, DKIM and DMARC.
• Keep volume modest (Smartlead's 2026 median mailbox sends 14 cold emails a day) and bounces under the 2% it suggests.
• For US recipients, follow CAN-SPAM, which the FTC says makes no exception for business-to-business email: a postal address, a clear opt-out, honored within 10 business days.
• Count replies, not opens, because Apple's Mail Privacy Protection hides opens from senders (why email reports look good while the pipeline stays flat).
• Keep cold email out of the newsletter platform, because Mailchimp's acceptable use policy asks for evidence of consent.

A worked example of the two-week count.

Example numbers: a five-email sequence starts 1,000 new contacts over two weeks, and fewer reach the late emails.

By raw count email 1 wins, and per recipient email 4 is the strongest. Emails 2 and 5 sit at the bottom one reply apart, which is a tie, and email 2 loses it on unsubscribes (6 against 3) and gets rewritten this round.

Common mistakes and the pushback a specialist would raise.

"Two weeks is too short to be significant." At these counts it usually is, which is why the sheet is a decision rule, not a significance test. A gap of one or two replies is a tie, and a rewrite stays only after it holds for two counts. With fewer than about 10 interested replies in a window, pool two windows before ranking, because at that volume a zero on a few hundred recipients is often chance.

"Why not A/B test the step instead?" Where the sending tool supports variants, as Instantly does, test one step at a time. Side by side, the season and the list are the same for both versions.

"What if every step drops at once?" A fall across all five usually starts with inbox placement, not copy. Check the Postmaster Tools spam rate and bounces first.

"Doesn't email 4 get credit for emails 1 to 3?" It does. The count credits the answered email, and earlier emails may have set up that reply, which is why a rewrite changes one email and the old version goes back if the sequence total falls.

A test you can run in the next four weeks.

1. Pick a sequence that has run long enough for every email to be sending, with at least 500 new contacts every two weeks, set up the sheet and write down what counts as interested.
2. Week 2: count and rewrite the lowest email, changing one thing.
3. Week 4: count again and compare that email's interested replies per 100 recipients to week 2.

A result is the rewritten email beating its old version by more than two interested replies at about the same number of recipients, while the sequence's interested replies per 100 contacts hold or rise and the list stays unchanged. With fewer than 10 interested replies a window, run it for eight weeks and compare the two four-week totals. If nothing moves, look at the list (fix the list before the copy).