Glossary · IP warm-up

IP warm-up

Building sending reputation by starting with small, engaged volumes and adding more over days; a cold dedicated IP loses to a shared one.

Definition

Warm-up builds reputation by sending small, engaged volumes first and increasing over days. A dedicated IP without warm-up is worse than a shared IP.

email docs/smtp

All glossary terms

A tall glass thermometer with a rising column, held in a steel clamp.

The short version

IP warm-up is the practice of starting a new sending IP at low volume with engaged recipients and increasing gradually while reputation builds.
It is not a checklist you complete once. Reputation is continuous, and a sudden volume spike on a warmed IP can undo weeks of work.
On a shared IP you inherit the provider pool reputation and cannot warm anything yourself. On a dedicated IP you own both the upside and the damage.
Send to your most engaged segment first. Receivers weigh engagement, complaints and unknown users, not just volume.
Authentication comes before warm-up: SPF, DKIM, DMARC, a matching reverse DNS and a consistent sending domain.
In India, send in the windows your audience actually reads, weight Gmail heavily, and remember that DLT governs SMS while email has its own rules.

What warm-up is, and what it is not

IP warm-up is a controlled increase in sending volume from a new IP address, designed to establish a reputation with mailbox providers before you ask them to handle your full list. The mechanism is behavioural: providers observe how much mail arrives from an address, how recipients respond, how many addresses are unknown, how many people complain, and whether the pattern looks like a legitimate business or a burst. A new IP has no history, so a large first send looks risky. A gradual send with good engagement looks normal.

The shape is simple to describe and harder to run. Start with a small volume to recipients who are likely to engage, hold the volume for a few days, increase in steps while the signals stay healthy, and stop increasing if complaints, bounces or deferrals rise. The ramp is not a fixed calendar, because the signals are the input. A team that follows a published fourteen-day table while its complaint rate triples is not warming up; it is automating the damage.

Figure 1. A loop, not a schedule. Volume only rises while the signals stay healthy.

What it is not: warm-up is not a substitute for authentication, a fix for a bad list, or a guarantee of inbox placement. It is not something you do to a shared IP, because you do not control the pool. It is not a one-time project with a completion date, because reputation is recalculated continuously. And it is not a reason to delay a launch by a month if you are sending a few thousand transactional messages, because at that volume the risk is small and the ceremony is disproportionate.

Shared versus dedicated: the calculus changes

On a shared IP, your mail leaves from an address used by many senders, and the provider manages its reputation as a pool. You inherit the pool history, good or bad, and you cannot warm anything yourself. The advantage is that a small sender gets the benefit of volume it could never generate alone, and the provider has an incentive to police abusive senders because they damage everyone. The disadvantage is that you do not control the neighbourhood, and a bad sender in the pool can affect your placement through no fault of your own.

On a dedicated IP, you own the reputation entirely. That is valuable when your volume is high and consistent, when your content is distinctive, and when you need predictable behaviour that a shared pool cannot give you. It is a liability when your volume is low or spiky, because an IP with long idle periods and sudden bursts looks suspicious, and because you now have to do the reputation work that the provider used to do. A dedicated IP without warm-up is usually worse than staying shared.

Figure 2. Shared pools hide the work and cap the control. Dedicated IPs give control and hand you the responsibility.

The threshold is a judgement, not a rule. Providers often suggest a dedicated IP when you send consistently at a volume in the hundreds of thousands per month, and many will advise against it below that. The better question is whether you can sustain a steady, engaged volume week after week. If the answer is no, a shared pool is the more honest choice, and the money is better spent on list quality. If the answer is yes, a dedicated IP plus a real ramp gives you control that is worth having.

What a receiver actually observes

A mailbox provider does not see your marketing plan. It sees connection-level facts and outcome-level facts. At the connection level it sees the IP, the HELO name, the reverse DNS, whether the IP appears on a public blocklist, whether TLS was used and whether the sending pattern is steady. At the message level it sees authentication results, the from domain, the links, the volume and the age of the domain. At the outcome level it sees bounces, complaints, replies and engagement, and those outcomes are what shape reputation over time.

Engagement is the signal teams underestimate. A provider that observes recipients opening, replying and moving mail out of spam learns that this sender is wanted. A provider that observes silence, deletions and complaints learns the opposite, even when the volume is modest. That is why warm-up sends to the engaged segment first: it is not a trick, it is the clearest available evidence that the mail is expected. Sending the first batch to a purchased list is the fastest way to teach a provider to distrust a new IP.

Figure 3. Reputation is built from outcomes, not declarations. The outcome layer is the one that compounds.

Consistency is the second signal teams underestimate. A sender that sends fifty thousand messages every Tuesday for a year looks predictable. The same volume sent in three bursts after two months of silence looks like a compromised machine or a spam run. If your calendar is inherently spiky, a shared IP absorbs that pattern better than a dedicated one, and a dedicated IP needs a plan to smooth it, such as spreading a large campaign across a day rather than firing it in ten minutes.

The shape of a ramp

A typical ramp runs over three to four weeks and moves in steps rather than a smooth curve. Day one might be a few hundred messages to the most engaged recipients, day two a few hundred more, and by the end of the first week a few thousand per day. The increase continues while complaint rate stays low, hard bounces stay low and deferrals do not spike. When a signal worsens, the correct move is to hold at the current volume for a few days, not to push through and hope.

The exact numbers depend on the list, the provider and the content, and no published table fits everyone. What transfers is the principle: the first sends should look like a small, wanted, well-authenticated stream. A steep ramp from zero to a hundred thousand in three days does not build reputation faster; it simply presents the provider with a burst before it has any evidence that the mail is wanted. The provider response may be throttling, deferral or silent filtering, and all three are harder to diagnose than a clean rejection.

Figure 4. An illustrative ramp shape, not a prescription. Steps hold or reverse when complaints or bounces rise.

Keep the content consistent during the ramp. Switching templates, link domains and subject styles every day makes it harder to attribute a change in signals, and it can look like several different senders sharing an IP. Warm up with the mail you actually intend to send, to the people who actually asked for it. A ramp built on a special quiet newsletter teaches the provider about that newsletter, not about your real programme.

Segmentation is the whole game

The first segment should be the people most likely to open, reply or click: recent buyers, recent openers, people who have written to support in the last quarter. Not because their attention is owed to you, but because their behaviour is the strongest available evidence that the mail is wanted. A hundred recipients who engage are worth more reputationally than ten thousand who ignore, and the provider can tell the difference over a few sends.

Build the tiers before you start. Tier one is thirty-day openers or clickers plus recent purchasers. Tier two is ninety-day engagement. Tier three is older engagement with a clear consent record. Tier four is the consented remainder who have never engaged, and it is usually the largest and least valuable group. Send tier one first, then two, then three, and treat four as optional. If a tier performs badly, do not promote the next tier on schedule; fix the reason first.

Figure 5. Illustrative tier sizes. The engaged minority carries the reputation; the never-engaged majority should not lead.

Suppress aggressively during the ramp. Hard bounces, complaints, unsubscribes and anyone who has not engaged in a year but has no recent consent should be out before the first send. A warm-up on a dirty list is a waste of the only clean reputation you will get, because the provider's first impression is formed from exactly the addresses you chose to include. This is the point where list hygiene and deliverability stop being separate projects.

The daily loop during a ramp

A ramp is run as a daily loop with a decision at the end of it. In the morning, check the previous day results: delivered, hard bounces, complaints, deferrals and reputation dashboards for the providers that expose them. Decide whether to hold, step up or step back. Send the day volume to the next slice of the current tier. Review again the following morning. The loop is deliberately boring, and boring is the point, because the alternative is discovering a problem after a hundred thousand messages have left.

Set explicit hold rules in advance so the decision is not a debate at nine in the morning. For example: hold if the complaint rate for the day exceeds the previous week by a wide margin, hold if hard bounces exceed your segment threshold, hold if deferrals rise sharply, step back if a provider dashboard moves from medium to low. The thresholds should be written down before the ramp starts, because the person watching the dashboard at the time will be tempted to explain away a bad number.

Figure 6. One decision per day, with the thresholds agreed before the ramp rather than argued during it.

Log the ramp in a table: date, volume, tier, complaint rate, bounce rate, deferrals, notable provider messages and the decision. That log is the only way to reconstruct what happened if placement falls three weeks later, and it is also the document that justifies the ramp to a manager who wonders why the whole list was not mailed on day one. The log is the deliverability equivalent of a deployment history.

Deferrals, throttling and when to stop

A 4xx reply during a ramp is not usually a failure. It is the receiving server asking you to slow down, and it is the clearest feedback a sender gets. Microsoft and Yahoo commonly defer rather than reject when volume arrives faster than expected. The correct response is to honour the retry, reduce the daily volume and lengthen the window, not to retry harder. Aggressive retrying of a deferred message is how a temporary throttle becomes a block.

Watch for the pattern rather than the single event. A handful of deferrals on a large send is normal. A rising share of deferred messages across several days, especially from one provider, means the ramp is ahead of the provider's comfort. Pause the increase, keep sending the same volume, and let the pattern settle. If a provider starts returning 5.7.x policy blocks, stop the ramp entirely and check authentication and blocklist status before sending anything else.

Figure 7. Three responses to three signals. Retrying harder is not one of them.

There is also a point to stop warming and accept a smaller programme. If the list is small, the content is niche and the engagement is modest, a shared IP will serve you better than a dedicated one. Warm-up is a cost, and it only pays when there is a steady, wanted volume to warm toward. Recognising that early is cheaper than spending six weeks building a reputation for a programme that sends twice a quarter.

Prerequisites before the first send

Authentication is not optional and it is not part of the ramp. Publish SPF for the sending infrastructure, sign with DKIM using a key you control, and publish a DMARC record that at least monitors and ideally enforces. Align the from domain with the DKIM signing domain, because alignment is what most large providers evaluate. Add a reverse DNS record that matches the HELO name, and make sure the IP is not on a public blocklist before you start, because inherited reputation from a previous tenant is a real and common problem.

Separate your streams. Transactional mail, such as password resets and receipts, should not share an IP or a subdomain with marketing, because a marketing complaint spike should not delay a receipt. Use subdomains for reputation isolation, and keep the link tracking domain consistent with the sending domain so that a link rewrite does not look like a redirect to an unknown host. If you use a new link domain, warm that too, because providers evaluate the domains in the message and not only the IP.

Figure 8. The checklist that comes before day one. A missing item here invalidates the entire ramp.

Finally, make sure the operational basics exist: a feedback loop enrolment where the provider offers one, a suppression process that runs within minutes, a complaint dashboard someone checks, and a rollback plan. Warming an IP without a complaint process is like driving without a brake. The first complaint is not the problem; the hundredth identical complaint that nobody suppressed is.

India: timing, providers and the rules that are not email

Indian inbox behaviour has a rhythm. Business mail peaks in the morning and early afternoon, consumer mail spreads into the evening, and a campaign sent at two in the morning may sit unread until the next day, which affects engagement signals during a ramp. Send in the window your audience actually reads, and keep that window stable during the warm-up so the signals are comparable. A ramp that changes send time every day is measuring the schedule, not the reputation.

Gmail carries the largest share of Indian consumer mail, so Gmail reputation is the one to watch most closely. Microsoft 365 matters for enterprise recipients, and its filtering is tenant-configurable, which means the same message can behave differently at two companies in the same city. Rediffmail, Yahoo India and Zoho appear in the long tail, and small business domains on shared hosting often have their own gateways. A ramp that looks healthy on Gmail can still be throttled at a corporate gateway, so watch provider-level signals rather than a single blended dashboard.

Figure 9. Five local factors that change a ramp. Provider mix and consent matter more than the calendar.

Two Indian rules get confused with email warm-up. DLT registration governs commercial SMS, not email, so a DLT template id has no bearing on whether your email is delivered. And the DPDP Act governs whether you may mail the person at all: consent has to be specific and withdrawable, and a suppression request has to be honoured across streams. Warm-up assumes a consented list. If the list is not consented, the ramp is simply a more organised way of sending mail people did not ask for, and the reputation outcome will reflect that.

Worked example: moving two lakh contacts to a dedicated IP

A Pune SaaS company sends about two lakh marketing emails a month on a shared pool and decides to move to a dedicated IP for control. The list has sixty thousand contacts with a ninety-day engagement record, forty thousand older engaged contacts, and a hundred thousand who signed up but have never opened. Authentication is already in place, including DMARC at quarantine, and the sending domain has two years of history, which helps because domain reputation travels with the message.

The team starts with the eight thousand most engaged contacts, at five hundred a day for three days, then two thousand a day for four days, watching complaints and deferrals. Week two moves to the ninety-day tier at five thousand a day, then eight thousand. Week three adds the older engaged tier and holds when Microsoft deferrals rise on the second day. Week four reaches twenty thousand a day across the consented engaged base and stops there. The never-engaged hundred thousand is left out and sent a separate re-permission note on the shared pool, because warming a new IP on silence would waste it.

Figure 10. A real ramp has holds. The value is in the log that shows why the volume stopped rising.

The log shows two holds, one caused by a deferral pattern at a corporate gateway and one caused by a complaint spike after a subject line change. Both were caught within a day because the daily loop existed. At the end of the month the dedicated IP carries the engaged programme, the shared pool still carries the re-permission and the transactional stream is on its own subdomain, and nobody has to explain a placement collapse to the leadership team. The cost of the caution was four weeks and a daily fifteen-minute review. The cost of skipping it is usually discovered later, in a quarter where the numbers quietly stop working.

Frequently asked questions

How long does IP warm-up take?

Usually three to four weeks for a mid-sized list, and longer for a large one or a domain with no history. The calendar is less important than the signals: volume rises only while complaints, hard bounces and deferrals stay controlled. A team that finishes in ten days because the table said so has not warmed anything, it has just sent faster. Plan for a month and treat an earlier finish as a bonus.

Do I need to warm up a shared IP?

No, because you do not control it. The provider manages the pool reputation and your mail inherits it. What you can control is your own behaviour: keep volume steady, suppress bounces and complaints quickly, and avoid sudden spikes that make your stream look anomalous inside the pool. If the pool is badly managed, the right move is to change provider rather than to attempt a warm-up you cannot perform.

When should I move from a shared to a dedicated IP?

When you send consistently at a volume that can sustain a reputation, often in the hundreds of thousands per month, and when you need control that a pool cannot give. If your volume is small, seasonal or spiky, a dedicated IP will be idle and then burst, which is worse than sharing. The decision is about consistency as much as size, and many teams move too early.

What volume should I start with?

A few hundred messages a day to your most engaged recipients, then increase in steps while the signals hold. The exact number depends on the list and the provider, and no published table is authoritative. The principle is that the first sends should look like a small, wanted, well-authenticated stream. Starting with a thousand engaged recipients is reasonable; starting with fifty thousand cold addresses is not a ramp.

Which segment should get the first send?

Recent openers, clickers and buyers, typically the last thirty days, plus anyone who has replied or contacted support recently. Their behaviour is the strongest evidence that your mail is wanted, and providers weigh engagement when building reputation. Never start with a purchased list or a never-engaged segment, because silence is the signal you are teaching the provider to associate with your new IP.

What should I do if complaints rise during the ramp?

Stop increasing, cut the volume and look at the segment and the content before continuing. A complaint spike usually means the audience was not as engaged as assumed or the message changed in a way that surprised people. Suppress the complainers, review the subject line and frequency, and only resume the increase when the rate returns to its previous level for several days.

Does warm-up guarantee inbox placement?

No. Warm-up builds a reputation that makes placement more likely, but content, links, authentication, list quality and recipient behaviour all still matter. A warmed IP sending a broken template with a mismatched link domain will still be filtered. Treat warm-up as one layer, and keep authentication, list hygiene and complaint handling as permanent operating disciplines rather than project tasks.

Is a new domain the same problem as a new IP?

Related but separate. Domain reputation travels with the domain, and a brand new domain has no history at all, which is often a bigger handicap than a new IP. Warm the domain alongside the IP, keep the visible from-name stable, publish authentication before the first send, and avoid changing the link domain mid-ramp, because providers evaluate the domains in the message as well as the address it came from.

How does DPDP affect a warm-up plan?

Warm-up assumes you are allowed to mail the people you are warming with. Under the DPDP Act that means specific, informed consent or another lawful basis, notice in clear language, and an easy withdrawal path, honoured across streams. A suppression request must stop the send even during a ramp. If the list does not meet that standard, fix the consent position first, because no amount of ramping makes unconsented mail welcome.

What does a warm-up cost?

Mostly attention. The sends themselves are billed like any other email, and on LoopNow overage is listed at ₹0.03 per delivered email with 18% GST added on the invoice. The real cost is the daily review, the segmentation work and the delayed campaign calendar. Set against a placement collapse that takes a quarter to recover from, four weeks of fifteen-minute reviews is usually the cheaper line item.