High-volume bursts without reputation damage

Symptom

A launch announcement to 150,000 addresses fires in twenty minutes. Providers defer two-thirds, the queue clogs behind retries, transactional alerts queue behind the blast, and placement for the next two weekly sends suffers. Analytics show a “successful” 40% open rate on delivered mail — hiding that a third never delivered promptly and the domain reputation absorbed the deferral pattern as volatility.

Follow-up sends then underperform for weeks, and the team concludes “email is dying” rather than “our burst mechanics are hostile.”

Cause

Bursts violate the consistency signals providers score: sudden 50× volume multiples from a steady baseline resemble compromised-account behavior. Shared streams force transactional mail to inherit bulk deferrals. Unsegmented blasts mix highly engaged and long-dormant recipients, so early complaints and bounces from the cold tail poison the reputation of the entire burst. Unbounded worker concurrency ignores per-domain throttling signals until hard deferrals force backoff.

Queue absence is the architectural cause: direct API fan-out with no pacing, prioritization, or per-provider shaping leaves throttling to chance.

Fix

Stage the burst: engaged cohorts first (recent openers, active users), then progressively older segments over hours or days, with per-provider rate shaping and deferral-aware backoff. Isolate transactional onto separate streams and keys so announcement throttling never touches alerts. Cap initial burst concurrency conservatively, watch deferral and complaint telemetry in real time, and pause expansion on threshold breach rather than “pushing through.” Pre-warm capacity: raise volume gradually in the preceding week so the spike multiple shrinks.

Split sends across days for non-urgent announcements; urgency is rarer than marketing calendars claim, and a two-day paced rollout routinely outperforms a one-hour blast on delivered reach.

Prevention

Maintain standing burst runbooks with segment order, rate caps, rollback triggers, and on-call ownership. Load-test queue throughput at 3× expected bursts, monitor queue depth and per-provider deferral as first-class metrics, and sunset dormant cohorts before they can join the next blast. Post-mortem every burst against placement and reputation telemetry — not opens — and fold the numbers into the next schedule.

Worked example

A dev-tools launch targets 150,000 addresses. Instead of one blast, the team stages three waves over 36 hours: 25,000 recent actives first, then 60,000 ninety-day engaged, then the remainder excluding 12-month dormants. Per-provider pacing caps Gmail and Microsoft throughput separately, transactional stays on isolated streams, and a 5% deferral threshold gates each wave.

Wave one delivers 98.5% with near-zero complaints, establishing a clean reputation baseline the later waves inherit. Total delivered reach beats the previous one-shot launch by 22%, support sees zero late-OTP tickets, and the next weekly send performs normally. The launch retrospective notes the counterintuitive lesson: slower sending delivered more inboxes, because providers reward the order that queues impose.

• D-07: 429 strategy
• D-10: warmup schedule
• D-06: reputation split