What does marketing email at scale actually require?
A marketing send of 5 million messages over 4 hours is a different infrastructure problem than 100 transactional messages per minute over 24 hours. The peak send rate has to be sustained without throttling, the recipient mailbox providers have to be willing to accept that rate from your IPs, and your reputation has to survive the campaign without contaminating tomorrow's send. This is where IP warmup discipline, suppression-list hygiene, and per-campaign segmentation stop being theory and start being operational reality. We build infrastructure for senders in this category to handle 100K-50M messages per month, with the kind of rate-limit headroom and reputation observability that the volume requires.
Why do marketing senders move from SendGrid, Mailgun and SparkPost?
The three common reasons we hear: (1) per-IP cost — at the scale where you need 12+ dedicated IPs, SendGrid's €80/IP and Mailgun's €59/IP add €600-1000/month to a sub-€500 plan; (2) reputation transparency — competitors expose top-line delivery rates but not the per-IP, per-mailbox-provider, per-day reputation that lets you debug specific deliverability problems; (3) operational rigidity — the moment you need to do something non-standard (a unique IP rotation policy, a custom DMARC enforcement workflow, an integration with a non-mainstream marketing automation tool), the only path forward at the big competitors is "upgrade to Premier" with a 2-month sales cycle.
What does OS Domains change for marketing senders?
Two things are tuned differently for marketing traffic on our platform. First, the rate scheduler is engagement-aware: high-engagement segments (recipients who have opened in the last 14 days) get prioritized in the send queue ahead of low-engagement segments, which means your inbox-placement signal to Gmail stays high through the entire send. Second, you can run multiple IP pools per account — typically one for new-subscriber welcome flows, one for high-engagement broadcasts, one for re-engagement campaigns — with separate reputation tracked per pool. This lets you isolate the reputation impact of a difficult send (re-engagement to dormant lists) from the rest of your traffic.
Which metric do mailbox providers weight most for marketing?
Mailbox providers weight spam complaint rate above almost any other signal, and the thresholds are now hard lines rather than guidance. Gmail filters senders whose complaint rate crosses 0.30%, Yahoo states the same number, and Microsoft enforces in the same range; Google recommends staying under 0.10% and suggests 0.08% to keep clear margin. Two details make this sharper than most teams expect. The threshold applies per campaign, not as a monthly average, so a single poorly targeted send can trip filtering even when your program looks healthy on paper. And the denominator is inbox recipients, not total sent, which means once Gmail starts routing you to spam, the people who would have complained never see the message, and your reported complaint rate can look fine while placement quietly collapses. Crossing 0.30% also removes access to Google's delivery mitigation support until you hold under the line for seven consecutive days. We track complaint rate per campaign rather than per month, score a list against its engagement history before the send goes out, and can throttle or pause a send mid-flight when the early complaint signal from the first cohort crosses a ceiling, which is the only point at which the damage is still containable.
Authentication only gets you to the starting line; engagement decides the inbox.
Google, Yahoo, and Microsoft now reject unauthenticated bulk mail outright rather than delaying it, with the hard-rejection behavior escalated through late 2025. Aligned SPF, a valid DKIM signature, and a published DMARC policy are the price of admission, and a missing one means the message does not arrive at all. What authentication does not do is decide the inbox. Once you clear the authentication gate, placement is governed by engagement: recent opens and clicks, the domain-level reputation Gmail surfaces in Postmaster Tools, and the ratio of wanted to unwanted mail your recipients signal back. The practical consequence shapes how a program should send. Continuing to mail a dormant segment to "keep the list warm" does the opposite, because the weak engagement from that segment drags down the domain-level reputation that governs placement for the engaged recipients who actually convert. Our scheduler routes the highest-engagement recipients first to set the day's reputation signal before the volume opens up, and we treat dormant-segment sending as a reputation cost to be budgeted deliberately rather than a free way to pad reach.
Inbox placement runs around 83%, varies by geography and vertical, and the average hides the problem.
Average inbox placement across the major ESPs sits around 83% in 2026, which means close to one message in six never reaches the inbox, and the average is the least useful number in the report. Placement varies sharply by mailbox provider — Gmail trended down from roughly 90% to 87% across 2024 as engagement filtering tightened, and Microsoft fell further for many senders after its May 2025 changes — and it varies even more by geography, with Germany around 97% and India near 71% in the same datasets. A global campaign measured on a single blended number hides the market where half your mail is missing. We report placement per provider and per region, so a sender shipping into both Frankfurt and Mumbai sees the two as separate problems rather than one averaged-out figure. Gmail adds a layer the raw numbers miss: a message can be accepted and counted as delivered while landing in the Promotions tab rather than Primary, which is visible to the recipient in theory and ignored in practice. We surface the signals that distinguish Primary from Promotions placement, so the cadence and framing that push a sender into Promotions can be corrected before a season depends on it.
What bounce rate gets a marketing sender rejected?
A bounce rate that used to cost a little reputation now costs delivery: Gmail moved to permanent 5xx rejections for senders whose bounce rate runs above two percent, a change that landed in late 2025. The cause is almost always list sourcing rather than content, and the worst offenders are predictable — recruiting and HR, B2B sales and lead generation, and real estate, the verticals that lean on third-party databases, purchased lists, and portal exports where addresses were never confirmed. The defense is hygiene applied before the send, not after the damage. Our bounce classifier returns a sub-code for every failure — hard bounce, spam-trap hit, content rejection, mailbox full — with the original SMTP response and a recommended action, which feeds suppression precisely rather than bluntly. We gate a validation pass on new or long-dormant segments before they go into a broadcast, because the cheapest spam-trap hit is the one that never gets sent, and a single programmatic send to an unvalidated purchased list is the fastest way to turn a healthy domain into a blocked one.
DMARC at p=reject is the minority position, and it is the BIMI gate.
DMARC presence has climbed past three-quarters of large-domain senders, but only about a third of those records sit at p=reject, the enforcement level that actually matters. The gap exists because reject feels dangerous: a misconfigured enforcement policy can block a legitimate sending source the business forgot it had, so teams publish p=none, collect the reports, and never move. That hesitation has a cost beyond authentication, because reliable Gmail placement leans on enforcement and a BIMI logo is not available below a policy of quarantine moving to reject. We run the migration the careful way through Managed DMARC: publish p=none, parse the rua reports to inventory every legitimate source, bring each one into SPF and DKIM alignment, step to p=quarantine with a percentage rollout, and reach p=reject only once the report data shows no legitimate mail would be caught. Done in that order, enforcement stops being a gamble, and it opens the door to the brand-indicator logo that lifts recognition in a crowded inbox.
The per-IP economics that flip the build-versus-managed decision at scale.
The migration economics are concrete once a sender needs real IP capacity. At the scale where you run a dozen or more dedicated IPs, the per-IP line items at the large platforms add up — on the order of seventy to eighty euros each at SendGrid and around sixty at Mailgun — which can bolt six hundred to a thousand euros a month onto a plan that was meant to cost less than five hundred. We price IPs without that markup, and we will talk you out of buying more than you can warm, because idle IPs do not hold reputation and an over-provisioned fleet sends worse than a right-sized one. The working guideline is one IP per hundred to two hundred thousand monthly messages, with a floor of three even at low volume to allow rotation. The honest counterpoint sits at the other end: below roughly half a million messages a month, a well-run shared pool at a mainstream ESP is usually the better economics, and we say so rather than selling dedicated infrastructure to a sender who does not yet need it. The build-versus-managed line is about whether the volume and the reputation control justify operating the fleet, not about the per-message rate.
Re-engagement is a controlled burn, not a list-wide blast.
A dormant segment is a reputation liability, and the instinct to win it back with one big "we miss you" blast to everyone inactive for a year is exactly the move that does the damage. That blast floods the program with low-engagement mail in a single send, and the weak signal drags down the domain-level reputation that governs placement for the people who still open. Re-engagement done well is a controlled burn. We isolate it on its own IP pool, separate from the broadcast reputation, so a predictably rough send cannot pull down the rest of the traffic. The daily volume is capped rather than dumped, the content asks plainly whether the recipient still wants to hear from you, and anyone who does not engage inside the window is suppressed rather than recycled into the next campaign. The unsubscribe rate roughly doubled across 2025, partly because one-click made leaving easy, and that is healthier than it sounds: a clean unsubscribe protects your complaint rate, while a recipient who cannot find the exit reaches for the spam button instead. We treat a rising unsubscribe rate on a re-engagement send as the system working, not failing. The list that shrinks on purpose, holding only addresses that still engage, outperforms the list that pads its size with recipients who stopped opening months ago, because placement is decided by the engagement ratio rather than the raw count.