This is a case study from a real B2B team we worked with between January and March 2025. Names and industry removed; numbers intact. They came to us with a cold outbound operation that was technically running but quietly broken: their bounce rate had climbed from around 4% to 12% over six months, and nobody could tell us why. In eight weeks we took it back under 2% and their reply rate nearly doubled as a consequence. Here is what actually happened.
The starting state (Week 0)
- 12.1% bounce rate across 14 sending mailboxes
- 4 sending domains, all authenticated on paper
- Apollo data pulled in bulk, no real-time verification
- Reply rate 1.8% (down from 4.2% six months prior)
- 3 domains on public blocklists (team didn't know)
The leadership was about to fire the SDR team. We asked for 8 weeks first.
Change 1: Real-time verification before every send (Week 1)
The Apollo data was 6-18 months old and had drifted. Every export was being sent without a freshness check. We plugged in ZeroBounce as a pre-send gate, rejecting any address that came back as invalid or "catch-all unknown". Bounce rate after week 1: 8.4%. Still high, but immediately better.
Change 2: Suppressing role addresses (Week 1)
About 9% of the list was role-based (info@, support@, sales@, contact@). These addresses bounce less but get flagged as spam more. We suppressed the entire category. Bounce stayed around 8%, but complaint rate started dropping immediately.
Change 3: Rest and rewarm for 3 damaged domains (Week 2)
Three of the four sending domains had burned reputation. We paused all cold sending on those three for 21 days and put them back into continuous warmup. The fourth domain kept going at reduced volume. Total sending capacity dropped 70% for three weeks, and nobody was happy about it. But it was the only way forward.
Change 4: Tighter targeting criteria (Week 3)
The team was exporting any prospect matching broad filters (title contains "director", company size 50-500). We tightened: specific titles only, verified last job change within 24 months, company had funding or hiring signals in last 6 months. The list shrank from 12,000 to 1,400. The quality jumped accordingly.
Change 5: Manual spot-checks on AI-written openers (Week 4)
The team was using GPT-4 to write openers and sending them without review. About 15% of the output was wrong: misread signals, hallucinated facts, awkward phrasing. We added a 30-second review step per email. Labour cost: 1 hour per day across the team. Quality impact: immediate and obvious.
Change 6: One-click unsubscribe + shared suppression (Week 5)
The suppression list was per-mailbox, not global. Prospects who unsubscribed from one mailbox were still receiving from others. We built a shared suppression table synced every hour. Complaint rate dropped another 30% within two weeks.
The results (Week 8)
- Bounce rate: 1.3%
- Reply rate: 3.5% (up from 1.8%)
- Booked meetings: +74% vs baseline, at similar send volume
- Postmaster spam rate: under 0.1% on all domains
- All three previously damaged domains back to healthy reputation
What we did not do
We didn't add new tools. We didn't rewrite the sequences. We didn't bring in a consultant to "rethink the strategy". The entire intervention was infrastructure hygiene and targeting discipline. That surprised the leadership, and it probably shouldn't have.
"The issue was never the reps. The issue was that nobody owned deliverability as a daily operational task."
The lesson
Most cold email problems look like content problems from the outside and turn out to be infrastructure problems on the inside. Before you rewrite your openers, before you change your sequences, before you retrain your team, check the boring technical layer. It is almost always the cheapest fix and almost always the one that matters most.