Every sending program eventually adds a new IP — because volume grew past what the existing pool can carry, because a new product line needs reputation isolation, because a dedicated relationship with an ESP reached the threshold where the shared pool stopped making sense, or because an acquisition brought sending infrastructure that now needs to be rebuilt from scratch. The twenty-one days that follow that decision determine the IP's reputation for years. Skip the warmup and the IP arrives at full volume carrying the specific pattern — sudden appearance, sustained throughput, indiscriminate recipient mix — that every mailbox provider has learned to read as "compromised account or spam campaign." Execute the warmup well and the same volume, delivered on the same IP, looks like a legitimate sender gradually coming online. This article is the design guide: what the curve actually looks like, which recipient cohorts to send to in which order, what to watch for during the twenty-one days, and what to do when the warmup runs into trouble halfway through.
Key takeaways
- IP warmup is the process of establishing reputation for a new sending IP by starting at low volume and growing it over three to four weeks. The goal is to produce a pattern of sending behavior that providers recognize as legitimate rather than as the fingerprint of a compromised or abusive account.
- The curve matters. Start with a few thousand messages on day one, roughly double every two to three days during the first week, then flatten the ramp as volume approaches steady-state. Total duration: typically 21 days for a target of 500k/day; longer for higher targets.
- Recipient selection during warmup is the lever most schedules ignore. Send to your most-engaged recipients first — openers and clickers from the past 30 days — and widen the cohort gradually. A warmup to an indiscriminate mix fails predictably; a warmup to engaged recipients succeeds predictably.
- Monitor five metrics daily during the ramp: deferral rate (4xx), hard bounce rate, complaint rate, authentication pass rate, and inbox placement at a seed panel. Set thresholds in advance that will cause the warmup to slow or halt — a Gmail deferral rate above 10% for two consecutive days, for example, should stop the ramp rather than push through it.
- Below roughly 500,000 messages per month of sustained volume, a dedicated IP is usually the wrong choice — there isn't enough signal density to build and maintain stable reputation. A shared IP pool at that volume produces better deliverability than a new dedicated IP that's always half-warm.
Why a warmup schedule needs to be designed, not improvised
Most delivery problems on new sending infrastructure trace back to the same root cause: someone turned on the IP and started sending at full volume. Providers interpret a sudden appearance of a sending IP at high throughput as a signal of compromise, and they respond with the caution that signal warrants. The result is predictable — spam folder placement, delivery deferrals, and a frustrated team that assumed the new hardware would just work.
The fix is not mysterious. Every major inbox provider documents, in broad strokes, that sending IPs develop reputation gradually and that new IPs should build volume over several weeks. The trick is that "gradually" is doing a lot of work in those recommendations. Senders who interpret it as "add a few thousand messages per day" get results; senders who interpret it as "double the volume every day" get penalized; senders who skip warmup entirely get blocked.
A designed warmup schedule is an artefact — an actual document with daily volume ceilings, recipient selection criteria and monitoring checkpoints. It is not a philosophy. This article walks through how to build one that holds up in production across the major providers in 2013. The broader context of why a team ends up at the warmup stage in the first place — the transition from shared hosting or ESP infrastructure to dedicated sending — is covered in scaling outbound email beyond shared hosting. If that transition is still ahead of you, read that first; this article assumes you've already decided that a dedicated IP is the right answer.
The shape of a warmup curve that receivers actually recognize
There are three qualities every workable warmup curve shares. First, it starts small — a few hundred to a few thousand messages on day one, not tens of thousands. Second, it grows meaningfully but not recklessly — roughly doubling every two to three days during the early phase, then slowing as volume approaches steady-state. Third, it reaches full production volume in two to four weeks, depending on the eventual target. Much faster reads as aggressive; much slower fails to build the sending history the IP needs.
A reasonable 21-day curve for an IP that will eventually carry 500,000 messages per day looks something like this: days 1 to 3 send 2,000 to 5,000 per day; days 4 to 7 move to 10,000 to 25,000; days 8 to 14 climb from 50,000 to 200,000; days 15 to 21 ramp from 300,000 to the full 500,000 target. The exact numbers matter less than the shape, which is an exponential climb that flattens as it approaches production volume.
Warmup curves also need to be provider-specific in spirit. Gmail, Hotmail, Yahoo and AOL all watch different things and tolerate different ramps. Gmail in particular weighs engagement heavily — a warmup that sends to engaged recipients first at Gmail recovers reputation faster than the same curve sent to an indiscriminate mix.
Provider-by-provider tolerance during warmup
The major providers each impose per-IP volume limits during the warmup period that aren't formally published but are well-known to experienced deliverability operators. The table below reflects the operational consensus as of October 2013, calibrated against the M3AAWG guidance and day-to-day observation. Hitting these ceilings produces deferrals (4xx responses); sustained pushing through the ceiling produces blocking.
| Provider | Day 1-3 | Day 4-7 | Day 8-14 | Day 15+ | Key signal |
|---|---|---|---|---|---|
| Gmail | 5,000 | 25,000 | 200,000 | 500,000+ | Engagement-weighted; tolerant of slower ramps with engaged recipients |
| Yahoo | 2,000 | 10,000 | 100,000 | 500,000+ | Complaint-rate sensitive; 421 deferrals common during ramp |
| Hotmail / Outlook.com | 5,000 | 25,000 | 200,000 | 500,000+ | SNDS filter result (red/yellow/green) is the truest feedback signal |
| AOL | 2,000 | 10,000 | 100,000 | 500,000+ | Strictest on complaint rate; target under 0.08% throughout |
| Comcast | 1,000 | 5,000 | 50,000 | 200,000+ | Smaller volumes; strict on bounce-rate signals |
These are not hard ceilings; they are the approximate points at which each provider's internal throttles begin to respond. A warmup that stays comfortably under these limits — ideally at 50-70% of the ceiling during the ramp phase — produces the signal of a careful, legitimate sender. A warmup that pushes against the ceiling produces the signal of urgency that providers treat with suspicion.
Recipient selection: the lever most schedules ignore
Volume is only half the warmup story. Who you send to during the warmup matters as much as how many messages you send. The rule of thumb is simple: the most engaged recipients first, the least engaged never during warmup.
Engaged recipients generate the signals that build positive reputation quickly — opens, clicks, replies, moves from spam to inbox. Inactive recipients generate the signals that suppress a new IP — deletion without reading, complaints, deferrals from addresses that no longer exist. A warmup schedule that mixes both is fighting itself. A warmup schedule that front-loads the engaged cohort and leaves the rest for after full warmup builds reputation several times faster.
In practice this means ranking the list before the warmup starts. Recipients who opened at least one message in the past 30 days go into the day-1 cohort. Recipients who opened in the past 90 days go into the week-2 cohort. Recipients who have not engaged in 90 days go into the post-warmup backlog, where they can be re-engaged carefully once the IP has stable reputation. Recipients who have not engaged in 180 days should probably not be mailed at all until they opt back in.
Cohort selection, concretely
| Phase | Days | Cohort | Why this works |
|---|---|---|---|
| Initiation | 1-7 | Opened or clicked in past 30 days | Highest likelihood of positive engagement (opens, clicks, moves from spam); builds reputation fastest |
| Widening | 8-14 | Opened in past 60 days | Still meaningfully engaged; adds volume without diluting signal quality too aggressively |
| Approach steady state | 15-18 | Opened in past 90 days | Full "active" definition; cohort is broad but still within the engagement threshold inbox providers reward |
| Steady state | 19+ | Full active list (whatever your policy defines as "active") | IP reputation now established; sending to the full addressable list reflects true operating volume |
| Post-warmup re-engagement | 30+ | Cold subscribers (90-180 days inactive) | Only after IP is stable; separate re-engagement program, not full-volume sending |
| Never during warmup | — | Inactive beyond 180 days, purchased lists, unverified addresses | Produces exactly the signals (bounces, complaints, non-engagement) that destroy warmup reputation |
Monitoring during warmup: the metrics that actually move
Track five metrics daily during warmup, broken out by receiving domain:
- Deferral rate. 4xx responses from the receiver mean "not now". A rising deferral rate is the earliest sign that the provider is skeptical.
- Hard bounce rate. 5xx "user unknown" responses. Target under 2%. Over 5% is a warmup-halting emergency.
- Complaint rate. Recipients marking your mail as spam. Target under 0.1%. Over 0.3% damages reputation substantially.
- Authentication pass rate. DKIM and SPF should pass for essentially 100% of sent messages. Anything below tells you the signing or DNS configuration has a problem.
- Inbox placement at a seed set. A small panel of test addresses at Gmail, Hotmail, Yahoo and AOL, checked after each send, tells you where mail is actually landing — primary, tabs, bulk or missing.
Two things to do with these numbers. First, set thresholds in advance that will cause the warmup to slow down or halt. A deferral rate above 10% at Gmail for two consecutive days, for example, is a signal to hold volume flat rather than double it. Second, keep a daily log — timestamp, volume, deferral rate, bounce rate, complaint rate, inbox placement — so that when something drifts you can see when it started. The broader model of how providers actually interpret these signals to build their view of a sender — and why complaint rate in particular is weighted so heavily — is covered in sender reputation fundamentals.
A worked schedule for a 21-day warmup
Day Target Volume Recipient Cohort
--- ------------- -----------------------------
1 2,000 Engaged past 30 days
2 3,000 Engaged past 30 days
3 5,000 Engaged past 30 days
4 10,000 Engaged past 30 days
5 15,000 Engaged past 30 days
6 25,000 Engaged past 30 days + top 30-60 day
7 40,000 Engaged past 60 days
8 60,000 Engaged past 60 days
9 90,000 Engaged past 60 days
10 125,000 Engaged past 60 days
11 160,000 Engaged past 60 days
12 200,000 Engaged past 90 days
13 240,000 Engaged past 90 days
14 280,000 Engaged past 90 days
15 320,000 Engaged past 90 days
16 360,000 Full active list
17 400,000 Full active list
18 440,000 Full active list
19 475,000 Full active list
20 490,000 Full active list
21 500,000 Full active list, steady stateThis is a template, not a prescription. Two things to adapt. First, the ceiling — 500,000 is illustrative; adjust to your real steady-state target. Second, the cohort widening — if your list does not have a deep tail of engaged recipients, slow the ramp so the cohort breadth can keep up.
One discipline worth naming: send at roughly the same time every day during warmup. Providers watch for consistency. A schedule that sends at 10am Monday, 6am Tuesday and 3pm Wednesday looks less established than one that sends at 10am every day. Once the IP is warm, send timing can flex.
Domain warmup runs parallel to IP warmup
The IP is the reputation surface receivers watch most closely in 2013, but it's not the only one. Every sending domain — the one in the From header, the one in the DKIM d= tag, the one in the Return-Path — carries its own reputation signal. At Gmail in particular, domain reputation has been steadily rising in weight over the past two years, and a new sending domain goes through its own warmup regardless of what IP is carrying the traffic.
In practice this means that if the warmup is for a new IP attached to an existing, reputable sending domain (for example, you've been sending from mail.example.com for years and are just adding a new IP to the pool), the domain reputation carries the IP through the early days more gracefully than it otherwise would. If the warmup is for a new IP attached to a new sending domain, you're warming both simultaneously — and the pace needs to be slower, because two new reputation surfaces at once is a harder posture to defend.
The practical implication: whenever possible, stand up a new IP under an established sending domain rather than changing both at once. If that's not possible — a fresh product launch, an acquired company, a rebranding — extend the warmup to four to six weeks rather than three, and be more conservative on the ramp during weeks two and three when both reputations are still thin.
Dual-warmup calendar adjustments
| Configuration | Target duration | Rationale |
|---|---|---|
| New IP, established domain | 14-21 days | Domain reputation absorbs some of the "new sender" signal; IP-specific warmup is the sole focus |
| New IP, new subdomain under established parent | 21-28 days | Subdomain inherits partial signal from parent domain; slightly slower ramp |
| New IP, entirely new sending domain | 28-42 days | Both reputations thin; slower ramp is essential; expect bulking at Gmail until week three |
| Previously-used IP with mixed history | 21-28 days, conservative ramp | Existing reputation can be positive, negative, or silent; start slow until you read the signal |
| Multiple IPs warming simultaneously | Each IP's own timeline, in parallel | Splitting warmup traffic across IPs does not speed any individual IP's warmup |
One operational detail worth pinning down: whichever domain carries the From header during warmup should be the one you plan to send from steadily afterward. Warming up on a "temporary" sending domain and then switching to the production domain at full volume on day 22 voids the warmup — the production domain has no reputation and the traffic pattern looks like a compromise. Plan the domain you'll be using in month twelve and use it from day one.
When the warmup gets into trouble, how to respond
Two failure patterns are common. The first is a complaint rate spike that appears around day 5 to day 7, when the cohort widens. The cause is usually that the list contains a cohort of once-engaged recipients who have forgotten about the sender. The response is to tighten the cohort definition and slow the curve for three to five days while the provider recalibrates.
The second is provider-specific deferrals — Hotmail starts batching mail to bulk, or Gmail starts deferring under SMTP 421 errors — while other receivers stay clean. The cause is usually that the provider's internal throttles have kicked in. The response is to hold volume to that provider flat, continue growing to others, and let the flattening send a signal that the sender is responsive to feedback.
Avoid the temptation to fight the provider. A warmup that pushes through deferrals by sending more aggressively turns a transient problem into a blocked IP. The correct posture is always to match the signal the provider is sending: if they slow you down, you slow down.
Warmup troubleshooting reference
| Symptom | Probable cause | Response |
|---|---|---|
| Deferral rate at Gmail climbing above 10% on day 7 | Cohort widening to 60-day engagement introduced recipients who aren't opening | Hold volume flat for 3-5 days; re-check engagement filter; resume at a tighter cohort |
| Hotmail mail quietly routing to bulk folder | SNDS showing yellow; reputation below the inbox threshold | Check SNDS filter result daily; confirm JMRP complaints are being processed; hold volume |
| AOL complaint rate above 0.15% | Subject line or content producing disproportionate complaints from AOL users | Halt warmup to AOL; review most-recent campaign content; resume after content change |
| Yahoo 421 deferrals across all IP | Connection-rate or per-hour throttle hit | Reduce concurrent connections; spread send across longer window; respect the 421 backoff |
| Bounce rate above 5% | List contains invalid addresses (common after long gap without list hygiene) | Halt warmup immediately; run list-validation against a third-party service; restart with verified list |
| DKIM failures on a subset of messages | Key mismatch, DNS propagation issue, or message-modification between signing and send | Check dig response for DKIM selector from multiple resolvers; confirm no post-signing modification |
| Seed test shows Gmail spam folder despite clean metrics | Domain reputation lagging IP reputation; content-based signals | Continue warmup; review subject lines; verify SPF/DKIM alignment to From domain |
| Inbox placement rate varies wildly day to day | Seed panel too small; natural variance masking real signal | Expand seed panel to at least 15-20 addresses per provider; smooth over 3-day moving average |
Frequently asked questions
How many IPs should I warm simultaneously? As many as you need for steady-state, but treat each one's warmup as independent. Splitting traffic across three warming IPs does not speed up warmup — it just means each IP warms at a third of the pace.
Can I pre-warm an IP with a seed campaign before using it for production? In principle yes, but the campaign has to be genuinely engaging to the recipients, or the metrics look worse than starting fresh. In practice, sending a real but small stream of important, well-received mail (transactional is ideal) is better than synthetic warmup.
What if the warmup completes and Gmail placement is still poor? That usually means the content itself, the domain reputation or the list quality is the remaining issue — not the IP. At that point the warmup has done its job and the investigation needs to shift to the other inputs.
Do I need to warm up IPs that have been used before? Yes, but differently. A previously-used IP may have residual reputation, positive or negative. Start with a conservative ramp while you learn what the IP's history tells the receivers. If metrics look clean, accelerate. If not, you may be inheriting problems that need managed deliverability attention.
How do I know the warmup is actually working? Three signals indicate a warmup is on track: deferral rates trending down as volume grows, inbox placement at the seed panel holding steady or improving, and no sudden complaint-rate jumps when cohorts widen. If all three are true at day 14, the warmup is working. If any one of them is failing, the warmup needs adjustment before the ramp continues.
Can the warmup run during a holiday or high-volume period? In principle yes, but the signal is harder to read because engagement patterns shift. A warmup that starts the week before Black Friday will run into recipient behavior that doesn't reflect normal operating conditions, and the metrics during those days are noise. If possible, schedule warmup during regular business periods; if not, extend the duration by a week to give the metrics time to settle back into representative patterns once the seasonal anomaly passes.
What about warmup during a ramp-down at a previous provider? Running new-IP warmup in parallel with wind-down sending from the old infrastructure is the standard migration pattern. The key is that recipient-level volume should remain roughly constant — if a subscriber was getting two messages a week from the old IP, they should get two messages a week from the combination of old and new during the transition. Doubling up because the old IP is still sending produces complaints that the warmup doesn't deserve.
Closing thought
Warmup discipline is unglamorous. It costs three to four weeks of patience at the start of every new sending deployment. The payoff is a sending IP that arrives at steady-state with a clean reputation trail, recognized by every major provider, and resilient enough to withstand the small missteps that inevitably happen in the months that follow. Senders who skip warmup pay for that missing history for the life of the IP.
Continue your evaluation
If this article maps to the sending layer you are building or operating, the pages below go deeper into the commercial and operational side of the same territory.