Most people answer this question by reading a blog post. It is arithmetic, so let me just do it.
Say you want to send 1,000 emails a day. You can send about 20 a day from a Google inbox before you are cooked. So 1,000 divided by 20 is 50 inboxes. You can hold a maximum of 2 inboxes per domain without the domain looking like a farm. So 50 divided by 2 is 25 domains.
That is the whole calculation, and if you would rather not do it by hand there is an inbox scale planner that runs the same sums. Everything below is the detail that decides whether those numbers hold.
Start from meetings, not from sends
The 1,000 a day above is the wrong place to start. It is a vanity input.
Work backwards instead. Decide how many meetings you need a month. Work out how many positive replies that takes. Work out how many sends that takes at whatever your reply rate actually is. Only then do you have a daily send target, and only then does the inbox maths mean anything.
Most people pick a send volume first because it feels like a plan. It is not a plan, it is a budget you have set yourself without knowing what it buys.
The per provider limits, which is where the generic formulas break
Nearly every guide gives you one rule of thumb and stops. The problem is that the two providers worth using behave completely differently, and using one set of numbers for both is how a domain dies in week one.
Google: 15 to 20 emails per day per inbox. Maximum 2 inboxes per domain.
Azure: 2 to 5 emails per day per inbox. Up to 50 inboxes per domain.
Read those twice, because the shape is inverted. Google gives you volume per inbox and almost no inboxes per domain. Azure gives you barely any volume per inbox and then lets you stack fifty of them on one domain.
Forty a day per inbox is a ceiling, not a target. If you are sending at the ceiling you have no room left when something goes wrong, and something goes wrong.
Run both. Two providers means no single point of failure, and that is not a theoretical concern. I will come back to it.
There is also a reason to prefer these two beyond the limits. Across 8 million cold emails of platform data, Azure and Google mailboxes pulled nearly twice the human reply rate of Outlook. That is not a preference, it is the measurement.
Buy more than the maths says
Buy 1.5 to 2 times the inboxes you think you need.
This sounds like a vendor telling you to spend more, so here is the reasoning. Domains get blacklisted. Inboxes burn out. When it happens mid campaign, the cost is not the replacement inbox, it is the two to three weeks of warmup before the replacement can send anything. If you have spares already warmed, you swap and keep going. If you do not, your campaign stops for a fortnight.
The way to run this is two sets. Set A goes through 14 day warmup and starts sending. Set B stays on long term warmup in the background, ready to slot in when mailboxes from Set A burn out. It is the cheapest insurance in outbound.
Warmup is a calendar problem
You can't buy your way out of this one.
Fourteen days minimum before anything cold touches those inboxes. In practice two to three weeks. Ramp to about 20 warmup emails a day per inbox during that window.
Use a private or isolated warmup pool rather than your sequencer's public one. In a public pool you inherit the behaviour of every other sender in it, which means someone else's bad campaign drags your reputation down while you are still warming up.
Skip warmup and you are in spam by day three. Not eventually. By day three.
Protecting the inboxes once you are sending
Buying the infrastructure is half of it. The other half is not wrecking it in week one, and this part costs nothing.
Spintax everything. Subject lines, opening lines, calls to action. No two emails going out should read identically. A thousand byte-identical messages landing in a thousand inboxes is the single clearest pattern a filter can see.
No links in the first touch. Links hurt deliverability badly enough that great copy with a link in it will lose to mediocre copy without one, because the mediocre one arrives. Move the link to a follow up or to your reply. This includes the one in your signature.
Turn off open tracking. Open tracking works by loading a pixel, and that pixel costs you inbox placement. You are trading a number you should not be optimising for anyway against the thing that decides whether anyone reads the email. Turn it off.
Plain text only. No HTML templates, no images, no branded header. If it looks like marketing it gets treated like marketing, and cold email is not marketing, it is a message from one person to another.
Validate the list before it goes out. Every time, whichever database you pulled from. Bad data means bounces, bounces mean a wrecked sender reputation, and a wrecked reputation means the fifty inboxes you just bought are worth nothing. This takes about two minutes and it is the highest return two minutes in the whole process.
Never send from your main domain
One rule with no exceptions and no cheap version.
If a cold campaign damages your sending reputation, and campaigns do, you want that damage contained somewhere that is not your invoices, your support replies and your recruiting email. Register secondary domains and route every cold campaign through those.
What to check before you buy from a vendor
The price being low is not the signal. What sits behind the price is.
- Are they an official Google Cloud or Microsoft partner?
- Are they reselling trial or education panel accounts? If yes, your deliverability tanks or the inboxes disappear.
- Is setup actually automated, or does the platform look instant and then deliver in two days?
- Are domains pre screened and blacklist checked before you buy them?
In August 2025 Google removed millions of inboxes overnight. People logged in to find responses gone and domains burned. The accounts that vanished were the ones built on non profit domain loopholes, education licences used commercially, and reseller panels priced at a dollar fifty a month. Clients on properly provisioned infrastructure carried on sending and lost nothing.
That is the whole argument for the boring version. Cheap infrastructure is not cheap, it is borrowed time.
What I do not do
I don't run placement tests, and I don't sit over a monitoring dashboard grading every metric.
I look at the overall picture instead of over analysing every detail. If reply rates tank, I go and look at the copy and the lead lists first, because that is usually where the problem started, and then I buy more infrastructure. You can lose a fortnight to dashboards that were only ever going to tell you what a week of sending would have told you anyway.
The disclosure
I am CRO of InboxKit, which sells cold email infrastructure, and some of the tool links on this site earn affiliate revenue. So my view on what you should spend is not neutral and you should read it accordingly.
The reason I ended up there is the same reason I write posts like this one. I was a customer of all of these tools first. I spent years buying infrastructure, watching it fail in specific and expensive ways, and building workarounds. I joined the company whose product solved the problem I had been having.
That doesn't make me objective. It does mean the numbers above come from running this, not from a pricing page. Where I think a tool is the wrong choice for you, I say so on the tools page, including for tools I earn from.
The short version
Work backwards from meetings. Google at 15 to 20 a day per inbox and 2 inboxes per domain, Azure at 2 to 5 a day and up to 50 per domain. Run both. Buy double. Warm for 14 days minimum on a private pool. Never send from your main domain. Check what your vendor is actually reselling.
Then stop optimising the infrastructure and go and fix your list, which is where the result actually lives. There is more on that in cold email personalisation: the list does the work.