Average Cold Email Bounce Rate: What Is Good, and How to Reduce It (2026)

On this pagev

• The receipt: two different measurements, the same number
• What a bounce actually is
• What counts as a good bounce rate
• Why catch-all domains hide bounces until you send
• How to measure your own rate
• How to cut your bounce rate
• Where Kustiq fits
• Key takeaways

Two different measurements landed on the same number, and they do not measure the same thing. Sparkle measured a 13.3% verifier bounce on Apollo database output in its 450-lead test. Our own probes refused 13.3% of already researched addresses that reached a definitive server verdict (82 of 617 probes with a definitive verdict, 2026-03-19 to 2026-07-05; across all 3,508 recorded outcomes that is 2.3%, because most probes ended without a verdict). We do not publish an industry average. 82% of our probe outcomes ended without a mailbox verdict at all (2,860 of 3,508 outcomes), and our sample covers one set of outbound work, not the internet. The familiar 2% to 3% ceiling, the commonly repeated rule of thumb, describes the verified list you send, while a raw purchased or researched list can sit far above it before cleaning. You close that gap by verifying close to send time, removing definitive refusals, and sending catch-all addresses as a separate segment. Catch-all covered 66.8% of resolved domains in our sample (141 of 211 domains).

We build Kustiq, the sales intelligence product mentioned later in this post. The benchmark numbers below come from our own SMTP probe logs, and we link the full dataset so you can check the denominators yourself.

The receipt: two different measurements, the same number

We recorded 3,508 SMTP probe outcomes across 4,127 processed contacts, and only 617 of those probes came back with a definitive server verdict. Inside that definitive set, servers accepted 502 addresses (81.4% of 617) and refused 82 (13.3% of 617). Each address had already passed research that made it look plausible. The receiving server still returned a refusal before any message existed.

Another 28 probes found no MX record left for the domain (4.5% of 617). That usually means the company changed its mail setup or stopped receiving mail there. A further 5 probes hit catch-all detection at probe time (0.8% of 617). Refused mailboxes and domains with no MX record together account for 110 of the 617 definitive verdicts. Those addresses were dead before a sequencer could send to them.

Sparkle ran a 450-lead test on Apollo database output and measured a 13.3% verifier bounce in that 450-lead sample, published in its Apollo vs ZoomInfo comparison. Apollo is a large B2B contact database; Sparkle's figure is a verifier bounce on database output, not an SMTP refusal rate, so the two numbers are not the same quantity. Two different lists and two different checkers gave the same refusal share to one decimal place.

People type "average cold email bounce rate" expecting a property of cold email itself. These two checks measure the state of address data at the moment someone checks it. A researched list and a database export both showed roughly one dead address in eight among the addresses that got a definitive answer at check time. Your sent bounce rate depends on whether you remove those dead addresses before the send or let the send find them.

We published the underlying counts, caveats and all, in our catch-all email benchmark. Read that post before quoting any figure from this one. The sample is Kustiq outbound work aimed at SMB and mid-market B2B companies. It does not cover enterprise lists or consumer mail, and we will keep saying so wherever the numbers appear.

What a bounce actually is

A bounce is a delivery failure report, and the report arrives at different moments depending on where the failure happened. Some receiving servers reject during the SMTP conversation itself, right when your server names the recipient. Others accept the message first and send a non-delivery report back later. The timing matters because a check can detect the first kind before sending, while it often cannot detect the second kind.

Hard bounces are permanent failures. In the response codes behind our logs (617 definitive verdicts), permanent failures sit in the 5xx range and include the familiar 550 response. A 550 at recipient stage usually means the mailbox does not exist on a server that does exist. The person left the company, someone mistyped the address when recording it, or someone built the address from a naming pattern the company never used. Retrying that address next week will not change the answer. Mailbox providers lower sender reputation when senders keep mailing addresses that return 5xx responses.

Soft bounces are temporary failures reported in the 4xx range in those same logs (617 definitive verdicts). The mailbox may be full or the server may be asking you to slow down. Greylisting returns a temporary failure to unknown senders on purpose and accepts the retry. A soft bounce is a reason to retry later and watch the pattern. An address that returns 4xx on every attempt for weeks often points to a permanent problem, and most sending tools retire it after repeated tries.

Policy rejections belong in a separate category. The mailbox can be real and the server can still refuse your message because authentication failed or your sending domain has a reputation problem. Gateway services add another step here, since the receiving company may never see a message its gateway rejected. Counting a policy rejection as a bad address makes your list hygiene data wrong. Counting it as a good address makes your deliverability data wrong. Record it separately. A raw bounce total from a sequencer export mixes these categories and can mislead.

Address construction causes many hard bounces. Of 70 domains with one identifiable email format in our benchmark, the single most common pattern, first.last@, covered only 15.7% of those 70 domains. The top three formats together covered about 40% of the same 70 domains. Ten formats shared a nearly flat distribution across that 70-domain set, running from first.last through first_last. Our post on how to verify an email without sending walks through the recipient-stage check that replaces the guess.

When the 28 probes in our set found no MX record (4.5% of 617 definitive verdicts), the domain itself had stopped advertising a mail server. No verification tool can rescue that address and no copywriting will matter. The company website still loads in many of these cases. A domain can keep its website after it stops accepting mail.

What counts as a good bounce rate

The 2% to 3% ceiling, the commonly repeated rule of thumb, describes sends. Read it as the hard bounce share your sending setup can take per campaign without mailbox providers treating you as careless. It says nothing about the list you started with. Our probes refused 13.3% of researched addresses before any send (82 of 617 definitive verdicts), and that refusal happened in a checker rather than in a campaign. Verification removes the difference between a raw list and a sendable list.

We refuse to print an industry average bounce rate. Our own logs cannot support one, since 82% of probe outcomes ended without a mailbox verdict through timeouts and proxy outages (2,860 of 3,508 outcomes). A number built only from servers that answered would omit most checks and describe a cleaner set of addresses than the one we probed. We publish the share without a verdict alongside every rate. Any vendor average you read elsewhere deserves the same question about what share of checks never returned a verdict.

Averages also hide how much the receiving platform matters. Confirmed-deliverable share across all processed contacts, infrastructure failures included, was 15.8% on Google Workspace (273 of 1,729 contacts) and 9.9% on Microsoft 365 (115 of 1,166 contacts). These shares mostly reflect probes that never got an answer. That is a statement about reachability rather than address quality. A single blended rate across platforms matches neither platform on its own. A list with many Microsoft 365 tenants behaves differently from one with many Google Workspace tenants, before you change a single word of copy.

Gateway-protected domains widen that range. Confirmed-deliverable share was 11.2% behind Proofpoint (16 of 143 contacts) and 3.3% behind Mimecast (5 of 154 contacts). Gateways return little mailbox information to probes. Treating a silent gateway as a verified address makes an unconfirmed address look verified. Treating it as a dead address throws away real people. Segment gateway domains and watch their replies separately.

Set your target at the level of the sent, verified segment. Hold hard bounces under that ceiling of 2% to 3% per campaign and investigate any segment that crosses it. Expect the raw list feeding that segment to start with more dead addresses. Sparkle found a 13.3% verifier bounce inside a 450-lead database export, and our probes showed 13.3% on researched addresses (82 of 617 definitive verdicts). Dead addresses at intake are normal. Sending to them after a check is a process fault.

Why catch-all domains hide bounces until you send

A catch-all domain accepts mail for every address at recipient stage, including addresses nobody reads. The server check that exposes a dead mailbox on a normal domain returns an accept here. Verification tools can detect that the domain behaves this way. They cannot tell you whether a specific person exists behind it, because the server returns the same accept either way.

Catch-all behaviour showed up on 141 of 211 resolved B2B domains in our benchmark, which is 66.8% of that 211-domain sample. Two of every three domains you prospect into may refuse to confirm any individual mailbox. A verification report that labels those addresses verified turns an unknown into a verified count. Label them catch-all, record the accept, and leave the person's existence open.

What happens at send time varies by domain. Some catch-all servers accept the message and discard mail for unknown recipients without sending a bounce. Your delivered count then includes mail that no one received. Others accept at recipient stage and bounce after reading the message headers. The verification report looks clean before the send, and the campaign records a hard bounce that no recipient-stage check could catch, because the server accepted the address at that stage. The deeper explanation lives in our email finder and verifier guide.

Send catch-all addresses in their own segment on their own schedule. A bad group of constructed addresses then cannot push a whole campaign over your hard bounce ceiling. Watch that segment for replies and out-of-office messages, since real human responses are the only per-address confirmation a catch-all domain will give you. When an address at a catch-all domain matters commercially, send one personalised message to the named person instead of putting the address in a sequence. One send creates at most one bounce for that address, while a sequence can send again to the same dead address.

Do not let a tool count catch-all accepts as verified. Ask wherever you buy or check data how the vendor labels catch-all results. If a vendor cannot answer, assume it counts them as good until it shows otherwise. On our side, every email gets a live SMTP RCPT TO handshake at delivery, and catch-all results are labelled rather than counted as verified. That labelling decision changes how you read every rate after it, because your verified segment then contains only mailboxes that a server confirmed.

How to measure your own rate

Define the denominator before you argue about the number. For sent campaigns we recommend hard bounces divided by attempted recipients in that campaign, with soft bounces tracked on a separate line. Mixing the two failure types into one rate hides opposite problems behind one figure. A high rate from soft bounces asks for slower sending and patience. A high rate from hard bounces asks you to remove dead addresses. The remedy differs, so the measurement must differ.

Segment by source before you segment by anything else. A purchased export, a hand-researched list and an inbound form fill group carry different failure profiles. Sparkle measured a 13.3% verifier bounce on a 450-lead Apollo export, while our probes refused 13.3% of pipeline-researched addresses (82 of 617 definitive verdicts). Your own sources will differ from both figures, and the difference between your sources is the part you can act on. If one supplier's addresses refuse at twice the rate of another's, you have a concrete reason to challenge the price or stop buying from that supplier.

Segment by receiving platform next, because platform behaviour distorts confirmation. Our Google Workspace contacts confirmed deliverable at 15.8% (273 of 1,729 contacts) while Mimecast-fronted contacts confirmed at 3.3% (5 of 154 contacts). A campaign aimed at larger companies behind gateways will show worse confirmation and messier bounces through no fault of the list builder. Platform splits keep gateway behaviour out of your judgement of the list builder.

Watch the verdict-free share as a metric in its own right. Our probes ended without a mailbox verdict in 82% of outcomes (2,860 of 3,508 outcomes) because servers timed out or proxies failed mid-check. If your verification tool reports only accepts and refusals, ask where the inconclusive checks went. A rising inconclusive share predicts messy campaigns before the bounce rate itself moves. It also tells you when to run the checks again rather than trust an old pass.

Mind the size of the campaign under the percentage. On a small send, one dead address can swing the reported rate by several points, and that swing means nothing about your list quality. Read rates across a rolling month of sends instead of reacting to a single Tuesday. Record the date each address was last verified alongside the result. When a segment's bounce rate rises, check that date first and re-run the check.

How to cut your bounce rate

Verify as close to send time as your process allows. D&B puts B2B contact data decay at about 2.1% per month, a figure still cited. Decay means a verification pass ages from the moment it finishes. People change jobs and companies retire mailboxes in the gap between list building and campaign launch. Checking at delivery shrinks that gap to hours. Checking at list purchase leaves the whole quarter open to change. Our business email lookup guide covers the finding side, but the checking side belongs as late as possible in your process.

Remove definitive refusals without negotiating with them. A recipient-stage refusal from a server that returns a definitive response is a clear signal. Our probes produced 82 such refusals among 617 definitive verdicts. Every one of those addresses removed before a send is a hard bounce that never touches your sender reputation. The same applies to domains with no MX record left, which accounted for 28 of those 617 verdicts. Delete them and log the reason. Do not add them again next quarter under a new campaign name.

Stop constructing addresses from pattern guesses at unknown companies. The most common single format covered just 15.7% of 70 single-format domains in our benchmark, and the top three formats together reached only about 40% of that same 70-domain set. Pattern construction creates addresses that return hard bounces while the work looks like research. Where you must construct, run a recipient-stage check and accept that catch-all domains will not confirm the result. Where the domain confirms nothing, consider whether that prospect belongs in a low-volume manual sequence instead.

Handle gateway-fronted domains as their own group. Proofpoint-fronted contacts confirmed deliverable at 11.2% (16 of 143 contacts) and Mimecast-fronted contacts at 3.3% (5 of 154 contacts). Low confirmation there reflects how little mailbox information gateway replies give. Send gateway segments separately so their results do not mix with the results from confirmed segments. Separate sends also protect your main sending reputation while you learn how each gateway group responds.

Keep your authentication records correct and your sending discipline steady. Correct SPF, DKIM and DMARC records will not fix a dead mailbox, yet missing records add policy rejections to your bounce reporting. Send from domains you can afford to rest when a segment misbehaves. Retire addresses that soft bounce repeatedly instead of letting a sequencer retry them without end. Mailbox providers set reputation from your past sends, so this record keeping and removal work decides how later campaigns land.

Re-verify the backlog on a schedule tied to decay. Given D&B's roughly 2.1% monthly decay figure, a list verified at the start of a quarter carries three months of job and mailbox changes by its end. Running recipient checks again before each major campaign costs less than restoring a sending domain's reputation after a dirty send. Ongoing monitoring helps here too, since account changes and hiring moves show up in public data between campaigns. We wrote about the research side of that workflow in our decision-maker email guide.

Where Kustiq fits

Kustiq enters this workflow at the checking step rather than the sending step. Paste a B2B domain and you get a company profile plus one SMTP-verified decision-maker contact, with verification done through a live SMTP RCPT TO handshake at delivery. Because the handshake happens when the profile is produced, the time between check and use stays small. Catch-all results come back labelled as catch-all and are not counted as verified, which keeps your segment math accurate. Kustiq does not send email, so your sequencer and sending domains stay under your control.

The commercial terms stay simple as of October 2026. A free account includes three profiles per week with no card required, and the free trial page works without an account at a lower weekly allowance. Paid plans on our pricing page run on credits with no per-seat fees, where one credit buys one profile. For list-scale work, Contact Finder checks uploaded company lists and returns up to three SMTP-verified contacts per company. If your current problem is database output quality rather than workflow, our cheaper Apollo alternatives and sales intelligence tools for small teams posts cover the other options.

Who should skip us? Teams that need a sequencer, since we do not send. Teams that want a large static database to browse by headcount bands will find purpose-built databases a better fit, accepting the verification burden Sparkle measured at 13.3% on a 450-lead export. Our fit is the smaller team that researches named accounts and wants each address checked at the moment of use. The benchmark behind this post exists because that checking produces logs, and the logs turned out to be more interesting than we expected. You can audit every denominator in the full benchmark post.

Key takeaways

• A pipeline-researched list and a database export both showed 13.3% in two different measurements (82 of 617 definitive verdicts in our logs, 2026-03-19 to 2026-07-05, and a 450-lead Sparkle test).
• We publish no industry average because 82% of our probes ended without a verdict (2,860 of 3,508 outcomes).
• The 2% to 3% ceiling, the commonly repeated rule of thumb, applies to the verified list you send, so measure hard bounces per sent segment.
• Catch-all covered 66.8% of resolved domains in our sample (141 of 211 domains), so label those addresses and send them separately.
• Pattern guessing rests on weak evidence, since the top format covered 15.7% of 70 single-format domains.
• Gateways return little mailbox information, with Mimecast at 3.3% confirmed deliverable (5 of 154 contacts), so segment gateway domains before judging list quality.

Frequently Asked Questions

What is a good bounce rate for cold email?

A verified sending list should hold hard bounces under the 2% to 3% ceiling, the commonly repeated rule of thumb. That ceiling describes the list you send after verification. Raw researched lists can start far higher, and one benchmark refused 13.3% of addresses before sending (82 of 617 definitive verdicts).

What is the average cold email bounce rate?

No honest industry average can be drawn from the data in this post. The underlying sample covers one outbound funnel aimed at SMB and mid-market B2B companies. Most probe outcomes ended without a mailbox verdict at all (2,860 of 3,508 outcomes).

Why do verified emails still bounce?

Catch-all domains accept every address at recipient stage and may discard or bounce mail for unknown recipients later. Such domains covered 66.8% of resolved domains in one benchmark sample (141 of 211 domains). Mailboxes also close between verification and sending as people change jobs.

What is the difference between a hard bounce and a soft bounce?

A hard bounce is a permanent failure, such as a mailbox that does not exist on a working server. A soft bounce is a temporary failure, such as a full mailbox or a server asking the sender to retry later. Hard bounces call for removal of the address. Repeated soft bounces call for retirement of the address after a fair retry window.

How does Kustiq help reduce email bounce rate?

Kustiq runs a live SMTP RCPT TO handshake at delivery for every email it returns. Catch-all results are labelled rather than counted as verified, so those addresses can be sent as a separate segment. A free account includes three profiles per week with no card required.