You have decided to run cold email, or you are already running it and the replies have quietly dried up. Every guide you open says the same five words: SPF, DKIM, DMARC, warmup, ramp. Very few of them tell you the order to do things in, how many inboxes you actually need, what "warmed up" means as a decision rather than a feeling, or what to check every week once the campaign is live.
This is the runbook for email deliverability in cold outreach, written as a sequence rather than a checklist. Each step has a gate: a condition that has to be true before you move to the next one. Most deliverability problems we see were not caused by a missing record. They were caused by skipping a gate, usually the one between "the records exist" and "the records pass," or the one between "warmup is running" and "this inbox is ready for real prospects."
One scope note before starting. Why your sending domain should not be your main domain, and how sending reputation actually attaches to a domain, is its own question, and we answered it in does cold email burn your main domain. This page assumes you already accept that model and want to build the thing correctly.
Step 1: Size the setup before you buy a single domain
The most common infrastructure mistake happens before any DNS record exists. Someone buys one domain and two inboxes, loads a list, and then discovers that the campaign they planned needs ten times the sending capacity they built. The fix is to work backward from the number of new prospects you want to contact each day.
Use four inputs, all of them yours:
- N, new prospects you want to start each business day.
- S, the number of emails in your sequence (the opener plus follow-ups).
- c, the daily cap you set per inbox for cold mail.
- m, the number of inboxes you put on each domain.
Once a sequence is running at steady state, every day carries the openers for today's new prospects plus the follow-ups for earlier ones, so daily volume approaches N × S. The inboxes you need are roughly N × S ÷ c, and the domains you need are that number divided by m. Round both up, then add spares, for reasons covered in step 6.
The value of c is a choice, not a law. The common practice among people who run cold email for a living is to keep each inbox's cold volume low, in the tens per day rather than the hundreds, and to put only a few inboxes on each domain. The logic is that a low per-inbox number looks like a person, and a few inboxes per domain means one domain's trouble takes down a small slice of your capacity rather than all of it. If the math produces a number of domains that feels absurd, that is useful information: it usually means the campaign is trying to send more than your market or your reply-handling can absorb, which is the subject of how many cold messages to send a month.
Gate: you have a written number for N, S, c and m, and the domain count follows from it. Not from what was on sale.
Step 2: Buy domains that pass the human check
Secondary sending domains should be recognizably yours. A variation of your brand on a common extension reads as a real company; a string of random words does not. Buy them on a registrar you can manage DNS on easily, because you are about to edit records on every one of them.
Then point each domain at your real website with a permanent redirect. Prospects who are curious will type the sending domain into a browser, and that domain needs to land somewhere that proves you exist. The redirect does not move sending reputation in either direction; it just passes the check a skeptical reader runs. If you do not have much of a site yet, one honest page is usually enough.
A newly registered domain has no sending history at all. That is not a penalty, but it is not a head start either. It is the reason step 4 exists, and it is why many operators buy domains a little before they need them.
Gate: every domain resolves to your site in a browser, and you can edit its DNS.
Step 3: Publish the records, in this order, and prove they pass
Create the mailboxes on your email provider first, because the provider tells you exactly what some of these records must contain. Then publish the records below for every sending domain. The order matters less than people think; what matters is that you check the result instead of trusting the setup screen.
| Record | What it does | The mistake that breaks it |
|---|---|---|
| MX | Tells the world where mail for this domain is received, so replies and bounces arrive. | Forgetting it on a domain used only for sending, which means replies to that domain vanish. |
| SPF (a TXT record) | Lists which servers may send mail for the domain. | Publishing two SPF records instead of one merged record, or chaining so many includes that it exceeds SPF's limit of ten DNS lookups. Either can make SPF fail outright. |
| DKIM (a TXT or CNAME record) | Publishes the public key that lets receivers verify your messages were signed by you and not altered. | Publishing the key but never switching signing on inside the email provider, so mail goes out unsigned. |
| DMARC (a TXT record at _dmarc) | States what receivers should do when SPF and DKIM do not line up with the From domain, and where to send reports. | Copying a strict policy before anything is verified, or omitting the reporting address so you never see what is failing. |
| Tracking domain (CNAME), if you track | Lets click or open tracking run on your own domain instead of a tracking host shared with thousands of senders. | Using the vendor's shared tracking domain by default. Many cold senders turn tracking off entirely and rely on replies. |
Alignment is the part that actually fails
SPF and DKIM can each pass and DMARC can still fail. DMARC does not just ask whether a check passed; it asks whether the domain that passed it is the same domain the reader sees in the From line. If your provider signs with its own domain instead of yours, or your return path belongs to the provider, the checks pass for somebody else. Configure custom DKIM for each sending domain so the signature belongs to you.
For a new sending domain, a DMARC policy of p=none with a reporting address is the usual starting point. It tells receivers you have a policy and sends you reports, without asking anyone to reject your mail while you are still learning what your setup looks like. Tighten it later if you want; do not start strict and discover a misconfiguration by losing mail.
How to prove it, in five minutes
Send a message from each new inbox to a mailbox you control on a different provider. Open the full message headers, often labeled "show original," and find the authentication results. You want SPF, DKIM, and DMARC each to say pass, and the DKIM domain to be your sending domain. Do this for every inbox, not one per domain. A typo on one mailbox's settings is exactly the kind of fault a spot check misses.
Gate: every inbox shows SPF, DKIM and DMARC passing, aligned to its own domain, in real headers.
What the mailbox providers actually require
Much of the urgency in deliverability content comes from the rules Gmail and Yahoo announced for bulk senders, and later rules Microsoft announced for its consumer mailboxes. They are worth reading in the original, because they are narrower and broader than the summaries suggest.
Google's published sender guidelines separate requirements for all senders from extra requirements for bulk senders, which Google describes as sending around five thousand or more messages a day to personal Gmail accounts. Everyone is expected to authenticate, use secure connections, and keep the spam rate reported in Postmaster Tools low, with 0.3% stated as a level never to reach. Bulk senders additionally need SPF, DKIM, and DMARC together, alignment, and one-click unsubscribe on marketing mail.
Two things follow for a cold B2B campaign. First, most small cold campaigns never come near the bulk threshold to personal Gmail addresses, because their recipients are on business domains. That is not a reason to skip anything in step 3. Business mail systems run their own filtering and treat unauthenticated mail with suspicion for the same reasons consumer providers do. Treat the bulk requirements as the baseline whether or not they technically apply to you. Second, the complaint ceiling is the number that matters most, and it is driven by who you write to and what you say, not by DNS. Records get you considered. Relevance keeps you there.
Whatever the threshold, give every recipient a plain, easy way to opt out, and honor it quickly. That is ordinary good practice and it costs you nothing but the people who were never going to buy.
Skip the DNS weekend entirely
A free 15-minute demo: the businesses we would reach in your niche, the messages we would send by text, email and phone, and the profit math. 3 booked appointments in your first 30 days or get your first month back.
Book your demo →Step 4: Warm up, and know what warmup does not do
Warmup means sending small, gradually rising volumes of mail from a new inbox and generating positive engagement on it, usually through a network of other inboxes that open, reply, and pull messages out of spam. The purpose is to give a brand-new sending identity a short history of being treated like a normal mailbox before it writes to strangers.
Before warmup starts, finish the inbox itself: a real person's name, a profile photo, and a short plain signature. An inbox that looks unfinished is one more signal that it exists only to send cold mail.
Timelines published by warmup vendors cluster around two to four weeks before a new inbox carries real campaign volume. Treat that as a range, not a promise, and judge readiness on evidence instead of the calendar: warmup mail landing in the primary inbox at your test accounts, no authentication failures in your DMARC reports, and no bounces from the warmup network itself.
Three things warmup cannot do, because people pay for it as if it could:
- It cannot fix a bad list. A warmed inbox that sends to unverified addresses will bounce, and bounces undo weeks of warmup quickly. The data problem is upstream of the inbox; why purchased lists fail covers it.
- It cannot fix copy that people mark as spam. Warmup engagement is manufactured. Your prospects' complaints are real, and they outweigh it.
- It does not end. Most operators leave warmup running at a lower level for as long as the inbox sends cold mail, so the inbox keeps a stream of positive engagement alongside the campaign.
Gate: each inbox has a finished profile, has warmed for your chosen minimum period, and its warmup mail is landing in the primary inbox.
Step 5: Ramp real volume like a person would
The switch from warmup to campaign is where most new setups get hurt, because it is tempting to go from zero to the full value of c on day one. Do not. Start each inbox at a fraction of its target cold volume and raise it in steps, holding at each level long enough to see bounce and reply data before the next increase. If anything moves the wrong way, hold or step back instead of pushing through.
Steady beats bursty. A domain that sends the same moderate amount every business day looks like a business; one that sends nothing for a week and then everything on Tuesday looks like a campaign tool, because it is. Spread sends across the working day and across inboxes, and keep weekends quiet.
The list you ramp onto matters as much as the ramp. Verify every address before it reaches an inbox, and start with your best-matched prospects rather than the leftovers, so the first real signals each domain receives are replies rather than bounces and complaints.
The message matters too. A short plain-text first email with no link and no images gives filters little to object to and gives prospects a question to answer. We have argued elsewhere that the booking link belongs later in the thread for reply-rate reasons; it happens to be the deliverability-friendly choice as well. And the more your messages read like a template merged across a thousand rows, the more they behave like one. The signs your outreach looks automated are mostly signs filters notice too.
Step 6: Monitor weekly, and plan for retirement from day one
Once the campaign runs, deliverability becomes a weekly review rather than a project. The useful signals are short:
- Bounce rate per inbox and per domain. A rising bounce rate is almost always a list problem. Stop and verify before you send more from that slice.
- Reply rate per inbox. It is the most honest deliverability metric you have. If identical copy to similar prospects gets replies from four inboxes and silence from the fifth, the fifth is probably landing in spam.
- DMARC aggregate reports. They show which sources sent as your domain and whether they passed. A failing source you do not recognize is worth investigating immediately.
- Google Postmaster Tools. Useful where you send meaningful volume to Gmail addresses, and Google's reported spam rate is where its 0.3% line is measured. At low volume it may show little or nothing.
- Placement tests with your own accounts. Periodically send to inboxes you control across the major providers and note where mail lands. Treat the result as a smoke alarm, not a measurement.
- Blocklist checks. Check sending domains occasionally, and immediately when replies drop off a cliff.
When a domain degrades, the playbook is the same every time. Pause its cold volume, keep warmup running, fix the cause, and bring it back slowly. Do not raise the volume on the remaining domains to make up the shortfall; that is how one bad domain becomes three.
This is why step 1 said to add spares. A cold email operation that runs for months needs a bench of domains already warmed and waiting, so that pausing or retiring a tired domain does not halt the campaign for a month while a replacement warms up. Budget for that bench as a normal operating cost, not an emergency; budgeting a cold outreach campaign walks through the rest of those line items.
Why email alone is the fragile channel
Run this whole runbook correctly and you have built a reliable way to get read by people whose inboxes are guarded by filters you cannot see and cannot appeal to. Every step above is about earning permission from a system that is designed to doubt you. That is the structural weakness of email-only outreach, and no amount of DNS work removes it.
The other channels fail differently. A text to a validated cell phone does not pass through the same inbox placement filters, and a call cannot land in a spam folder. Each has its own rules and its own costs, which we compared in cold email vs SMS vs calls and costed per appointment in cold SMS vs cold email. The point here is narrower: when email is one channel among several, a bad week for one domain is a dip, not an outage.
Build it yourself, or buy the outcome
If you enjoy infrastructure, everything above is learnable, and doing it once teaches you a lot about how your own market reads mail. Expect the first build to take longer than a weekend, and expect the weekly monitoring to stay on your calendar for as long as the campaign runs.
If what you actually want is meetings, the infrastructure is a cost of getting them rather than the goal. That is the trade TaskBlink exists for. We find businesses that match your ideal customer from real-time business data and Google Business Profile signals, validate that each phone number is a real working cell, and run AI-powered outreach by text, email and phone that books ready-to-buy prospects onto your calendar. The deliverability work in this runbook is our problem, not yours. Agencies of every kind use it, from marketing agencies and SEO agencies to web designers and AI automation agencies. Current plans are on our pricing page, and the guarantee is simple: at least 3 booked appointments in your first 30 days, or your first month back.
Want the meetings without the mail server?
In 15 minutes we will show you the real businesses we would contact in your niche, the exact messages, and what the math looks like for your deal size. 3 booked appointments in 30 days or your first month back.
Book your free demo →Frequently asked questions
How long should I warm up a new inbox before cold emailing?
Warmup vendors commonly publish two to four weeks. Judge readiness on evidence rather than the date: warmup mail landing in the primary inbox at test accounts you control, authentication passing in real headers, and no bounces. Then ramp campaign volume gradually instead of switching straight to your full daily cap.
How many inboxes do I need for cold email?
Work backward from your own numbers. With N new prospects a day and an S-step sequence, steady-state volume approaches N times S emails a day. Divide by the daily cap you set per inbox to get the inbox count, divide that by inboxes per domain to get the domain count, round both up, and add warmed spares.
Do Google's bulk sender rules apply to cold B2B email?
Google's bulk sender threshold refers to roughly five thousand or more messages a day to personal Gmail accounts, which many small B2B campaigns never reach. Google's requirements for all senders still apply, and business mail systems filter on the same signals, so the practical answer is to meet the bulk requirements anyway: SPF, DKIM, aligned DMARC, and a low complaint rate.
What DMARC policy should a new cold email domain use?
Most operators start a new sending domain at p=none with a reporting address. That publishes a policy and sends you aggregate reports without asking receivers to reject mail while you confirm that SPF and DKIM pass and align with your From domain. Tighten the policy later once the reports are clean.