Home / Outbound Benchmarks

Outbound Benchmarks

Measurements from our own B2B outbound operations: email verifiers tested against real bounce data, bounce rates by security gateway, data-provider coverage, and the signals that actually predict replies. Every page states its sample size, method and date. No vendor-supplied numbers.

AgentsVerify vs No2Bounce vs BillionVerify on the same 455 real bounce outcomes

AgentsVerify's 'valid' gate kept 270 of the 280 addresses that delivered (BillionVerify kept 230, No2Bounce strict kept 140) and brought the bounce rate of the kept set from 38.5% to 32.2%. When it rejects an address, it is right 84% of the time (47 real bounces among 56 rejections). It still labeled 128 of 175 real bounces 'valid', 78 of them on domains behind security gateways that no verifier can see.

October 26, 2026 · 455 email addresses with a verified send outcome (175 bounced, 280 delivered); 10 invented addresses on catch-all domains; first production run of 290 catch-all addresses

How much of your site can an AI crawler read? 116 company sites measured: Wix 0.95% text, others 5.62%

Fetched with an AI crawler user agent and stripped of scripts, styles and tags, Wix-built homepages exposed 2,754 characters of readable text inside 268 KB of HTML (0.95%), versus 5.62% for non-Wix sites with a similar word count. 43% of the 116 sites served no H1 tag in the raw HTML, and 3% served an error page to the crawler.

September 21, 2026 · 116 company websites from one B2B prospect list

BillionVerify vs No2Bounce: tested against 455 real bounce outcomes

On 455 addresses with a known delivery outcome (175 bounced, 280 delivered), BillionVerify returned 'valid' for 131 of the 175 addresses that had actually bounced (75%). Filtering on its 'valid' status only lowered the bounce rate from 38.5% to 36.3%. No2Bounce's strict gate caught 68 of 175 bounces but also rejected 140 of the 280 good addresses.

August 20, 2026 · 455 email addresses with a verified send outcome (175 bounced, 280 delivered), plus a second fresh pool of 300

Cold email bounce rate by recipient security gateway: Mimecast, Sophos, Proofpoint, Microsoft 365, Google

Recipient domains whose MX points to Mimecast bounced on 90 domains against 1 accepted in one campaign and 30 against 1 in another, independently of email verification. Excluding Mimecast and Sophos domains with a free DNS lookup took a campaign from 16.95% to 0.84% bounce per lead. Proofpoint, Microsoft 365, Google, Barracuda, Cisco IronPort and others all sat around 1%.

August 7, 2026 · 2 cold email campaigns (2,856 and 623 leads) with per-lead bounce status, then a 5,650-send baseline; 714-lead follow-up test

Sending a Calendly link in a cold email reply: 10% book, 9% show, measured on 129 prospects

129 prospects received our scheduling link inside a human reply between August 2025 and October 2026. 13 booked (10%), 1 cancelled, 12 calls held (9%). Including invitations sent directly, 15 of 129 (12%). Nearly every booking happened within 1 to 9 days; after two weeks without a booking, none came.

October 30, 2026 · 129 prospects who received the link in a manual reply, August 2025 to October 2026

Which email verifiers actually detect catch-all domains? The invented-address test

A random, non-existent local part sent to accept-all domains came back 'valid, is_catchall=false' from BillionVerify on 9 of 9 domains (August 2026) and 'invalid' from AgentsVerify on 10 of 10 (October 2026). No2Bounce does not call it valid but returns 'Deliverable/AcceptAll, catchall=true' with a score of 99, which is the same as saying 'cannot decide'.

October 26, 2026 · 9 catch-all domains (August test) and 10 catch-all domains (October test), 1 invented address each; 307 confirmed catch-all addresses re-submitted to No2Bounce (September)

Yahoo, Hotmail and AOL addresses flagged catch-all bounce at 18%; Gmail verifies cleanly

A campaign of 192 leads on personal addresses that a verifier had labeled catch-all (Yahoo, Hotmail/Outlook, AOL, Comcast and similar) bounced 35 times, 18% per lead, and was suspended automatically 30 minutes after activation. The sibling campaign on addresses rated deliverable (mostly Gmail) bounced at 0.2%.

October 12, 2026 · 192 leads (catch-all tier) and the sibling deliverable campaign on the same sending inboxes

How much of a 'verified' B2B contact list is actually catch-all or dead: GetLeads, Apollo, LeadMagic measured

Sent without any extra verification, six verticals from a 'VALID' discovery pool bounced between 6.7% and 23.3% per lead on the same inboxes that bounced at 1.2 to 4.5% on verified campaigns. Re-verified before sending, 33% of GetLeads VALID addresses, 48% of Apollo 'verified' DTC addresses and 38 to 51% of LeadMagic 'valid' addresses were catch-all.

July 31, 2026 · 6 campaigns sent unverified (about 1,400 leads); 1,133 GetLeads VALID addresses; 9,788 Apollo verified addresses; 99 LeadMagic valid addresses; 2,666 GetLeads VALID addresses in UK and UAE; 1,201 leads scored 99

GetLeads job_titles filter is fuzzy: 29% of 'Founder, CEO, Owner, President' results were not executives

Filtering GetLeads on job_titles = Founder, CEO, Owner, President, Principal returned about 16,000 contacts of which roughly 29% were not decision makers: 'Vice President of Finance', 'Principal Engineering Program Manager', 'VP Business Development', 'Independent Business Owner'. The filter is a substring match, not a role match.

October 24, 2026 · About 16,000 exported contacts, US companies, 6 industry segments

GetLeads phone number coverage: 76% in the US, 4% for Gulf founders

Marketing agencies in the US with a valid email: 6,559 of 8,604 contacts had a phone number (76%). Founders in the Gulf with a valid email: 879 of 21,790 (4%). The whole Gulf pool, any title: 8.6%. A promise like '1,000 leads with phone numbers' is trivial on a US list and structurally impossible on a Gulf list.

August 17, 2026 · 4 target definitions counted in full (8,604; 21,790; 349 and about 1,060,000 contacts); one 86-contact UK export in October 2026

Google Workspace '550 daily user sending limit exceeded' turns every send into a false bounce: 424 leads lost in two days

On October 7, 2026 at 21:28 UTC, 39 of 40 sending mailboxes on a Google Workspace tenant returned '550-5.4.5 Daily user sending limit exceeded'. The platform kept trying and marked 424 leads as bounced in two days (92% of one day's sends), on addresses that had passed a strict verifier. Read per lead, the campaign showed 32% bounce. The sibling campaign on the same mailboxes the same day showed 0%, which is the test that proves the list was not the cause.

October 16, 2026 · 40 mailboxes on one Google Workspace tenant; one campaign of 1,440 leads; a sibling campaign on the same mailboxes; a 1,019-lead re-send test

The 'companies that are hiring' signal selected 283 software vendors and 36 real clinics: measured on a healthcare outbound list

Target: US clinics and care centers, 11 to 500 people, owner-operated. Filter: open roles in operations or customer success, healthcare industries, C-level to director, valid email. After our fit filter and strict verification: 319 contacts. Manual quality check: 36 were care providers, 283 (89%) were software and services vendors selling to healthcare, one of them a direct competitor of the client.

July 29, 2026 · 319 contacts after filtering (from a 19,400-row raw pool); second case: 714 contacts in a marketing-leader list

IcyPeas email finder: it returns an address even when it found nothing. Real hit rates on three lists

On 512 searches, IcyPeas returned 373 addresses, of which 92 carried certainty 'not_found' (a pattern guess, not a find). Keeping only 'ultra_sure' and 'probable' leaves 281 (55% of searches). On 19 searches for large consumer brands, 3 were sure; on 90 searches for small accounting firms with the right domain, 50 were sure (55%). Guesses are not charged; sure finds cost one credit.

October 27, 2026 · 512 searches (October 16), 19 searches (October 26), 90 searches (October 27)

Instantly's dashboard bounce rate vs bounce per lead: a 2.9% campaign was really 17.6%

A campaign displayed at 2.9% bounce in Instantly's analytics (bounced_count over emails_sent_count) was at 17.6% per lead when recounted from the lead export: 3,803 leads, 30 inboxes all bouncing between 14% and 20%. A second workspace showed a campaign at 5.1% per send and 13.8% per lead, still active on an account that already had 5 suspended campaigns.

September 4, 2026 · Two workspaces: one campaign of 3,803 contacted leads (plus a sibling of 201), one workspace of 6 campaigns including 349 leads in the worst

LeadMagic 'unknown' is not a dead address: 54% pass a strict No2Bounce gate

Of 1,332 addresses, LeadMagic returned 'unknown' for 426. No2Bounce rated 228 of those (54%) as strictly deliverable. In the other direction, 15 addresses LeadMagic called 'valid' were dead for No2Bounce. 'Unknown' means the server refuses to answer the probe, not that the mailbox is gone.

August 10, 2026 · 1,332 email addresses (university, hospital and government staff from a conference attendee list)

Low open rate? Cut by mailbox provider first: Gmail 5.7%, Google Workspace 40.9%, Office 365 93.3% on the same send

On one send, overall unique open rate was 6.10% (2.20% when excluding prefetched opens). Gmail was 97.4% of the volume and opened at 5.72%; Google Workspace recipients opened at 40.86% and Office 365 recipients at 93.33%. Same message, same pixel, same sending domains: the tracking works, the consumer Gmail placement does not.

September 8, 2026 · One newsletter send to an opt-in B2C list (consumer software users), split by recipient mailbox provider

MoltSets LinkedIn-to-email: 46% hit rate on profiles the discovery database had no email for, but its grade does not predict bounces

On 300 LinkedIn profiles for which the discovery database returned no work email, MoltSets found 138 business emails (46%): 103 grade A, 28 C, 3 B, 4 F. On 150 bounced and 150 delivered addresses already verified elsewhere, it gave grade A to 61 of the bounced (41%) and to 104 of the delivered (69%): the grade is not a deliverability signal. When the reverse lookup returned a different current work email than the one we had, 36 of 49 such addresses had bounced (73%).

October 22, 2026 · 300 LinkedIn profiles without email; 300 addresses with known outcome (150 bounced, 150 delivered); 110 found domains compared with the database's company domain

Cold email open and click rates are both inflated: 141 opens per 100 emails, 15% to 46% click rates from scanners

Open and click are polluted by two different mechanisms. On one send, 100 emails produced 141 opens (Apple Mail Privacy Protection prefetches the pixel). On another, 271 delivered emails produced a 15.0% click rate while a human click rate on cold email sits at 1 to 3%. On a later program, every subdomain and list showed 36 to 46% clicks, with 40 clicks in the first 4 hours: uniform and immediate equals automated scanning.

October 20, 2026 · 100 emails (opens test, July 2026); 271 delivered emails (clicks, August 4, 2026); 8 sending subdomains across US and Gulf lists (October 2026)

PrismCrawl vs Serper for Google search results: 4x cheaper, 10x slower, and 'site:' is sometimes ignored

PrismCrawl's Google search endpoint costs $0.23 per 1,000 requests against $1.00 at Serper's entry tier (Serper reaches $0.30 only after $3,750 of purchases). It answers in about 13 seconds per call against 1 to 2 seconds. On 60 brand-presence checks its Amazon endpoint agreed with Serper 75% of the time, all 14 disagreements caused by truncated product titles. Its Google endpoint sometimes ignores the site: operator, so an empty-looking result must be retried before being read as 'absent'.

October 28, 2026 · 60 brands checked on both providers (September 2); repeated queries on a UK marketplace-presence task (October 28)

Typesafe AI (Jev) vs LLM agents for ICP qualification: 89% agreement on 3,141 companies for $0.13

One 'choice' question (9 company types) alone agreed with the LLM verdicts on 77% of 3,141 records; adding 4 atomic yes/no questions on the records classified 'brand' raised agreement to 89% with 90% precision and 82% recall. Total cost $0.13, about 5 minutes. The LLM agents themselves were far less stable: four batches of the same composition returned 133, 20, 112 and 36 'keep' decisions.

September 28, 2026 · 3,141 company records with prior LLM verdicts (benchmark); 400 records in a second vertical (calibration); 4 LLM batches of identical composition (variance test, September 30)