COLD EMAIL INFRASTRUCTURE
Cold Email
Infrastructure
at Scale
Working backwards from replies and pipeline
Revenue
Pipeline
Demos
Replies
Foundation
PRACTITIONER WEBINAR
The result we want
Start with the business outcome. Then work backwards.
01
Replies
02
Demos
03
Pipeline
04
Revenue
The chain behind one reply
At every step, ask: what had to be true immediately before this?
Our Intended Business
result
Person
replies
Seen and
relevant
Useful inbox
placement
Trusted
sender
Healthy
foundation
Work backwards
Domain + mailbox foundation
ICP + data + offer + copy
The infrastructure and the campaign are one connected system.
We want to be here.
Seen by the prospect and interesting enough for the prospect to open the email.
Bursting some myths
At the start let’s make sure we are not focusing on myths.
There is a secret hack to making cold email work - there is actually none.
Cold email doesn’t work - it actually works.
Cold email is impossible - yes it is hard but not rocket science, it is focusing and doing the basics again and again.
Deliverability rate tells us how our emails are doing - Wrong, it is placement rates. We want our emails to be placed in the inbox and not delivered to spam or other tabs.
Monitor Open rates - no we don’t. Those are false opens and they prevent emails from reaching the primary box.
Personalization matters in cold email - no it doesn’t. What matters is relevance and timing.
ESPs like gmail and microsoft hate cold emails - no they don’t, for them it is an email. They stop people from getting irrelevant potentially dangerous emails.
Reply-rate maths - the expectation setting.
Cold email can work economically even when the percentage is small.
200
well-targeted emails
about 1
human reply
0.5% planning assumption
Only a small part of the market may be actively buying at any one time.
In-market prospects: 4%to 5% of the TAM
Domain and mailbox architecture
Protect the company domain by separating scaled outbound.
MAIN COMPANY DOMAIN
poso.ai
Normal business communication
Keep this domain away from scaled cold outbound
OUTBOUND DOMAINS
getposo.ai
mailbox 1
mailbox 2
mailbox 3
useposo.com
mailbox 1
mailbox 2
mailbox 3
useourposo.co
mailbox 1
mailbox 2
mailbox 3
Current working rule: up to 3 mailboxes per outbound domain
Capacity planning for 10,000 prospects
Scale by adding controlled capacity, not by pushing one inbox harder.
10,000
prospects
25
sending days
400
sends per day
9
domains
27
mailboxes
15 cold emails / inbox / day
3 inboxes / domain = about 45 sends / day
Send on weekends and holidays if your ICP is startup and smb’s
Authentication and sender identity
SPF or DKIM must authenticate a domain.
Mailbox
identity
SPF
DKIM
DMARC
Domain
alignment
Complete every mailbox identity
First name, last name and a clear profile picture
ANONYMISED DNS SNAPSHOT
poso.ai
SPF
poso.ai.
v=spf1 include:_spf.google.com include:spf.sender-service.example ~all
DKIM
selector1._domainkey.poso.ai.
v=DKIM1; k=rsa; p=MIIBIjANBgkqh...REDACTED...IDAQAB
DMARC
_dmarc.poso.ai.
v=DMARC1;p=reject
SPF and DKIM values altered for security. Do not copy these example values into DNS.
Mailbox provider choice
Use mature sending ecosystems for this outbound model.
PREFERRED
Google Workspace
+
Microsoft 365
EXTRA OPERATING BURDEN
Self-managed SMTP
We want to be good at outbound. We do not want to become a mail-server company.
Buy only one sending infrastructure?
We need to diversify and match esp providers.
Buy both Google Workspace
+
Microsoft 365
We match providers. Microsoft emails get emails from microsoft.��We don’t want to fail.
Buy more than two sequencers (if doing more than 100K sends a month)
Mailbox provider purchase
Buy from the right providers, not resellers and not the cheapest ones
PREFERRED
Directly from
Google Workspace
+
Microsoft 365
We want to generate a sizeable pipeline worth thousands of dollars and not save $100 a month.
Or from providers:
Inboxkit
Zapmail
Hypertide
These often have a great deliverability, flexibility, speed and monitoring including.
Warm-up belongs in the foundation
Build activity gradually before asking a new mailbox to carry real cold volume.
Observed warm-up activity and inbox placement
Working figures: 14-60 days, 20%-30% replies, 20 warm + 10 cold
Warm-up score and inbox placement
These signals answer different questions.
Warm-up dashboard
Provider placement test
about 62%
100% Google inbox
Warm-up dashboard result
Google-specific provider result
One mailbox, two different views of placement.
PRACTITIONER EVIDENCE - DO NOT GENERALISE TO EVERY WARM-UP PRODUCT
What if you don’t want to wait for 14 days to start sending?
You can buy pre warmed up mailboxes.
They cost around $12 to $15 per domain annually and then around $8/mailbox/domain
Pros:
Start sending instantly
Cons:
Costly
None branded sending domains
We still need to keep warming up those domains.
Warming up never stops.
Inbox Placement Tests Results
A warmed mailbox still needs a launch gate.
Use tools like glockapps, your sequencer and google postmaster to understand deliverability and placement impact.
Blacklists we care
Many blacklists providers list domains to earn money. They don’t impact our real deliverability.
Warmup and Real Send Pattern
What makes emails get better deliverability
We slowly switch from warmup to real sends.
Before launch we will have 40 warmups going every day.
On the day of the launch we will have 30 warmups and 10 real sends.
After a week it will be 20 warmups and 20 real sends.
We keep it at that.
If we see good results we can do 20 warmups and 25 sends.
Even increase to 30 sends while keeping 20 warmups.
Warmups help us absorb the spam and bounce rates.
Launch validation
A warmed mailbox still needs a launch gate.
Placement test
Known auto-reply
check
Bounce + provider
errors
LAUNCH
GATE
Launch
Investigate
Hold
Working placement target: above 95%
95% PREFERRED TARGET AND ORIGINAL 90% LAUNCH GATE REQUIRE VERIFICATION
Sender lifecycle and reserve capacity
Prepare replacement capacity before a sender degrades.
Provision
Warm
Test
Active
Degraded
Pause
Recover
Replace
ACTIVE FLEET
Sending controlled cold volume
RESERVE FLEET
Warming in parallel
Working reserve: about 50% of active capacity
If you have say 100 mailboxes sending from currently
Then keep 50 mailboxes always warming up in parallel.
Operating guardrails after sending begins
When a metric breaks the internal range, pause and investigate.
<0.1%
Spam complaints
<1%
Hard bounces
>95%
Inbox placement
>99%
Delivery
Pause first
Diagnose the source
Resume carefully
Volume comes second.
The monitoring stack
Bring signals together and act before a small issue becomes a reputation problem.
Recipient outcomes
Human replies, positive replies, OOO, complaints
Placement + provider feedback
Inbox placement, spam, deferrals, throttling
Sender health
Domain, mailbox, bounce and error health
TOOLS
Google Postmaster
Microsoft SNDS
Placement tools
Sequencer + error data
Enterprise email adds another gate
The message may meet a security layer before it reaches the corporate mailbox.
STARTUP / SMB
Sender
Google or Microsoft
Prospect
ENTERPRISE
Sender
Security gateway
Corporate mail system
Prospect
Proofpoint | Mimecast | Cisco | Barracuda
Clean history can help. Domain age helps.
Placement gives the message a chance
Infrastructure gets the email seen.
Inbox
Recipient
reaction
ICP + data + offer + copy
The recipient decides what happens next.
ICP, list quality and verification
Wrong people create poor reactions. Bad data creates bounces.
Target ICP
Multiple data
sources
Verify 1
Verify 2
Catch-alls
Suppression
ONE GLOBAL SUPPRESSION LAYER
An unsubscribe, complaint or permanent bounce must apply across the whole sender fleet.
VERIFICATION OUTPUT
821 valid
188 invalid
16 catch-all
Ideal: 0% hard bounces | Operational ceiling: below 1%
Buying timing and the same ICP
Sometimes the list was right and the timing was wrong.
MONTH 0
MONTH 2
Same ICP
Different buying window
A relevant market can contain different in-market buyers later.
Working cadence: revisit about every 2 months
Copy and message hygiene
Keep the first email simple and easy to process.
From: First Last
Subject: quick question
Hi Name,
Short, relevant message about a real problem.
Simple question.
First Last
Title, Company
Replace this mockup with a real anonymised email
If a URL must appear, the notes suggest a plain-text form such as poso(dot)ai.
PLAIN-TEXT URL FORMAT IS A WORKING PRACTICE, NOT A UNIVERSAL RULE
Template similarity and real personalization
Keep the offer consistent. Avoid sending thousands of exact copies to restrict copy fingerprinting.
IDENTICAL
CONTROLLED VARIATION + RELEVANT CONTEXT
Role
context
Company
trigger
Relevant activity
Variation cannot rescue the wrong ICP or a weak offer.
Reply-first measurement
Measure the path to a business conversation.
Human replies
Positive replies
Demos
Opportunities
Pipeline
OPEN RATE
CLICK RATE
Open and click rates do not drive our decisions.
Separate any reply from a positive reply.
Controlled sending behaviour
Outbound should look controlled, not like a burst.
09:00
10:15
11:30
12:45
14:00
15:15
SPACE SENDS ACROSS THE DAY
1-2 emails
per sequence
Fresh follow-up
angle
Test timing
by ICP
9-11 minute spacing | 6+ hour window | 3-5 day follow-up
The cold email feedback loop
Healthy foundation
Useful placement
ICP + offer + message
Recipient reaction
Future reputation
Next
send
Infrastructure and campaign quality feed each other.
Diagnosing what is actually broken
Start with the observed failure and move one step backwards.
Weak pipeline
Check reply quality and sales conversion
Few replies
Check ICP, timing, offer and copy
Poor placement
Check sender health, authentication and sending behaviour
Bounces or complaints
Check data, suppression and relevance
Work backwards from the result
Want to connect with me?
Linkedin Profile: https://www.linkedin.com/in/abhisheekpatra/