Bounce storms need a diagnostician.
A purpose-built model reads every bounce cluster in parallel — by default. Nothing to configure. Nothing missed.
Every way mail fails, catalogued
Checks run on every bounce cluster.
Reputation & throttling
9 modes — blocklist hits, throttling and sender-score decay.
Hard bounces & suppression
11 modes — dead addresses, typos and stale lists.
Auth & alignment
8 modes — SPF, DKIM and DMARC failures by hop.
Content filtering
7 modes — spam triggers and URL reputation flags.
Rate limits & deferrals
6 modes — 4xx greylisting and per-IP ceilings.
Vendor-side incidents
10+ modes — provider outages and upstream queues.
Why bounce storms linger
Cryptic SMTP replies
A 5.7.1 from one vendor means six different things — and the queue does not translate itself.
Vendor finger-pointing
SendGrid says content, the receiver says reputation. Meanwhile the queue keeps growing.
Fixes by folklore
The last person who understood the bounce codes left two reorganizations ago.
Forty-five minutes, or forty seconds
median manual triage per bounce storm
to an LLM diagnosis, every time
Collect, read, fix
Three steps between a bounce spike and a closed incident — the model does the reading, you keep the approve button.
Collect
Bounce webhooks from every vendor stream into clusters the moment they land.
Read
The model classifies root cause against vendor docs, recent changes and routing history.
Fix
It proposes a routing change with the evidence attached — you approve with one click.
See it on your own bounce logs
Thirty minutes, your delivery data, real answers.