Cold email infrastructure setup: domains, mailboxes and build order

Infrastructure before volume, and why that order is not negotiable

The sequence that fails is familiar. Somebody buys a sequencing tool, connects a mailbox, uploads a list and starts sending. Replies are thin, so volume goes up. Deliverability degrades quietly, because nothing announces it, and by the time anyone notices the pattern the domain has a reputation problem that takes longer to repair than it took to create.

The fix is unglamorous: build the whole sending estate before the first campaign, and accept that the build has a fixed minimum duration you cannot buy your way out of. Everything below is the specification GrowthStack works to when a company is opening a new market, most recently for a diagnostics equipment manufacturer entering the United States.

Two notes on that example, because it shapes several decisions later on the page. It is a hardware business rather than software, and it sells the same equipment to genuinely unrelated buyers. Both facts change the build.

How many sending domains, and why never your main one

Cold outbound generates spam complaints even when it is done carefully. Some proportion of strangers will mark an unsolicited message as spam no matter how relevant it is, and that is a cost of the channel rather than a sign of failure.

The decision that follows is about where that cost lands. If you send from your primary domain, the reputation damage attaches to the domain that also sends invoices, contracts, support replies, password resets and everything else the company depends on. The failure mode is not that outbound stops working. It is that your finance team's emails start going to spam and nobody connects the two.

So: buy separate domains for sending. Close in spelling to the main one, so a recipient who checks recognises the company, and registered properly rather than through a throwaway service. More than one, because spreading mailboxes across several domains means a problem on one does not take the whole programme down, and because a single domain carrying every mailbox concentrates exactly the risk you are trying to distribute.

These are consumable assets. If a domain's reputation is damaged beyond repair, it gets retired and replaced, which is a manageable event when it is a sending domain and a serious one when it is the company's.

How many mailboxes, and the question people ask backwards

Six to eight mailboxes across those domains is a sound starting architecture for a company testing a new market. That is the number, and the reasoning behind it matters more than the number itself.

The usual question is "how many emails can I send from a mailbox". The useful question is the reverse: how much volume does the motion need, and how many mailboxes does it take to deliver that volume while keeping each individual mailbox well below anything that looks unusual. Volume per mailbox is the most common self-inflicted deliverability wound, and multiple mailboxes exist to reduce it, not to multiply the total.

You will find confident daily figures published everywhere. Treat them carefully: the safe rate for a mailbox depends on how old the domain is, how far through warm-up it is, how the list was built, and how people are responding. A mailbox on a six month old domain sending to a verified, well-targeted list behaves differently from a new one sending to a scraped export, and no universal number covers both.

Start conservatively, watch the per-mailbox numbers weekly, and increase slowly. The ramp is part of the build rather than something that happens after it. How long warm-up takes and what to monitor is covered in the cold email deliverability checklist, along with the authentication that has to be in place first: SPF, DKIM and DMARC on every sending domain, no exceptions, before a single campaign message goes out.

Google Workspace or Microsoft 365

Either works when it is configured properly, and the honest answer is that this decision matters far less than the three that surround it.

The one real consideration is where your buyers read their mail, since provider-to-provider delivery is the path being tested. Beyond that, running a mix across the estate is defensible, because it avoids concentrating your entire sending reputation with a single provider and gives you somewhere to move if one side develops a problem.

What is not optional either way is that these are properly licensed business mailboxes on domains you control, with authentication configured per domain. The shortcuts that exist in this area are the reason the category has a reputation problem.

Data, verification and the suppression list nobody builds

Three components, and the third is the one routinely skipped.

• A data source for identifying and building the list. For a US programme this is usually Apollo at this scale, with the list built against an agreed ICP rather than pulled as a broad export. The tool is the cheap part of this decision. The targeting is not.
• Email verification before anything is sent. Bounce rate is one of the fastest ways to damage a new domain, and unverified data bounces. Verification runs on every list before it enters a sequence, not once at the beginning of the quarter.
• A suppression framework. A single list, maintained across every campaign and every mailbox, of people who unsubscribed, bounced, asked not to be contacted, are existing customers, are in an active sales conversation, or are a competitor. Without it you will eventually cold-email a current customer, which is the sort of thing that reaches the founder the same afternoon.

Suppression is the least interesting item on any infrastructure list and the one whose absence causes the most avoidable damage. Build it before the first send, because retrofitting it means reconciling records across several tools after the fact.

One product, several motions

This is the part of the build that is strategy rather than plumbing, and it is where most infrastructure-first programmes still fail.

The manufacturer above sells the same imaging equipment into markets that have almost nothing in common. Clinical and veterinary buyers care about throughput, image quality and patient workflow. Industrial inspection buyers care about detecting defects in components, compliance and inspection cycle time. And a third group is neither: the existing installed base, who already own the equipment and are candidates for upgrades or a service contract, and who need a completely different conversation because they are already customers.

The tempting approach is one campaign describing the equipment, sent to everybody. It underperforms with all three, because a message written to work for a vet clinic and an industrial inspection lab is necessarily written about the product rather than about either buyer's situation.

So the outbound is separated into distinct motions from the start: different lists, different messaging, different sequences, tracked separately. The infrastructure supports this directly, because separate mailboxes and domains make it straightforward to keep motions isolated and to see which one is producing replies and which one is producing complaints. That is also why segmentation appears in the setup phase rather than being treated as an optimisation to do later. Deciding it afterwards means rebuilding the lists.

The general method for defining those segments is in ICP, persona and messaging and the B2B ICP framework.

Reply routing and reporting

Two things that get postponed until the first campaign is live, at which point they are urgent.

Reply routing decides what happens when someone answers. Across six to eight mailboxes on several domains, replies arrive in six to eight places. Without a defined route into one inbox with one owner, the interested replies sit unread next to the out-of-offices, which wastes the only outcome the whole build exists to produce. The standard worth holding is a first response within one business hour, measured in the buyer's working day, which is covered in speed to lead.

Reporting has to run per mailbox and per domain, not only per campaign. Campaign-level reporting tells you whether the message is working. Mailbox-level reporting tells you whether the infrastructure is degrading, and those fail independently. A campaign with a healthy reply rate on five mailboxes and a quiet problem on the sixth looks fine in aggregate right up until it does not.

What it costs to run, and whose name it is in

For a build of this shape, third-party costs come to roughly $250 a month: domains, mailbox licences, a sequencing tool, a data source and email verification. The exact figure moves with the final mailbox count and how much data the programme consumes.

The more important point is whose name those accounts are in. At GrowthStack, third-party software, data, domains and email infrastructure are paid for by the client and stay with the client. The reason is not administrative. A sending estate takes weeks to build and accumulates reputation over months, and if it is registered to whoever ran your outbound then ending that relationship means starting the warm-up clock again from zero. Infrastructure held in the client's name is the difference between changing supplier and rebuilding.

That is separate from the cost of the work itself, which is set out on the pricing page.

Why this splits into two phases

The build and the running of it are different jobs with different shapes, and combining them into one commitment obscures both.

Phase one is the setup above: domains, mailboxes, authentication, warm-up, data, verification, suppression, segmentation, the first messaging architecture, reply routing and reporting. It has a defined end, and the output is an asset the client owns whatever happens next.

Phase two is running it: account and contact research against the agreed ICPs, list building and hygiene, copywriting and sequencing, execution and optimisation, mailbox and deliverability monitoring, reply management, follow up, qualification, handing over meetings, and reporting on what is working.

A ninety day initial engagement is the recommendation, and warm-up is the reason. The configuration takes days. The reputation-building cannot be compressed, because it is observed by mailbox providers rather than completed by you. Ninety days leaves enough room to build correctly, test the market properly, find which segment and message actually work, and then put volume behind the one that does. Anything shorter tests the infrastructure rather than the market.

Related reading

The rules this build has to satisfy, including the Google and Yahoo bulk sender requirements and the spam rate threshold, are in the cold email deliverability checklist. What goes into the sequences once the estate is ready is in outbound sequence structure, and the full ninety day shape is in the outbound 90 day plan.

For the engine this sits inside, see B2B lead generation services. For US market entry specifically, B2B lead generation for the US.