If you've ever bought a new domain on a Friday and tried to do 500 cold sends from it by Monday, you know how cold-email infrastructure actually works. The protocol is simple. The reputation system is not. Gmail, Microsoft, and every business-grade spam filter judges a sender by behaviour over time — and behaviour means a graph of who-talked-to-whom, who-replied, who-marked-spam, who-archived without reading.
The warmup pool is how cold-email tools cheat this graph in your favour. Mailboxes from different customers cross-send each other, mark the messages as read, move them out of spam, and reply ~30% of the time. To Gmail, your brand-new mailbox suddenly looks like a sender that has actual, two-way human relationships.
The problem is that the same mechanism, run naively, is exploitable. One new mailbox sending obvious junk into a pool of established senders can drag the whole pool's reputation down. Every cold-email tool that runs a single-graph warmup pool has had this happen. We've watched it happen to two of them on the way to building ours.
The constraint that drove the design
The brief was simple: a customer with five-year-old Gold-reputation mailboxes should not be paired with a customer who signed up yesterday. Both customers should get value from the pool, but the value should be proportional to what they put in.
So we don't have one pool. We have three.
Tier is per-mailbox, not per-customer. A given customer can have a Gold mailbox they've been running since 2023, a Silver mailbox from last quarter, and a Bronze mailbox they connected this morning. Each one earns its tier on its own behaviour.
New mailboxes enter at Bronze. After 14 days with no spam complaints, no quarantine triggers, and a sustained bounce rate under 1%, they're promoted to Silver. After another 30 days clean, they hit Gold. A spam complaint or a sustained bounce rate above 3% sends a mailbox to quarantine for 7 days; quarantined mailboxes still send their own campaigns, they just don't pair with anyone.
Why Bronze can't reach Gold
This is the part that confuses people. If a Gold mailbox is the most desirable thing in the pool, why won't we pair it with a brand-new Bronze sender?
Three reasons.
1. Spam-trap risk.The probability that a brand-new sender is going to mass-import a list with spam traps in it is non-trivial. We've seen it. Spam traps are addresses that don't belong to real humans; they exist to detect senders who scraped. If a Gold mailbox sends to a spam-trap address via the warmup pool, Google will absolutely punish the Gold mailbox's reputation, not just the new sender's.
2. Behavioural pattern matching. Established mailboxes have established sending patterns — usually 5–8 messages a day to known correspondents, with replies. A brand-new mailbox suddenly sending 50 warmup emails into a Gold inbox stands out. By only letting Bronze pair with Bronze and Silver, the volume rises in line with the established baselines.
3. The incentive is wrong otherwise.If brand-new mailboxes paired with Gold immediately, there'd be no reason to ever build mailbox reputation organically. The whole system would collapse into "spin up a new mailbox, warm it for 24h in the Gold pool, send your campaign, throw it away." We don't want that.
The cost: brand-new senders take 44 days to fully enter the Gold-pairing band. That's deliberate. If you need to send today, you can buy already-warm mailboxes from ColdRelay — our sister product — and they'll enter as Silver. Otherwise, patience.
The three quarantine triggers
Quarantine is the most aggressive action a customer can experience without our intervention. We took our time on what fires it.
- Spam complaint.Any "Mark as spam" from a recipient warmup mailbox fires immediately. There's no threshold; one complaint is enough. We re-evaluate after 7 days clean.
- Sustained bounce spike.Bounce rate above 3% sustained over 24 hours. Lower bounce rates are normal — you'll see 1–1.5% on a healthy list. 3% sustained means the import was bad, or the mailbox itself has a deliverability problem.
- Cross-customer pattern flag.If multiple customers in the pool independently flag a sender as suspicious (the "this looks spammy" button in the unibox), we quarantine. This is rare and currently fires at four flags over 48 hours.
Quarantine doesn't kick a mailbox out of the pool. It just stops it from pairing with anyone new while the team and the customer have a look. The mailbox keeps sending its own campaigns, throttled to the same 90s cadence as before. After 7 clean days, it's released back to its previous tier.
What we deliberately don't do
A few decisions we made on purpose, against the conventional cold-email tool advice:
- We don't auto-rotate IP addresses.Most warmup pool tools pretend each warmup message is from a different IP. Real warmup is supposed to look like a real conversation between two real mailboxes — those don't bounce between IPs.
- We don't fake conversations. Some tools generate fake threads to make a mailbox look like it has back-and-forth replies. We use real customer mailboxes responding to each other; the conversation is artificial but the SMTP envelope is real.
- We don't run warmup during the campaign send window.Warmup happens at low-volume hours for the local TZ. It shouldn't compete with the actual outbound a customer is paying us for.
- We don't recommend warmup beyond 30 days. Once a mailbox is sending to real recipients, the warmup graph adds diminishing value. We turn it down automatically.
Numbers we watch
The pool is in continuous operation across 2,847 mailboxes (as of last Tuesday's snapshot). It exchanges ~84,000 warmup messages a day. Median open rate within the pool is 91% (because every recipient is configured to open). Reply rate is set per-mailbox; default is 30%.
Pool composition right now: 45% Gold (1,284 mailboxes), 34% Silver (980), 21% Bronze (583). 14 mailboxes are in quarantine — the typical churn floor. The number we care about most is the rate of Bronze → Silver promotions; if that drops, we know something in the broader email ecosystem has shifted and we re-tune the bounce/complaint thresholds.
The boring answer to "why is your deliverability so good" is: we never let anyone shortcut to the established pool. The dramatic answer is: we built three pools instead of one.
If you want to read more about the send-pipeline side of this — the per-campaign-lead send function, the per-mailbox local-midnight quota resets — we'll have a piece up next week. Until then, the warmup pool is opt-in per mailbox in the dashboard; if you're on a trial, try it on one mailbox first.