Read past the title: ApartmentIQ's new "Email Marketing Manager" is really hiring someone to keep an AI-sequencer outbound machine — 20 reps, Clay, Smartlead — out of spam. "Warmup" is now a line item in the job description.
TL;DR AI collapsed the cost of producing outbound, so volume exploded while the inbox got stricter. The constraint moved from how much you can send to how much can land — and the two things that decide it, ramp and list quality, are the two nothing in the AI stack measures.
This month ApartmentIQ — a multifamily PropTech data platform — posted for an "Email Marketing Manager." The title reads like newsletters and campaign calendars. The job description reads like something else entirely: the role supports a 20+ person SDR team — the same shape as deliverability getting a director elsewhere running on AmpleMarket, Clay, Smartlead, Instantly, HubSpot, and n8n, with Claude and ChatGPT drafting the copy. Two requirements give the game away — the same greenfield a founding lifecycle hire inherits — "expert knowledge of sender reputation management, inbox warmup, and global privacy regulations," and, under nice-to-haves, "email deliverability expertise: DMARC, DKIM, SPF, warmup strategy, ESP migrations."
That's not a marketing hire. That's a deliverability engineer wearing a marketing title.
Here's what changed. AI didn't make sending email free — every one of those platforms still bills by volume, and more mailboxes and domains cost more, not less. What AI collapsed is the cost of producing outbound: researching accounts, writing personalized copy, building multi-step sequences, and spinning up the mailboxes and domains to send from. The work that used to cap how much a team could send is now nearly free, so volume exploded. The inbox went the other way — Gmail and Yahoo's 2024 bulk-sender rules raised the bar on authentication, complaint rates, and sender reputation. The constraint moved. It used to be "how much outbound can we produce?" Now it's "how much of it actually lands?"
When inbox warmup, sender reputation, and SPF/DKIM/DMARC show up as requirements in a role titled "Email Marketing Manager," the company has quietly discovered that deliverability is its own function now — not a toggle inside the sequencer. Someone has to own whether the machine's output reaches a human at all. ApartmentIQ is staffing that role on purpose. Most teams discover they need it only after reply rates have been sliding for a month.
And it's the part that's easiest to get catastrophically wrong. Point a 20-rep sending operation at your primary domain, skip the readiness check, ramp the volume too fast, and you don't just lose the cold outbound — you torch the sender reputation of every email your company sends, including the invoices and password resets that actually have to arrive. Warmup isn't a switch you flip before you scale. It's a posture: are your authentication records right, are you sending from a domain you can afford to put at risk, and are you ramping at a pace your reputation can absorb?
This is the same signal we keep seeing, now in cold outbound instead of lifecycle: as AI pushes every company's volume up, "did it reach the inbox?" becomes a first-class operational question. Right now, companies answer it the way ApartmentIQ is — by hiring a human to own warmup and reputation full-time. Increasingly, they'll answer it by automating it.
Three things moved at once, and only one of them was in anyone’s control:
Ramp and list quality — and neither is measured by anything in the sending stack. Authentication is the visible half and it is largely solved: per our scan of 117 companies hiring for this work, 92% publish SPF and 97% publish DMARC. Nobody is failing because they forgot a TXT record. They are failing because volume arrived faster than reputation could absorb it, aimed at addresses nobody profiled.
| What AI made cheap | What it did not change (per Google’s guidelines) |
|---|---|
| Researching accounts | How fast a new domain can gain reputation |
| Writing personalised copy | The 0.30% complaint ceiling at Gmail |
| Building multi-step sequences | Whether the addresses are real and consenting |
| Spinning up mailboxes and domains | That each new domain starts at zero history |
The mailbox-and-domain point is the one teams get wrong most often. Spreading volume across many fresh domains looks like risk distribution; it is actually the opposite. Per how reputation accrues, each new domain starts with no history, so you have multiplied the number of cold assets rather than diluted the load on a warm one. Before scheduling the ramp, it is worth checking whether the domain can absorb the volume at all.
| Ramp decision | Feels like | Actually |
|---|---|---|
| Split across 10 fresh domains | Risk distribution | 10 cold assets, 0 warm ones |
| Full volume in week 1 | Fast payback | Week 1 is the reputation |
| Same domain as transactional | Simplicity | A bad campaign takes password resets with it |
Everything the enrichment vendor guessed at, at machine scale. Pattern-generated addresses, role accounts, long-dead domains and carrier gateways all enter the file faster than anyone reviews them. Two categories in particular pass every syntax check and still cost you: email-to-SMS gateways, where the message arrives as a text and carriers treat unsolicited traffic as a complaint, and the share of the list that is a personal mailbox rather than the business address the campaign assumed.
The reflex is to buy verification, which means uploading the list to a third party — often the exact thing a regulated buyer cannot do. Most of what matters is computable locally: syntax, duplicates, role accounts, disposable and carrier domains, and the personal-mailbox share all run in the browser, with only bare domain names ever hitting DNS. That is what the contact list reality check does, and it is the half of deliverability that job postings keep asking for.
Three list-side checks, drawn from the published CC0 domain datasets, are worth running before any AI-built file is loaded into a sequencer:
The metric that justifies the whole machine is the one quietly breaking. Open rates already absorbed one shock when Apple’s Mail Privacy Protection pre-fetched images. The next is structural: assistants increasingly read, triage and summarise mail before a human sees it, so an “open” may be a model, and a reply may be drafted by one. If both ends of a cold-email sequence are automated, the engagement number that proves it works is measuring machines talking to machines.
The practical consequence is that copy has to survive being read by something that does not skim. If an assistant cannot extract the ask, the deadline and the next action, the message fails before a human is involved — you can check what an agent extracts from a message directly. None of this is a reason to stop sending. It is a reason to stop treating open rate as the scoreboard.
Not by being AI-generated. It hurts by changing the volume curve: producing sequences, copy, mailboxes and domains became nearly free, so send volume rose faster than sender reputation can be built. The failure is almost never authentication, which most senders have already configured, but ramp speed and list quality, which nothing in the sending stack measures.
Usually the opposite. Each additional domain starts with no sending history, so spreading volume across several fresh domains multiplies the number of cold assets rather than diluting load on a warm one. Reputation accrues per domain and per IP; splitting traffic splits the evidence that you are a legitimate sender.
Pattern-generated addresses that were never verified, role accounts, long-dead domains, and carrier email-to-SMS gateways, all imported faster than anyone reviews them. Two categories are especially costly because they deliver successfully and so pass every syntax check: gateway addresses that arrive as text messages, and personal mailboxes on a list the campaign assumed was business addresses.
Less every quarter. Apple's Mail Privacy Protection already broke it by pre-fetching images, and assistants that read, triage and summarise mail before a human sees it break it further. When outbound is machine-generated and increasingly machine-read, an open rate can measure automation on both ends. Reply quality and intended-action-taken are more honest signals.