Cold Email Bounce Rate

Most pages answering this question quote a benchmark from somebody else's study. This one reports our own production numbers, because we run cold outreach as a product and the database is right there. Every figure below comes from 11,477 send attempts between 7 March and 25 August 2026, counted once each after deduplicating the two rows our system writes per send.

The more useful answer than any single figure is the shape of the curve, because it includes a month where we got it badly wrong.

Our bounce rate, month by month

Of those 320 bounces, 143 were classified hard and 182 soft. Almost all of this is genuine first-touch cold email: 10,977 of the attempts were first contact, and only 295 were follow-ups to a thread already open.

March is a small sample and should be read as noise. The number that matters is June.

What happened in June

In June our bounce rate hit 10.1%, which is roughly five times what it should be and well into the range where a sending domain starts accumulating real damage. We are publishing it because a page about bounce rates that only shows good months is not worth reading.

We are not going to invent a tidy single cause for it. What we can say precisely is what we changed afterwards, and what the rate did next: 2.8% in July across four times the volume, then 2.0% in August across more volume again. Holding a lower rate while quadrupling sends is a harder test than holding it at a steady volume, and that is the part we would point at.

The three changes that did the work

Verify immediately before sending, not when the address was found. A prospect list decays. An address scraped six weeks ago and verified six weeks ago is not a verified address any more, because people leave and mailboxes get retired. Moving verification to the moment of send rather than the moment of collection removed a whole category of bounce.

Block risky addresses instead of sending anyway. Verifiers return a middle category that is neither clean nor clearly dead. The tempting thing is to send, because the address might be fine and the prospect might be good. The arithmetic does not support it: the proportion that bounces is high enough that the handful of extra deliveries is not worth what the failures cost your domain.

Never re-send to an address that soft-bounced once. A soft bounce reads as temporary, so retrying feels correct. In practice the retry usually fails the same way and you have now spent two failures instead of one on the same dead end.

What 17,507 verified addresses look like

The bounce rate is a lagging measure. The leading one is what verification finds before anything is sent. Across 17,507 addresses we checked, all through the same verifier:

The headline figure is that 17.1% of the contact addresses we find on real blogs are already dead. Roughly one in six. If you build a prospect list and send to it without verifying, that is your floor bounce rate before any other problem is counted, and it is far above the level any sending domain tolerates for long.

The second figure is the awkward one. 34.7% are catch-all, which means the receiving domain accepts mail for any address whatsoever and a verifier cannot tell you whether the mailbox exists. Excluding them wholesale throws away a third of the pool. Sending to them does not produce a bounce, which sounds good until you realise the failure mode is silence: mail to a non-existent mailbox on a catch-all domain is simply swallowed. That is worse than a bounce, because a bounce at least tells you.

A separate note on who you are writing to: 26.6% of the addresses in that set are role inboxes rather than named individuals, the info@ and editor@ and hello@ addresses. Conventional cold email advice is to avoid them. Our data does not support avoiding them, and we send to them deliberately.

Why the hard and soft labels mislead

A hard bounce is supposed to mean the address does not exist and a soft bounce is supposed to mean try again later. In practice the classification describes the response, not the cause.

A permanent rejection is issued for plenty of reasons that have nothing to do with the mailbox. A receiving server that does not like your domain, your IP range, or your sending pattern will refuse the message permanently, and that refusal is recorded as a hard bounce against a perfectly real address. So of our 144 hard bounces, an unknown share are policy blocks rather than dead mailboxes. If you prune your list by deleting every hard-bounced address, you are also deleting good contacts at servers that simply did not want mail from you that day.

The practical version: treat the classification as a hint about what to do next, not as a fact about the address.

Bounces are only half of what damages a sending domain. The other half is reputation, which is what email warmup builds, and we publish our placement numbers there too.

The threshold that actually matters

Benchmarks are less useful than a stop line. We gate sending at 7%: sustained above that, the correct action is to stop and fix the list rather than push through and hope the average recovers. Under 3% is healthy. Under 2% is where you want to sit.

Worth saying plainly: the bounce rate is a proxy for list quality, not a goal in itself. You can drive it to nearly zero by only emailing addresses that are trivially verifiable, which usually means large companies with predictable address formats, which usually means the least relevant prospects. A slightly higher bounce rate against a genuinely relevant list beats a perfect rate against an irrelevant one.

The other number we watch is complaints rather than bounces. Across the same period we have recorded zero spam complaints. Bounces damage your domain slowly; complaints damage it quickly.

How to lower yours

In the order that gives you the most improvement per unit of effort:

Verify at send time. Not at list-build time. This is the single largest lever and it is mostly a scheduling decision rather than a tooling one.

Actually act on the risky category. Blocking it costs you volume and saves you far more.

Stop retrying soft bounces. One failure per dead address is enough.

Decide your catch-all policy deliberately. There is no right answer, but drifting into one by accident means a third of your sending is unmeasurable.

Watch the monthly curve, not the lifetime average. Our lifetime figure of 2.8% hides a month at 10.1%. A lifetime average is exactly the shape of number that lets a problem run for weeks before anyone notices.

Frequently asked questions

What is a good cold email bounce rate?

Under 3% is healthy and under 2% is good. Ours was 2.8% across 11,477 attempts and 2.0% in August 2026. We stop sending entirely at 7%.

What causes cold emails to bounce?

Mostly addresses that were never valid. 17.1% of the addresses we verify come back invalid before we send anything. The rest is policy blocks, full mailboxes, and reputation rejections.

How do you reduce a cold email bounce rate?

Verify immediately before sending, block risky addresses rather than sending anyway, and never re-send to an address that soft-bounced once. Those three took our monthly rate from 10.1% to 2.0%.

Should you send to catch-all addresses?

Cautiously, and deliberately. They are 34.7% of what we verify, so excluding them discards a third of the pool. They do not bounce, but mail to a non-existent mailbox on a catch-all domain vanishes silently, which is harder to detect than a bounce.

Is a hard bounce always a dead address?

No. Policy blocks produce permanent rejections against real addresses. Pruning every hard bounce from your list will delete good contacts along with dead ones.

Method

Figures cover 7 March to 24 August 2026 from our production database. Sends are deduplicated on the provider message ID, because each send writes two rows internally and a naive count doubles the total. Send attempts are delivered messages plus bounces. Verification results are from a single third-party verifier across 17,507 addresses collected during live outreach. Reply figures are deliberately excluded from this page: our conversation records cannot currently be reconciled against sends with enough confidence to publish a rate, and we would rather leave it out than print a number we cannot stand behind.

Related reading