Prove delivery before you test anything
Every cold email campaign is a GTM experiment: does this segment care about this offer? But a campaign can only answer that if the emails arrive. If they land in spam, a zero-reply result says nothing about your market. It only says something about your mail setup.
This chapter covers the sending setup an experiment needs: how many mailboxes, which domains, SPF, DKIM, DMARC, reverse DNS and warmup. Then it shows how to confirm delivery before you spend your real sample, with a case where a missing DNS record looked exactly like a failed offer.
Why delivery comes first
A reply rate looks like one number. It is really two questions multiplied together: did the email arrive? and did it persuade? You can't tell them apart by looking at replies.
Say you send 300 emails and get no replies. Maybe the offer is wrong. Maybe the segment doesn't have the problem. Or maybe 290 of those emails sat in spam folders and nobody saw them. All three look the same in your dashboard.
If you read that zero as a market verdict, you drop an offer that might work. Then you rewrite the copy, pick a new segment, and send again into the same spam folder. Each round burns fresh companies you can't email again for a clean test.
So the rule is simple: a zero is only a verdict once delivery is proven. Everything in this chapter serves that rule.
How many mailboxes an experiment needs
Size your setup to the experiment, not to a monthly sending target. An experiment needs a few hundred sends over one to three weeks. That is a small setup.
Keep each mailbox at 25 cold emails per day or fewer, and no more than 4 mailboxes per domain. Above that, one bad day on one mailbox can hurt the whole domain.
Example: Sizing One Experiment
| Planned sample | 300 sends |
| Test window | 10 business days |
| Sends per day | 30 |
| Per mailbox, with headroom | 10–15 per day |
| Mailboxes | 3 |
| Domains | 1–2 |
Headroom matters. Warmup keeps running on the same mailboxes, and a follow-up step adds sends on later days. If a mailbox has a problem, you want the others to carry the test without pushing past the daily limit.
The sample size itself comes from your economics and the threshold you set, not from your mailbox count. Chapter 4 explains how. Work it out first, then buy the mailboxes that sample needs.
The free Experiment Planner turns your deal size, conversion and cost per email into a threshold and a sample size. That tells you how many sends, mailboxes and days one experiment needs before you register a single domain.
Plan your experimentDomains and mailboxes
Never send cold email from your main domain. If your product lives at yourcompany.com, send from separate domains like yourcompany.io or tryyourcompany.com. If an outreach domain gets flagged, your product emails, invoices and support replies keep arriving.
Choosing domains
- Use a reputable registrar with reliable DNS management.
-
Stick to common TLDs like
.com,.io,.coor.net. TLDs often abused by spammers draw extra filtering. -
Stay close to your brand. A prospect who checks the sender domain should recognize it.
getacme.ioworks for a company called Acme. - Keep WHOIS details public. A hidden registration is what anonymous senders do. Public details look like a real business.
- Redirect each domain to your website. Someone who types the sending domain into a browser should land on your real site.
- Register early. New domains have no reputation. Give them two to four weeks before the experiment starts.
Where to host mailboxes
Google Workspace and Microsoft 365 are familiar, but their terms don't allow cold outreach. An account can be suspended mid-test, which leaves you with half a sample and no verdict. Dedicated outreach mailboxes are built for this kind of sending and keep problems isolated to one domain.
Whichever you pick, the host decides your sending IP. That IP needs its own checks, covered in the next section.
Naming and profiles
Use addresses that look like a person: james@getacme.io or sarah.m@getacme.io. Avoid info@, sales@ and noreply@. Give every mailbox a real display name and a plain signature with your company name and website.
Sellvance registers outreach domains, sets up redirects and DNS records, and creates mailboxes from one dashboard. Daily sending limits are enforced per mailbox, so an experiment never pushes one address too hard.
See email infrastructureDNS: SPF, DKIM, DMARC and reverse DNS
Gmail, Outlook and Yahoo check these records before they decide where your email goes. Since 2024, Google and Yahoo enforce authentication for bulk senders. Without these records, copy and targeting don't matter, because nobody sees the email.
SPF (Sender Policy Framework)
A DNS TXT record listing the servers allowed to send as your domain. The receiving server checks it. If your sending server isn't listed, the email fails.
Example:
v=spf1 include:_spf.google.com ~all
DKIM (DomainKeys Identified Mail)
Your mail server signs every email with a private key. The public key sits in your DNS. The signature proves the email came from your domain and wasn't changed on the way. Without it, your email looks the same as a spoofed one.
DMARC (Domain-based Message Authentication)
Tells receiving servers what to do when SPF or DKIM fails: nothing, quarantine or reject. It also sends you reports on who sends as your domain.
Example:
v=DMARC1; p=quarantine; rua=mailto:dmarc@yourdomain.com
Start with p=none while you check the reports. Move to p=quarantine once everything passes.
Reverse DNS (PTR) for the sending IP
SPF, DKIM and DMARC are about your domain. Reverse DNS is about the server. The IP address your mail leaves from needs a PTR record that points to a hostname, and that hostname must point back to the same IP.
Receiving servers treat an IP with no PTR record as a likely spam source. Many send its mail straight to junk. Your domain records can all pass, and the email still never reaches the inbox.
Check it: find the sending IP in the Received headers of a test email, then run dig -x <ip>. No answer means no PTR. Only the owner of the IP (your mail host or server provider) can set it.
Warming up your mailboxes
A new mailbox has no history. If it starts sending 25 cold emails on day one, it looks like a hacked account. Warmup builds a history first: the mailbox exchanges real-looking emails that get opened, replied to and moved out of spam.
Plan at least two weeks of warmup before the delivery test, and three to four if you can. Warming up a few mailboxes by hand with colleagues works, but it takes time. Most teams use a warmup service.
What to look for in a warmup service
- A private, vetted sender pool. Open pools full of low-reputation accounts can teach providers to distrust you.
- Gmail, Outlook and Yahoo addresses in the pool, since that is where your prospects read email.
- Varied, natural content, not repeated placeholder text.
- No tracking pixels or tracked links in warmup emails.
- A slow ramp, starting at a few emails a day.
Warmup timeline
| Week | Warmup Emails/Day | Cold Emails/Day | Focus |
|---|---|---|---|
| Week 1 | 5–10 | 0 | Warmup only |
| Week 2 | 10–20 | 0 | Warmup; run inbox placement tests |
| Week 3 | 15–20 | 5–10 | Delivery test (100–150 emails) |
| Week 4+ | 10–15 (ongoing) | Up to 25 | The experiment, once delivery passes |
Keep warmup running during the experiment. If you stop it, reputation slowly drifts down while the test is still in progress, and the second week of sends arrives worse than the first.
How to confirm delivery
Passing DNS checks and finishing warmup don't prove your emails arrive. They only remove the obvious reasons they wouldn't. Before the real experiment, get direct evidence. Use all three checks below.
1. Inbox placement tests on the real message
Send the exact email you plan to test, from the exact mailboxes you plan to use, to a set of seed accounts at Gmail, Outlook and other providers. Then check where it landed: inbox, promotions or spam.
Test the real message, not a "hello, test" email. Links, wording and formatting all affect filtering. A clean test email tells you little about the copy you will actually send. Rerun the test after any change to copy, mailboxes or mail host.
2. Out-of-office replies as a delivery signal
Placement tests use seed accounts. Out-of-office auto-replies come from your real prospects' mail servers. In any list of business addresses, some people are on holiday or away. Their servers answer your email automatically, without anyone reading it.
That makes auto-replies a clean delivery signal. They don't depend on whether the offer is any good. And Gmail and Outlook don't send them for mail that went to spam. So out-of-office replies mean your emails are reaching inboxes. A long run of sends with none at all means they probably aren't.
To use this, measure your own baseline: the share of sends that come back with an auto-reply when delivery is known to work. Then compare each campaign against it. Don't count auto-replies as interest, and don't delete them before you count them.
3. A delivery test before the experiment
Send 100–150 emails to prospects you don't plan to use in the experiment. Use the real message and the real mailboxes. You are not reading replies for interest here. You are checking two things:
- Bounces stay low and don't cluster on one mailbox or domain.
- Out-of-office replies start coming back.
If auto-replies show up, start the experiment. If none come back, stop. Rerun placement tests, check SPF, DKIM, DMARC and the PTR record, and fix what you find. The test costs a hundred or so companies. Skipping it can cost the whole experiment.
Sellvance runs inbox placement tests on the real message from your real mailboxes. Its Signal Check reads out-of-office replies as a delivery verdict with a confidence interval, and blocks the interest verdict while delivery fails.
See inbox placement testsA real case: the zero that was a DNS record
In July 2026, we ran our own outreach at Sellvance: 337 emails across 13 campaigns, all testing one offer. Replies: none.
That looked like a clear answer. Thirteen campaigns, nothing. The easy conclusion was that nobody wanted the offer.
Then we checked delivery. Across those 337 sends there were zero out-of-office replies, against a baseline we had measured at 1% when delivery worked. Getting none by chance is unlikely (p = 0.034). Delivery failed, so the reply rate could not be read at all.
What We Found
| Emails sent | 337 (13 campaigns, one offer) |
| Out-of-office replies | 0 (baseline 1%) |
| Delivery verdict | Fail (p = 0.034) |
| Interest verdict | Could not be read |
| Cause | Sending IP in South Africa with no reverse DNS record |
SPF, DKIM and DMARC all passed. The domain was fine. The server wasn't: the IP had no PTR record, so receiving servers filtered the mail before anyone saw it.
The zero looked like a verdict on the market. It was a verdict on a PTR record. Without the out-of-office check, we would have dropped the offer and moved on to a new one, sending into the same filter.
Before you start testing an offer
- Mailboxes sized to the planned sample, at 25 cold emails per day or fewer
- Outreach domains separate from your main domain, redirected to your site
- SPF, DKIM and DMARC passing
- PTR record on the sending IP, pointing back to the same IP
- Two or more weeks of warmup, still running
- Inbox placement test passed on the real message
- Delivery test of 100–150 emails with out-of-office replies coming back
The Experiment Playbook walks through the full loop: prove delivery, define a segment, hold one offer fixed, and read one verdict. Delivery is the first step, for the reasons in this chapter.
Read the Experiment PlaybookGet to a Delivery Verdict First
Sellvance sets up domains, DNS and mailboxes, runs inbox placement tests, and won't read interest until delivery passes. So a zero means what you think it means.
Start your first experiment