Learn / Lead lists
Segment, test, and cut: finding the list that converts
Most cold email list segmentation starts the same way. Open Apollo, add a few filters for your ideal customer, build one big list, start sending. Those filters come from who you THINK your buyer is, or from the people who found you through inbound. Neither one describes a stranger. Someone who came to you already knows your name and already suspects they have a problem. The person you are about to cold email knows neither.
So a profile built on a hunch quietly aims an expensive campaign at people who were never going to answer. This page is the written version of the training below: how to cut a list into segments built from clients you actually got results for, how much volume each one needs before its number means anything, and how to read the outcome honestly. It is the same system behind over 3,000 sales calls booked in the past 12 months.
Build the segments from results, not assumptions
Start with case studies. Yours, or your client's. Pull every one of them, get the websites, and then ask for the commonalities as a matrix rather than a list. The breakdown that holds up: business model, then industry, then company size.
Run that on a marketing agency with a deep case studies page and it falls out fast. Three business models: direct to consumer, multi-channel retail, and subscription. Underneath them the industries, apparel, wellness, home, travel. Across them the size bands, under 50 employees, 11 to 50, 200 to 500. Every cell carries an example client they already won with.
What you have now is better than an ideal customer profile. You have a distribution: which cells hold the most clients you have proof for. That is your testing order, and the market built it instead of a guess in a filter box.
From there, two ways to turn a cell into a list. Take one (direct to consumer, travel accessories, 11 to 50 employees) and convert it into keywords and job titles you can plug into Apollo by hand. Or feed a single seed company into a lookalike tool like Company Enrich, choose broad or targeted, add employee count and location, and let it hand you the companies. One seed returned 157 luggage companies in the United States. Tighter targeting means a smaller, better list, which is the trade, and lookalike lead lists covers how to work it.
Whichever route you take, count the reachable prospects in every segment before you commit to testing it. A segment you cannot reach at volume is not a segment, it is a preference. Sizing that properly is the same math one level up.
The number nobody wants to hear: 1,000 sends per sub-segment
This is where most segment tests quietly die. People split the list, send a bit to each bucket, then read numbers that cannot carry a conclusion. Send 100 emails, get 2 positive replies, and you genuinely do not know whether that audience is strong or whether you got lucky twice. You will still plan the next quarter around it.
The floor is 1,000 sends per sub-segment. Not per segment. Per SUB-segment, because the entire reason to split a list this finely is being able to say which exact filter combination produced the result.
Five segments, fourteen sub-segments, and the volume each one owes you:
| Segment | Sub-segments | Prospects to validate |
|---|---|---|
| Direct to consumer | 6 | 6,000 |
| Multi-channel retail | 2 | 2,000 |
| Subscription | 2 | 2,000 |
| Marketplace | 1 | 1,000 |
| Financial | 3 | 3,000 |
| Total | 14 | 14,000 |
Two emails per contact makes 28,000 sends. Run the whole test across two weeks, which is 10 sending days, and you need roughly 2,800 sends a day.
Size the infrastructure to the test
2,800 a day sounds heavy until you do the arithmetic. All Google, at 15 sends per inbox per day, is 187 accounts, roughly $350 to $400 a month. All Microsoft, at 5 sends per inbox, is 560 accounts and comes in slightly under Google. A mix of the two is the real answer and lands between $250 and $350 for both. Buy the domains, hand them to an infrastructure provider, fill in an order form. Google publishes the bulk sender requirements you are sending against, so read them once instead of guessing at what volume is safe.
Under-testing a segment does not save money. It buys you a number you cannot use.
Reading the results without flattering yourself
Here is a launch we ran recently. Five segments across two data sources, with one deliberate constraint: every segment got the same offer and the same copy. The only things allowed to vary were who we wrote to and where the list came from. Round one validates the audience and the source, nothing else.
Three campaigns went to creator sub-niches, with the lists built from LinkedIn. They came back with 19, 10, and 24 interested replies, a 12 to 18% interested rate.
Two campaigns went to agencies, one to email agencies and one to ecommerce agencies. Both of those segments came straight off the client's best case studies, so on paper they were the safe bets. Both lists came out of Apollo. The reply rates were fine, 5% on one and almost 7% on the other. Interested replies landed at 1% on both. We sent roughly 5,000 on one and 8,000 on the other, and no segment ran under 1,000 prospects, so those are results rather than noise.
The creator segments beat the agency segments about tenfold. Same offer. Same copy.
Reply rate is the trap in that data. The two losing campaigns had the better reply rates and almost nothing worth booking, which is a busy inbox and an empty calendar. The reply rate benchmarks are worth knowing, but the ranking never changes: interested replies beat replies, and booked calls beat interested replies.
The cut line, and what a loss actually tells you
One booked call per 1,000 messages sent. That is the line I look for. Above it, the segment survives and earns more volume in the next round. Below it, it goes, and the volume it was eating goes to a winner.
Both agency campaigns were well under. The easy read is "agencies do not work," and it would have been wrong. The agency lists came from Apollo, which is about as inundated as a data source gets. Everyone mails those people, so there is a good chance the offer had already walked past them in somebody else's email. The creator lists came from LinkedIn engagement: post commenters, competitor community followers, people who left a public trace of what they care about.
So we kept the target and changed the source. The agencies we still want are being reached through alternatively sourced data instead of written off on one bad round. "Cut this segment" and "cut this data source" are different conclusions, and holding every other variable still is the only thing that lets you tell them apart.
Then run it again
Phase 2 changes exactly one thing. Job titles, maybe, now that the audience is settled. Or the offer itself, now that segment and source are validated and a change in results can finally be attributed to the copy. One variable per round, 1,000 sends per sub-segment, cut the losers, push their volume into the winners, repeat. Every round shrinks the part of the machine you are still guessing about.
The honest caveat, better from me than at 28,000 sends: this needs volume. If your whole market cannot support 14,000 prospects across the segments you want to test, run fewer segments, or accept that you are making a judgment call instead of reading a result. Both are fine. Pretending the second one is the first is not.
tl;dr: build segments from clients you already got results for, give every sub-segment at least 1,000 sends, hold the offer and copy constant, and cut anything under one booked call per 1,000 sent. Then change one variable and go again.
FAQ
How many emails do I need to send to test a cold email segment?
At least 1,000 sends per sub-segment, and the sub-segment is the unit that matters. Below that, two positive replies could be a strong audience or two lucky sends, and you cannot tell which. If you are testing 14 sub-segments, that is 14,000 prospects, and 28,000 sends at two emails per contact.
How do you know when to cut a cold email segment?
One booked call per 1,000 messages sent is the line. Above it, the segment stays and earns more volume next round. Below it, it goes. Judge on booked calls rather than reply rate, because a segment can produce a 7% reply rate and a 1% interested rate at the same time, which is a busy inbox and an empty calendar.
How should I split a cold email list into segments?
Build the split from clients you already produced results for, not from an assumed buyer profile. Pull every case study, get the websites, then group them by business model, then industry, then company size. Each cell of that matrix is one testable segment, and the number of case studies in each cell tells you which one to test first.
Why do Apollo-sourced lists underperform?
Because everyone mails them. Agencies and ecommerce operators in particular get contacted daily off the same filters, so your offer has probably already walked past them in somebody else's email. In one launch, identical offer and copy pulled roughly ten times the interested replies from LinkedIn-sourced lists as from Apollo-sourced ones.
Should I test different copy and different segments at the same time?
No. Hold the offer and the copy identical across every segment in round one, so the only variables are the audience and the data source. If the copy moves at the same time, a losing campaign tells you nothing about who to keep. Change one thing per round: segment first, then source, then titles, then the offer.
If you would rather have the segmentation, the lists, the infrastructure, and the copy run for you, that is what we do: over 3,000 sales calls booked for clients in the past 12 months. Or keep learning free, every training grouped by topic lives on the trainings hub, and the full video for this one is on YouTube.
PS - one client booked 47 calls in 45 days. Segmenting the lead list was only one part of that build, but it is the part that decides who ever sees the offer in the first place.