I broke our email deliverability. Rebuilding it took two months.
When it worked
Outbound was small and deliberate. I sourced contacts through Apollo, verified every address with ZeroBounce, and built campaigns for a hundred to three hundred people at a time. Each one was written for a narrow cohort rather than a broad list.
A representative campaign from February: 307 emails, 98.7% delivered, 1.3% bounce, 27.7% opens. No unsubscribes, no spam complaints. Nothing spectacular, but healthy, and it produced real conversations.
Then I was asked to scale it
Growth targets went up and outbound had to go up with them. Campaigns went from a few hundred contacts to a thousand or fifteen hundred. I did not have enough verified contacts to fill that, so I bought a list. I also spread the sending across three or four addresses to carry the extra volume — all of them on our primary company domain.
Both of those decisions were wrong, and the second one is why it took two months to undo rather than two days.
The numbers moved the wrong way, fast
| February Verified list | May Purchased list | |
|---|---|---|
| Bounce rate | 1.3% | 6.9% |
| Unique opens | 27.7% | 8.6% |
| Clicks | 0.7% | 0% |
| Spam complaints | 0 | 0 |
Two campaigns from the same sending infrastructure, three months apart. The May figures come from a deliberately small test batch — once I suspected something was wrong, sending to a larger list would only have deepened the damage.
A bounce rate above 5% is the threshold at which mailbox providers start treating you as a sender who does not know who they are writing to. I had gone past it.
The clue that cracked it
What made me look properly was a contradiction I could not explain away.
The same people were receiving our newsletter through a different platform, sent from that platform’s infrastructure rather than ours. There, opens ran between 19 and 32 percent across editions. On our own domain, the same audience was opening at under 15.
Same humans, same subject matter, wildly different results. That ruled out the audience and the content, and pointed at the only variable left: who was doing the sending.
It is also the one open-rate comparison in this story I would defend. Both platforms were running in the same period, to the same people, counting opens the same imperfect way. Whatever inflates one inflates the other, so the gap between them is real even if neither number is exact.
So I ran inbox placement and deliverability tests. The result was unambiguous.
| Where our mail landed | ||
|---|---|---|
| Spam folder | 75% | Gmail, Yahoo, AOL |
| Inbox | 25% | Outlook only |
Outlook still trusted us. Gmail had stopped completely — and most of the people we needed to reach were on Gmail.
What I had actually done
Only then did I go back and audit the purchased list properly. It was full of invalid addresses, catch-all domains that accept anything and confirm nothing, and generic free-mail accounts. Exactly the profile that generates bounces.
But the bounces were the symptom. Laying our whole audience out by source showed the real problem:
| Existing customers | 185 |
| Known prospects | 220 |
| Partner-sourced contacts | 181 |
| Two internal product lists | 744 |
| Purchased list | ~3,500 |
1,330 contacts we had earned. Roughly 3,500 we had bought.
Gmail was no longer learning who we were from our best contacts. It was learning from our weakest ones.
That is the whole failure in one sentence. Mailbox providers judge a sender on how recipients behave. By letting a bought list become more than seventy percent of the audience, I had handed Gmail a sample of people who mostly did not exist, did not care, or could not be verified — and asked it to form an opinion of us from that.
And because I had sent it all from our primary domain rather than a separate sending domain, the reputation I damaged was the one the entire company relied on.
The conversation I had to have
I paused every campaign that day. Then I went to leadership and said, roughly: I have made a blunder, it will only get worse if I do not get two months to fix it, lead generation is going to be slow for the next quarter, and it should improve from there.
They agreed. I think the reason they agreed is that the ask came with a diagnosis, a plan and a date rather than an apology.
Telling them a quarter early cost me an uncomfortable afternoon. Telling them a quarter late would have cost considerably more.
The rebuild
Verify everything. I put the full list of roughly ten thousand contacts through MillionVerifier and enriched what survived, so that every remaining record was both real and correctly attributed.
Stop sending from the primary domain. We already owned a couple of domains that had never sent a single email, so we used those and built fresh mailboxes on them. Cold outreach now runs on infrastructure that is separate from the company’s real correspondence. If I damage it again, I damage something replaceable.
Warm up slowly, and let the tool enforce the rules. We moved sending to Smartlead, which handles warmup and will only send to addresses that pass its own verification — not foolproof, but a second gate between me and a bad record.
Then two months of sustained warmup before returning to real volume. There is no way to shorten this part. Reputation is rebuilt at the pace the mailbox providers decide.
What the fix cost
| Verifying ~10,000 contacts (MillionVerifier) | ~₹4,000 |
| Enriching the surviving records | ~₹5,000 |
| Domains (two already owned, unused) | ~₹2,500 |
| Six mailboxes, monthly | <₹1,000 |
| Smartlead, monthly — warmup and sending | ~₹4,000 |
Around ₹11,500 one-time and roughly ₹5,000 a month.
That is the part worth sitting with. The repair cost about as much as one good dinner out. The shortcut that made it necessary cost a quarter of lead generation and two months of a marketing team’s output.
Where it stands
| At the low point | After the rebuild | |
|---|---|---|
| Bounce rate | 6.9% | 0.45–1.65% |
| Inbox placement test | 75% spam | Passing |
Open rates have also risen sharply, but I am not going to lead with that number. The new platform reports opens differently from the old one, warmup traffic opens its own mail, and image-based open tracking has been unreliable across the industry since mail clients began pre-fetching images automatically. Bounce rate and placement tests are measured the same way in both periods, so those are the numbers I trust.
What I can say honestly is that our email reaches inboxes again rather than spam folders. Whether that converts back into pipeline at the rate it did before is too early to tell, and I would rather say so than claim a recovery I cannot yet evidence.
What I would do differently
Never send cold outreach from the primary domain. This is the one that actually mattered. Separate sending domains are cheap, take an afternoon, and mean a mistake stays contained. Had I done this first, the same error would have been a week’s inconvenience.
Treat volume as the output, not the input. I was asked to scale and I reached for more names, because that is the fastest-looking route. The number of people you can credibly contact is a constraint, not a dial. When the verified list runs out, the honest answer is that outbound has hit its ceiling and growth needs another channel.
Watch bounce rate as a leading indicator. Opens fall for a dozen reasons. Bounces have one meaning: you are writing to people who are not there. It was visible in the data weeks before I acted on it.
Escalate on the diagnosis, not the outcome. The only part of this I would keep unchanged is going to leadership early with a timeline attached. That bought the two months. Waiting for the numbers to explain themselves would not have.