Deliverability Incident Response

Deliverability Incident Response

Things break. When they do, you need a playbook — not panic. This skill is the triage decision tree.

The five incident types

Decision tree: "my reply rate dropped"

Step 1: Quantify the drop

Copy code
Was this week's reply rate < 50% of last 4 weeks' average?
  Yes → real drop, continue triage
  No  → noise, wait another week

If drop is real:

Step 2: Is it campaign-specific or fleet-wide?

Copy code
Are ALL campaigns dropping, or just one?
  All → fleet-level issue (infrastructure or copy pattern)
  One → campaign-specific (targeting, copy, list)

Step 3: Fleet-wide drop — check these in order

1. Smartlead Smart Delivery spam test (/email-deliverability-audit → run-spam-test.ts)

   • If inbox placement <70% → real deliverability issue, skip to Step 4
   • If inbox placement >85% → deliverability is fine, look at copy/targeting instead

2. Check domain authentication (/email-deliverability-audit → check-domain-auth.ts)

   • Missing DKIM on any domain → fix immediately (see /zapmail-domain-setup-public for reconnect steps)
   • DMARC policy=reject with alignment failures → temporarily lower to quarantine

3. Check warmup status (/smartlead-inbox-manager → list-health.ts)

   • If multiple inboxes blocked in warmup → warmup network flagged you
   • If reputation dropped "good" → "fair" on many inboxes → slow down sending volume

4. Check bounce rate (/email-deliverability-audit → audit-performance.ts)

   • If bounce >3% → list quality degraded
   • If bounce <1% but reply rate low → emails landing in spam (Step 1 result)

Step 4: You have a spam-placement problem — what to do

Time-ordered actions:

1. Immediate (today): Pause the highest-volume campaign. Stop damage.
2. Day 1: Run spam placement test on 2-3 sender subsets (tag=active vs tag=new). Find which cohort is worst.
3. Day 1: Check the spam-filter-details report from Smart Delivery — specifically which filters are firing. Common: DKIM_INVALID, HTML_MESSAGE, LINK_REDIRECT. Each has a specific fix.
4. Day 2: Fix the identified issues:
   • DKIM_INVALID → verify DKIM records across all domains, re-publish if needed
   • HTML_MESSAGE → simplify email body (fewer fonts, no inline CSS, no tracking pixels)
   • LINK_REDIRECT → remove or reduce links in email 1 (ideally 0 links)
5. Day 3-14: Slow down send volume. Cut daily send per inbox by 50% for a week. Reputation rebuilds slowly.
6. Day 14: Re-test spam placement. If back above 85% → resume normal volume.

Decision tree: "bounce rate spiked"

Step 0: Do you have enough bounces to read?

50 sends before a bounce rate means anything. Below that, one dead address reads as 2% and two read as 4% — straight through the 3% line. 30 classified bounces attributed to that specific sending domain before the bounce composition means anything. A domain with no sampled bounces has a 0% sender-originated share purely by construction, which looks exactly like a rotten list — and that is how an unmeasured domain gets a confident "go revalidate your list" verdict.

Step 1: Classify the bounces by SMTP code, not by hard/soft

Hard-vs-soft is too coarse to act on. The enhanced SMTP status code in the bounce DSN tells you whose fault it is. Pull the DSN from the bounced lead's message history and take the first [45].\d+.\d+ code in the body — DSN templates repeat the code and often quote unrelated examples further down in the "more info for admins" boilerplate.

⛔ 5.7.133 is noise, full stop. It means "the group only accepts messages from inside its organization" — someone else's distribution-list setting. It is frequently the single most common code in a real sample, and bucketing it as a sender problem will manufacture "burned domain" cancel candidates out of nothing.

Then compute, per sending domain (not per client — a client-level split cannot tell you which domain to act on):

Copy code
sender_share = SENDER / (SENDER + LIST + BURNED)      # noise and throttle excluded
burned_share = BURNED / (SENDER + LIST + BURNED)

Copy code
burned_share > 20%  (≥30 classified bounces)
  → BURNED, at any bounce rate. Evaluate this FIRST, before the rate gate — a domain
    that is majority-blocklisted at a 2.5% bounce rate is still burned.

bounce_rate > 3% AND sender_share < 25%
  → list quality problem. Revalidate.

bounce_rate > 3% AND sender_share 25-50%
  → both diseases at once. Fix the auth gap FIRST, then revalidate. Do not go chase
    the list when 40% of the failures are yours.

bounce_rate > 3% AND sender_share > 50% AND auth incomplete
  → SENDER_AUTH. Fixable. Fix the record, re-check in 7 days.

bounce_rate > 3% AND sender_share > 50% AND auth clean
  → BURNED. Configuration is clean and we are still refused.

bounce_rate > 5% either way
  → stop the campaign immediately to prevent ISP suspension, then classify.

rising 4.x share, low permanent bounce
  → throttle. Cut volume. Do not cancel.

Step 1b: Sampling — do not enumerate every bounce

A bounced row does not tell you which domain sent it; attribution only exists after you pull the message history. So you cannot stratify by domain up front. Spend a global budget and enforce the floor afterwards: sample evenly across the in-window roster (roster[::step], never offset=0 only, which is time-biased), budget roughly 30 × (domains you need a verdict on) × 1.5, and mark any domain that ends up under 30 attributed bounces as insufficient sample rather than judging it.

⚠️ Bounced-lead rosters typically return a campaign's entire history with no date parameter. Filter client-side on the sent date, or bounces from a campaign that finished six months ago will drive this week's decision.

⚠️ Message-history fetches fail transiently. Retry 3+ times with backoff before recording "no DSN available" — on a real run, 61 of 64 apparently body-less leads returned a body on retry.

Step 2: If list problem

1. Export remaining leads from the campaign
2. Run through MillionVerifier (batch, ~$2 per 1,000)
3. Only reuse emails with status "ok"
4. Discard the rest

Step 3: If reputation problem

Slow way down. Cut daily volume 50% per inbox for 2 weeks. Run warmup more aggressively.

Decision tree: "domain blacklisted"

Step 1: Confirm the blacklist

Check:

• Spamhaus, Barracuda, SURBL via /email-deliverability-audit → spam-test report → blacklist detail
• MX Toolbox (https://mxtoolbox.com/blacklists.aspx) for a second opinion

Step 2: What tier of blacklist?

• Domain on Spamhaus DBL or SURBL → serious. You may need to replace the domain entirely.
• Sending IP on a DNSBL → usually the IP pool's fault (not yours, if on Zapmail shared IPs). Zapmail rotates IPs; wait 1-2 weeks.
• Minor list (e.g., Barracuda) → submit delisting request at their portal. Usually resolved in 3-7 days.

Step 3: Replace vs repair

• Domain <30 days old + blacklisted → replace. Not worth the cleanup effort. (Note this is the only case where a sub-30-day domain gets replaced: it is a confirmed blocklisting, not a reply-rate verdict. A young domain is never cancelled on performance — see /inbox-lifecycle-manager.)
• Domain >90 days old + blacklisted → try repair. Stop sending for 7 days, submit delisting requests, slowly resume.
• If you replace: archive the old domain, buy a new lookalike via /zapmail-domain-setup-public, warm it for 2 weeks before reusing.

Decision tree: "inbox blocked in warmup"

Step 1: Why is it blocked?

Smartlead's is_warmup_blocked: true flag usually means:

• Your warmup emails looked like spam to the warmup network
• Too many warmup peers marked them as spam
• The inbox type/provider is rate-limiting

Step 2: Triage

1. Check the blocked_reason field (if populated)
2. If warmup network issue → the inbox reputation may be damaged. Cost-benefit:
   • Young inbox (<30 days) → retire, provision new
   • Established inbox (>90 days, previously good reputation) → try disabling and re-enabling warmup with lower total_warmup_per_day (try 15 instead of 40)
3. If ISP rate-limit → wait 48h, re-enable warmup

Step 3: Retire workflow (via /smartlead-inbox-manager)

bashCopy code
# Tag as retired
npx tsx scripts/tag-inboxes.ts --ids=<id> --add-tag=retired --remove-tag=active

# Disable warmup
npx tsx scripts/set-warmup.ts --mode=disable --ids=<id>

# Replace — buy new domain, create new inbox
# See /zapmail-domain-setup-public

Decision tree: "Gmail marking as promotional"

Step 1: Test on a fresh Gmail account

Send the campaign's email 1 to a fresh @gmail.com account (yours, not in the campaign). Where does it land?

• Primary → you're fine
• Promotions → this is the issue
• Spam → you have a bigger deliverability problem (see above)

Step 2: Promotional-tab triggers (and fixes)

• Multiple links in email → drop to 0-1 links in email 1
• Images / inline images → remove them
• Heavy HTML styling → simplify, fewer tags
• Marketing-style phrases ("click here", "act now", "limited time") → remove
• Unsubscribe in header (List-Unsubscribe) — actually helps deliverability, but formatting must be correct
• Mass signature blocks with logos → simplify to text-only

Step 3: A/B test

Create two versions — your current and a stripped-down version. Send 50 leads each. Check which has better reply rate after 7 days.

The 72-hour "total failure" checklist

If NOTHING is working and you don't know why:

1. Pause all campaigns. Stop damage.
2. Audit: run /email-deliverability-audit full suite
3. Spam test: run Smart Delivery on 2 sender subsets
4. Check SPF/DKIM/DMARC on every domain — if ANY are missing, fix before resuming
5. Check Zapmail health dashboard — if their IPs are in trouble, everyone on their pool is too
6. Reduce volume 75% for the restart
7. Use a known-good copy — don't launch new copy during recovery
8. Watch reply rate daily for 7 days post-restart

When to call in experts

• You've been in spam for >2 weeks despite fixes → infrastructure level issue, consider migrating sending platforms
• Multiple domains permanently blacklisted → IP pool issue, switch providers
• Domain reputation never recovers after 4 weeks → retire domain, replace

What to do next

Re-run /email-deliverability-audit --days=7 in 7 days to confirm recovery. Domain/inbox reputation rebuilds slowly — don't re-audit before the 7-day window.

Meanwhile: continue the weekly rhythm via /cold-email-weekly-rhythm, which catches new issues as they emerge.

Or wait: if the incident required replacing domains/inboxes, wait 2 weeks for warmup on the replacements before expecting full recovery.

Related skills

• /email-deliverability-audit — the diagnostic suite you run first
• /inbox-lifecycle-manager — when the verdict is "this domain is burned", the cancel/replace/buy decision
• /smartlead-inbox-manager — tag, rotate, retire inboxes
• /zapmail-domain-setup-public — replace a burned domain
• /positive-reply-scoring — confirm recovery (reply rate back to baseline)

The 1% rule sanity check

After fixes, give it at least 200 sends at your normal volume before you judge whether the fix worked. Below 200, a reply rate cannot distinguish a broken domain from an unlucky week.

• Reply rate back over 1% — recovered. Resume normal rhythm.
• Still under 1% after 200+ sends — the fix did not take. Either start the playbook over, or accept the domain is burned and hand it to /inbox-lifecycle-manager for replacement. (It will still refuse to cancel a domain under 30 days old — young is not the same as burned.)

And give it time: reputation rebuilds slowly. Re-auditing after 2 days tells you nothing.