Rebuilding Deliverability on a New Sending Domain

A personal care brand we manage migrated to a new sending domain this summer. A new domain has no reputation, which means every inbox provider is deciding in real time whether you are a sender worth delivering. Eight weeks later the account's standard campaigns open at 38% instead of 27%, Gmail opens are up 18 points from the post-migration baseline, and monthly campaign revenue is higher than it was when we were mailing twice as many people per send.

This is the full playbook for recovering email deliverability on a new sending domain, with the real numbers: what we cut, what we kept, what it cost (almost nothing), and the one mistake mid-recovery that got 24% of a send's Gmail volume bounced in an afternoon.

The Starting Point: A Fresh Domain and a Dead Tail

Before the changes, a standard campaign went to roughly 68,000 people and opened at 26-28%. Broken down by inbox provider, Gmail sat at 26% opens and Microsoft (Hotmail and Outlook) at 18% with bounce rates over 1%. Not a crisis, but on a domain with no accumulated trust, every unengaged recipient you mail is a vote against you. The fastest way to burn a new domain is to keep force-feeding the people who stopped looking.

So we ran a seven-send audience test before deciding anything, tagging engagement and intent layers on every campaign so we could read revenue per layer. Three findings made the decision for us:

• Engagement fell off a cliff at 30 days. The 0-30 day engaged layer opened 48% and carried 89% of its strategy's revenue. The 31-day-plus layers combined were 48% of send volume but 11% of revenue, opening at 1-3%.
• Recent buyers were the revenue. Buyers from the last 180 days earned $0.14 per recipient. Every other intent bucket was around $0.01.
• Email engagement, not site activity, separated alive from dead. A cut of no-site-activity profiles who were email-engaged captured 100% of that bucket's orders. The rest produced zero.

Move 1: A Layered Core, Not an Engaged-Only List

The new standard campaign audience became a union of three dynamic segments: email-engaged in the last 30 days, OR purchased in the last 180 days, OR browsed in the last 30 days. Everyone else stopped receiving standard campaigns. Nobody was suppressed or deleted; the moment a dormant profile opens or clicks anything (a flow, a sale email), they re-enter the engaged segment and start receiving campaigns again automatically.

The core is a union, not a filter chain. The tail is parked, not suppressed: dynamic segments pull anyone back in the moment they show signs of life.

The union matters more than it looks. We have tested engaged-only sending before on a different account, and it quietly leaked real revenue: opens looked great inside Klaviyo while Shopify showed money leaving. Engagement filters miss the buyer whose opens do not track and the lapsed customer who converts twice a year. Layering recent buyers and browsers into the core is what protects the revenue that engagement metrics cannot see. It is the same logic as segmenting by purchase behavior, applied to deliverability.

Eight weeks of results, comparing the four full-list standard sends before the switch to the nineteen layered-core sends after it:

Read that middle row twice. We cut 47% of recipients per send and revenue per recipient did not move. The ~32,000 people removed from each send were contributing almost nothing except negative engagement signals to Gmail and Microsoft. Total monthly revenue went up because cadence carried it: smaller, healthier sends, slightly more often.

Move 2: Tighten Only the Providers That Are Punishing You

A month in, the per-provider breakdown still showed Gmail and Microsoft as the weakest placements. Instead of shrinking the whole core, we added a provider-scoped exclusion segment: profiles on Gmail or Microsoft with no opens in 14 days, no clicks in 30, no site activity in 14, and no orders in 30. About 7,000 recipients per send. We applied it to roughly every other campaign and compared treated sends against untreated ones:

Cutting 7,000 dead profiles from a send cost zero revenue. The treated sends earned more per send, not just per recipient. The honest caveat: this is a small sample and the treated sends came later in the recovery, so some of the open-rate lift is the domain healing. The bounce reduction is the cleanest signal, and it is unambiguous.

Mid-migration and seeing the same per-provider pattern? Talk to us before you cut anything list-wide.

Move 3: The Full List Still Has a Job

The dormant tail is not abandoned; it is reserved for the moments it can actually convert. For a tentpole sale, the open and close sends go to the full list. The middle sends stay on the core. When we ran this during a sitewide sale, the full-list blasts opened at 25-29% while the same week's core sends opened at 35-36%. Even with a sitewide discount in the subject line, the tail barely opens, which is the whole premise of the strategy confirmed on live data. Two full-list sends per sale is signal the domain can absorb. Ten a month is how you end up back where we started.

Below the campaign layer, the usual hygiene kept running: an automated sunset flow suppressing the truly dead, and the fundamentals from our deliverability guide (authentication, consistent volume, engagement-weighted sending) underneath everything.

The Incident: How to Bounce 24% of Gmail in One Send

Mid-recovery, one campaign went out and 3,681 Gmail recipients (24% of the send's Gmail volume) came back as bounces with SMTP 550 "UnsolicitedMessageError." On a warming domain, that error message stops your heart.

The cause was not the audience strategy. The campaign had been cloned from an old template that carried the brand's previous from address, on a domain that no longer matched the DKIM signing domain. Gmail saw bulk mail from an unauthenticated domain and refused it, exactly as DMARC intends. Google Postmaster Tools confirmed the real domain's reputation stayed High through the whole episode, which is how we knew it was a one-send authentication problem and not a reputation collapse.

Two lessons worth stealing. First, from addresses ride along on clones: if you have ever changed domains, every pre-migration campaign in the account is a landmine, so verify the from address on anything cloned. Second, per-provider bounce forensics plus Postmaster is how you tell "we made a config mistake" from "the domain is burned." The first is fixed in an hour. Panicking about the second when it is actually the first leads to exactly the wrong moves.

The Playbook, Compressed

• Test before you cut. Tag engagement and intent layers on real sends and read revenue per layer. Our cut line (30-day engagement, 180-day buyers, 30-day browsers) came from the data, not a best practice.
• Union engagement with purchase behavior. Engaged-only sending leaks revenue. Engaged-or-recent-buyer-or-recent-browser did not lose a cent per recipient.
• Tighten per provider, not per list. Reputation is scored per provider. Fix Gmail with a Gmail-scoped exclusion, not a list-wide one.
• Keep a full-list valve. Sale open and close sends re-engage the tail and re-enter dormant profiles into the core automatically.
• Watch per-provider bounces and Postmaster weekly. Blended open rate hides everything that matters during a recovery.

One more thing that mattered as much as the strategy: enforcement. Campaign names lied more than once (sends labeled full-list were actually running the core, and vice versa), so every config claim above was verified against the API, not the campaign list. That verification layer, and the assistant that runs it week to week, is its own story: meet Jarvis, our AI account manager.

Eight weeks, no paid tools, no deliverability consultants: a layered core audience, one exclusion segment, and discipline about who gets the full list. If your account is mid-migration or your provider breakdown looks like ours did, reach out and we'll tell you which of these moves applies to your situation.