We sent 251 cold emails with our own tool: 5 bounced, 0 replied, and our data has holes
Every number from 251 cold emails GrowFast sent for our own products in October 2026: bounces by mailbox type, what the replies were, and what we failed to record.
GrowFast is a cold-email tool, so before selling it we ran it on our own products. This is the full result, including the parts that make us look bad: no genuine replies, a delivery log that is empty for more than a third of the sends, and one configuration mistake of our own. Every count comes from our production database (read-only, queried 2026-11-17); the SQL is at the end.
TL;DR
- 251 emails sent between 4 and 23 October 2026 (191 first emails and 60 follow-ups, to 191 recipients). 5 bounced. 0 genuine replies were recorded. The only inbound messages were 8 automatic replies and 2 test messages from our own domains.
- 4 of the 5 bounces were role mailboxes such as
hello@andcontact@. Among sends with a stored delivery result, role mailboxes bounced 4 of 69 (5.8%) and named addresses 1 of 87 (1.1%). That is a signal, not a proven effect: p = 0.17 (Fisher's exact test). - 95 of the 251 sends have no stored delivery result, so every bounce rate here is approximate. Google's and AWS's published guidance (below) matters more than our small sample.
What we sent
| Item | Count |
|---|---|
| Rows in the outreach table (created 4-23 Oct 2026) | 256 |
| Sent | 251 |
| Failed before sending | 3 |
| Skipped as duplicates | 2 |
| Sent: first emails (one per recipient) | 191 |
| Sent: follow-ups (to 38 of those recipients) | 60 |
| Campaigns / products promoted | 47 / 7 |
Recipients are B2B contacts found by GrowFast's lead finder from addresses published on company websites. We are not naming products or recipients.
What bounced
Five sends bounced, to five different recipients. No follow-up went to an address that had bounced.
| Delivery result stored | Sends |
|---|---|
| Delivered | 151 |
| Bounced | 5 |
| No result stored | 95 |
The bounces were on 21 October (2), 22 October (2) and 23 October (1). The 36 sends on 19 and 20 October were all delivered.
Role mailbox or a person's address?
"Role" here means the part before the @ is one of hello, support, contact, sales, info, help, hi, office, admin. Everything else is "named".
| Address type | Sends | Delivered | Bounced | No result | Rate, all sends | Rate, sends with a result |
|---|---|---|---|---|---|---|
| Role mailbox | 126 | 65 | 4 | 57 | 3.2% | 5.8% (4 of 69) |
| Named address | 125 | 86 | 1 | 38 | 0.8% | 1.1% (1 of 87) |
The last column is the honest one. Missing results are not spread evenly: 45% of role-mailbox sends have none against 30% of named ones, so dividing by all sends flatters role mailboxes. Inside the role group, hello@ had 3 bounces in 37 sends with a result (8.1%) and contact@ 1 in 19 (5.3%). support@ had 27 sends and no results at all, so it tells us nothing.
Five bounces cannot settle this. Fisher's exact test gives p = 0.17 on the right-hand counts and p = 0.37 on the all-sends counts.
A cross-check, with lower confidence. The 63 sends on 4 October also have no stored result. A comment we wrote in our source code after that batch records 6 bounces in the first 59 cold sends: 3 of 12 to hello@, 3 of 26 to support@, none of 16 to other role mailboxes. Our repository is private, so you cannot check it. If you accept it, the combined tally is about 10 bounces in 123 role-mailbox sends against 1 in 92 named sends (p = 0.03); the 5 early named sends are inferred by subtraction.
What replied
| Inbound message | Count | Notes |
|---|---|---|
| Automatic (out-of-office style) | 8 | 1 from an address we had mailed, 7 from other addresses |
| Labelled "interested" | 2 | Test messages sent from our own domains while we tested reply handling |
| Genuine replies | 0 |
The honest answer to "what replied" is nothing. The two "interested" rows are the kind of number that ends up in a case study when nobody checks where it came from, so we checked: the sender domains are ours.
Two caveats. GrowFast only records replies that reach the address it manages for the sender; a reply to some other mailbox would not appear. And one send row carries the stop reason "replied" with no reply timestamp and no matching inbound message. Our scheduler writes that label for any contact already silenced for another reason, so we do not count it. We have no click or signup attribution for these sends and make no claim about either.
What went wrong in our own measurement
- No delivery result for 95 sends (all sends from 4-17 October). The 156 sends from 19 October on all have one.
- A configuration mistake. One send failed with "Domain gmail.com is not verified": the tool tried to send from a personal address on a domain we do not control.
- Two more refused rows. The sending provider refused one recipient already suppressed after a previous bounce and one whose domain has no mail servers.
- Two duplicates were skipped, as intended.
What the public guidance says
Authenticate your domain. Since 1 February 2024 Google requires SPF or DKIM from every sender to personal Gmail accounts, and SPF, DKIM and DMARC, a spam rate under 0.30% in Postmaster Tools and one-click unsubscribe for marketing mail from anyone sending more than 5,000 messages a day (Google, Email sender guidelines). We send far below that, so DMARC is not formally required of us, though Google recommends all three for everyone. The specifications are RFC 7208 (SPF), RFC 6376 (DKIM) and RFC 7489 (DMARC). Google says DKIM keys must be at least 1024 bits and recommends 2048.
Warm up by ramping slowly. Google's page never says "warmup". It says to start at a low volume to engaged recipients, raise it gradually, avoid bursts, and cut volume if mail starts bouncing or being deferred. GrowFast's managed sending address ramps the same way: 15 emails a day for the first 3 days, rising to 50 a day after a week (growfast.sh homepage).
Know the bounce thresholds. Amazon SES says to keep the bounce rate below 2%, reviews accounts at 5% and may pause sending at 10%, and tells you to remove bounced addresses immediately and never buy, rent or share lists (AWS, SES sending review FAQ). Our 5 of 251 is 2.0% on all sends and 5 of 156 (3.2%) on sends with a result. AWS counts only hard bounces to unverified domains over its own "representative volume", so you cannot reproduce its number from a table like ours.
Cold email is still regulated. CAN-SPAM has no business-to-business exception. Each commercial message needs accurate headers, a truthful subject, a valid postal address and a working opt-out honoured within 10 business days; penalties run up to $53,088 per email (FTC guide). EU and UK rules differ, and we are not lawyers.
What we changed
- Service inboxes (
info@,support@,help@,admin@and similar) are never emailed. Shared inboxes (hello@,hi@,contact@,team@,sales@) rank below named addresses and nothing is written for them without a contact name. - Every address gets an MX lookup first.
- A run halts if at least 10 results are in and over 15% bounced, and a new campaign will not start if the sending account has no room under a 1% bounce ceiling over the last 14 days.
- Sending is paused. The last row in the table is from 23 October 2026. We will rerun this after a new batch of a few hundred sends, with the delivery result stored from the first message.
FAQ
Are role mailboxes worse than personal addresses for cold email?
Our data points that way but does not prove it: 4 of 69 against 1 of 87 among sends with a known outcome, p = 0.17. A published hello@ is often a website convention and not a mailbox anyone created, which is a plausible reason we have not tested.
What bounce rate is acceptable? AWS says below 2%, with review at 5%. Gmail's published limit is on spam complaints, not bounces: keep the Postmaster Tools spam rate below 0.30%. Anything near 5% on a list is a reason to stop and clean it.
Do I need to warm up a domain before cold emailing? Google prescribes no schedule, only gradual increases and no bursts. Start with a few emails a day to the most likely real recipients, watch bounces and spam rate, and raise volume only while both stay low.
Is zero replies from 191 recipients normal? We have no benchmark we trust and will not quote one. Ours was zero, we changed who we write to, and we have not sent a second batch.
Can I rerun this on my own data? Yes. The SQL needs only a table of sends with a status, an address and a delivery result.
The SQL
Read-only; no address is selected. delivery is the result the sending provider reported for each message.
-- Status and delivery result of every row
select status, coalesce(delivery,'(none)'), count(*) from outreach_send group by 1,2 order by 3 desc;
-- Role mailbox vs named address, sent rows only
select split_part(lower(email),'@',1) in ('hello','support','contact','sales','info','help','hi','office','admin') as role,
count(*) sent,
count(*) filter (where delivery='delivered') delivered,
count(*) filter (where delivery='bounced') bounced,
count(*) filter (where delivery is null) no_result
from outreach_send where status='sent' group by 1;
-- Inbound messages: category, automatic or not, sent from a domain of ours?
select category, automatic,
split_part(lower(from_email),'@',2) in ('example-our-domain-1.com','example-our-domain-2.com') as from_own_domain,
count(*)
from inbound_message group by 1,2,3;
The p-values use a two-sided Fisher's exact test (for example scipy.stats.fisher_exact([[4,65],[1,86]]) in Python).
Methodology and data
- Data date: queried 2026-11-17 from GrowFast's production database, read-only. Sends run 4-23 October 2026; there are no later rows.
- Included: all 251 rows with status "sent". The 3 failed and 2 skipped-duplicate rows are listed separately.
- Excluded from replies: 2 inbound messages labelled "interested", because both came from our own domains. They are shown, not hidden. Replies delivered outside GrowFast's managed address are not in the data.
- Role mailbox definition: local part is one of
hello, support, contact, sales, info, help, hi, office, admin. Other generic-looking local parts (for exampleinquiry,operations) count as named; none bounced, so moving them would lower the role rate by under half a point. - Statistics: two-sided Fisher's exact test (Python). With 5 bounces, none of these comparisons is conclusive.
- Not in the dataset: recipient names, addresses and domains, products promoted, clicks, signups.
- Earlier post: our first write-up covered one week (160 sends); this one covers the whole table.
Disclosure: we make GrowFast. The product is named here only because this is the data from running it.
Related: our first write-up covered one week of this data, hello@ vs named person. To check your own sending domain, the free SPF checker, DMARC checker and deliverability audit read live DNS and need no account.
From the same team
SEOBolt
Technical SEO and AI search audits with ready-to-apply fix plans.
Try SEOBolt →Bugsly
AI-powered error tracking that explains your bugs.
Try Bugsly →AutoShort
Turn a topic into a faceless short: AI script, voiceover and captions.
Try AutoShort →RemixViral
Paste a long video and get back the moments worth posting, cut vertical and captioned.
Try RemixViral →