Yesterday I set up the sending identity for The Gap, the weekly newsletter I write. It took about twenty minutes and one decision, and the decision was the interesting part.
The newsletter does not send from the address I email clients from. It never will.
That sounds like a small technical choice. It is actually the thing that decides whether the emails you care about most keep arriving at all.
The two jobs an email domain does
Your domain does two completely different jobs, and they pull in opposite directions.
The first job is correspondence. You reply to a client, send a proposal, answer a question. Low volume, high stakes, sent to people who already know you and asked to hear from you. If one of these lands in a spam folder, a deal quietly stalls and nobody tells you why.
The second job is bulk. A newsletter, a sequence, an announcement. Higher volume, sent to a list, and some of those people will not open it. Some will mark it as spam, not because it is spam, but because that button is closer than the unsubscribe link.
Every mailbox provider keeps a reputation score against the domain that sent the message. It does not care that you think of these as two different activities. Same domain, same score.
So the moment you send bulk from your correspondence domain, you have connected the two. A bad week on the newsletter, a list that went stale, one complaint spike, and the score drops. What drops with it is not just the newsletter. It is the proposal you sent on Tuesday.
What I actually did
The Gap sends from a dedicated subdomain, set up for exactly this purpose, with its own authentication records. The reply address is my normal working address, so anyone who replies reaches me the way they expect.
Three things follow from that split, and they are the whole reason to do it:
The reputations are separate. If the newsletter has a bad month, the damage is contained to the subdomain that sends it. My client correspondence is scored on its own behaviour, which is low volume and almost never complained about.
I can be honest about list hygiene. Because the downside is contained, I can send to a real list and let people leave. The alternative, where every unsubscribe feels like a threat to your main domain, quietly pushes you toward the worst habits: not offering an easy unsubscribe, holding on to addresses that have not opened anything in a year, sending less often than you should because each send feels risky.
The unsubscribe link is not optional anyway. POPIA has its own requirements about direct marketing consent, and the large mailbox providers now require a working one-click unsubscribe for bulk senders regardless of what the law where you are says. I would rather build the thing that makes compliance the easy path than rely on remembering. I wrote about that shift in thinking when I stopped shipping features and started reading the law.
The part I got right by accident
I did not plan this well. I planned it late.
The newsletter had already sent five issues before I sat down and thought properly about sending identity. It worked fine, because five issues to a small list is not enough volume to hurt anything. The reason to fix it now is that the cost of this decision is asymmetric in a way that is easy to miss.
Setting it up correctly at the start costs twenty minutes. Setting it up correctly after your main domain has been marked down costs months, because reputation recovers slowly and there is no support desk to appeal to. You cannot buy your way out of it and you cannot rush it. You just send carefully and wait.
That asymmetry is the entire argument. Not "this is best practice", but "one version of this is cheap now and the other version is expensive later, and you do not get to choose which one you are in once you are in it".
It is the same shape as most of the operational things I have got wrong and then fixed. Cheap to do properly up front, disproportionately expensive to retrofit, and completely invisible until the day it is not. Running my own marketing has taught me more of these than I expected, which is roughly what I found when I started using AI agents to run my own marketing.
If you send anything to a list
You do not need my setup. You need the split.
Whatever you send in bulk, send it from something other than the domain your clients reply to. Give it its own authentication. Keep a working unsubscribe that takes one click. Let people leave without making it hard, because the people who stay are the only ones who were ever going to matter.
And if your newsletter is currently going out from the same address your invoices do, that is worth twenty minutes this week. Not because something has gone wrong. Because the whole point is to do it before it does.