Most people start cold email by writing the email.
That’s step seven.
I’ve worked with over 250 B2B SaaS companies on their outbound, and the single most common mistake I see is teams jumping straight to copywriting without doing any of the foundational work first. They open a Google Doc, write three emails, load them into their sequencing tool, and wonder why nothing happens.
The emails aren’t the problem. The order is.
Every cold email campaign has seven steps, and they need to happen in sequence. Each one feeds into the next. Skip step two and you’ll get stuck on step five. Rush past the persona work and your copy will be generic no matter how talented the writer is.
Here’s the process I use with every client. It’s not glamorous. But it works.
Step 1: Estimate your total addressable market
Before you think about infrastructure, messaging, or anything else, you need to know the size of the opportunity you’re working with.
How many companies could you possibly sell to? Not how many you want to sell to. How many could realistically be a fit for your product or service?
This matters because it directly affects your infrastructure decisions (more on that in step two), your campaign strategy, and how quickly you’ll burn through your market.
If your TAM is 200 companies, you need a very different approach than if it’s 10,000. With 200, every email matters and you can afford to be highly manual. With 10,000, you need systems and automation to reach enough people.
A rough estimate is fine here. You don’t need a 40-page market sizing report. You need a number that feels right based on your geography, industry, company size, and the type of problem you solve. You can use tools like Clay, Apollo, or even LinkedIn Sales Navigator to get a ballpark.
The point is to ground yourself in reality before you start building.
AI can help here. You can ask Claude or ChatGPT to estimate your TAM based on the filters you describe (industry, geography, company size, business model). It’ll give you a starting point in seconds. But don’t take the number at face value. Cross-reference it against a quick search in LinkedIn Sales Navigator or Apollo. AI gives you speed. You bring the judgement on whether the number feels right.
Step 2: Set up your email infrastructure
Now that you know your TAM, you can make sensible decisions about infrastructure.
If your addressable market is small (under 500 companies), you might get away with sending from your primary domain with a few safety precautions. If it’s large (5,000+), you’ll need secondary domains, inbox warming, and a sending tool that handles rotation and deliverability.
I’m not going to go deep on infrastructure here because it changes constantly and the specifics depend on your setup. But the key decisions at this stage are: how many emails do I need to send per day to hit my targets? How many sending domains and inboxes do I need to support that volume safely? And what sending tool am I going to use?
The mistake I see most often is teams setting up infrastructure for volume they don’t need, or worse, blasting thousands of emails from their primary domain and torching their deliverability in the first week.
Match the infrastructure to the TAM. Not the other way round.
Step 3: Define your ICP and personas
This is where most teams think they’ve already done the work. They haven’t.
Your ideal client profile is not “B2B SaaS companies with 50 to 500 employees.” That’s a demographic filter, not an ICP. A proper ICP describes the type of company that gets the most value from what you sell, stays longest, and is most likely to say yes to a cold email right now.
Look at your best existing customers. Not all customers. Your best ones. The ones that renewed, expanded, and actually use the product. What do they have in common? Industry, stage, team size, tech stack, business model, growth trajectory? Those patterns are your ICP.
Then within that ICP, you’ll have multiple personas. The CEO of a 30-person company has different problems than the VP of Operations at the same company. They care about different things, they respond to different messages, and they have different buying authority.
Each persona needs its own campaign. One sequence for all personas is the default, and it’s why most cold email underperforms. An e-commerce agency owner and a head of operations at the same company might both be in your ICP, but sending them the same email is like giving the same presentation to the board and the engineering team. It doesn’t work.
I use a simple ICP and persona worksheet with my clients. For each persona, we map out their role, their day-to-day responsibilities, the problems they’re likely facing, and what a good outcome looks like for them. This becomes the foundation for everything that follows.
AI can help you structure this. Once you’ve done the interviews and identified the patterns from your best customers, you can feed that raw thinking into Claude and ask it to organise it into a clean persona profile. But the inputs have to come from you. AI is brilliant at structuring your thinking. It’s terrible at replacing the thinking itself. If you skip the customer conversations and ask AI to “create a persona for a Head of Legal at a mid-market SaaS company,” you’ll get something that sounds plausible but is built on assumptions, not evidence.
Step 4: Map problems and challenges to each persona
This is the step that separates good cold email from bad cold email, and almost nobody does it properly.
For each persona you identified in step three, you need to answer: what are the two or three biggest problems this person is facing right now that your product or service can help with?
Not features. Not benefits. Problems.
If you sell legal contract automation, the Head of Legal’s problem isn’t “they need better contract software.” Their problem might be that manual review processes that worked when the company was doing 20 deals a quarter are starting to break at 50. Or that they’re spending four hours on every contract redline that should take 30 minutes. Or that non-standard terms are slipping through approvals because there’s no consistent playbook across the team.
Those are real problems. And when your email opens with one of those, the recipient thinks “this person understands my world” rather than “this person wants to sell me something.”
Where do you find these problems? Talk to your customers. Talk to your sales team. Read G2 reviews (yours and your competitors’). Look at job descriptions for the persona you’re targeting. Look at what they’re posting about on LinkedIn. The language your customers use to describe their problems is more compelling than any copy you’ll write yourself.
I typically identify two or three distinct problems per persona. Each one becomes a separate email in the sequence (more on that in step seven).
This is another area where AI genuinely helps. You can feed G2 reviews, job descriptions, and LinkedIn posts into Claude and ask it to identify the most common pain points for a specific persona. It’s excellent at synthesising large amounts of text and pulling out patterns. But you still need to decide which problems to lead with. AI can surface 15 problems. Your job is to pick the two or three that are most likely to get a response from the type of company you’re targeting right now.
Step 5: Build your campaign hypothesis
Now you bring it all together into a campaign plan.
I use a simple table for this with my clients. It’s one document that forces you to think through the entire campaign before you write a single word of copy. Here’s what goes in it:
The ICP segment this campaign targets. The specific persona within that segment. The problems and challenges you’re going to reference. The trigger or signal that makes this timely (if applicable). The hypothesis you’re testing. The sequence structure (how many emails, what gaps, what channels). And the success criteria (what does “working” look like?).
The hypothesis part is important. Every campaign is a test. You’re saying: “I believe that [persona] at [ICP segment] cares about [problem] right now, and if I reach them with [message], they’ll respond.” That’s your hypothesis.
If you can’t articulate the hypothesis clearly, you’re not ready to build the campaign. Go back to steps three and four.
One thing I want to flag here: you’ll need more than one type of campaign running at the same time. I typically build three types with my clients.
Signal-based campaigns target companies showing specific buying signals (new leader, funding round, hiring activity). These are small volume but high relevance. The challenge is you’ll typically only get 5-10 qualified companies per month from a list of 1,000. That’s not enough pipeline on its own.
Volume-based campaigns target the broader ICP with strong, relevant messaging. This is your pipeline engine. It fills the gaps that signals can’t.
Marketing-led campaigns tie outreach to an event, webinar, content launch, or partnership. They give you a reason to reach out that isn’t purely cold.
Most teams only run one type. That’s why pipeline is inconsistent.
Step 6: Build your list
With the campaign plan in hand, you know exactly who you’re looking for. Now go find them.
List building is a whole discipline in itself, and I won’t cover every detail here. But a few things that matter:
Quality over quantity. A list of 200 well-researched, verified contacts who genuinely match your ICP will outperform a list of 2,000 scraped from a database with no validation. Every time.
Verify before you send. “Verified” in Apollo or ZoomInfo doesn’t always mean the email is actually deliverable. Run your list through a dedicated verification tool before loading it into your sequencer. Bounced emails destroy deliverability, and once your domain reputation tanks, it takes weeks to recover.
Suppress properly. Check your list against existing customers, active pipeline, and anyone currently in another live campaign. I’ve seen teams accidentally email their biggest customer’s CEO with a cold prospecting message. It happens more often than you’d think.
Check the angle. Every contact on the list should support the campaign hypothesis. If the campaign targets Heads of IT, you can’t have facilities managers in the sequence. This sounds obvious, but it’s one of the most common errors in real outbound programmes.
Do an enrichment sanity check. Does the data feel current? Are the company details plausible? Are people still in the roles your data says they’re in? People change jobs constantly. A list that’s even three months old can have 15-20% of contacts in different roles.
Step 7: Write the copy
Now, and only now, you write the emails.
If you’ve done steps one through six properly, the copy almost writes itself. You know who you’re writing to. You know their problems. You know what makes this timely. You know what hypothesis you’re testing. All the hard thinking is done.
For cold email sequences, I recommend three emails maximum (four if you have a strong reason for a fourth). Here’s how I structure them:
Email one focuses on problem number one. The biggest, most urgent problem you identified for this persona. Keep it under 75 words. Lowercase subject line. No pitch, no demo request. Just acknowledge the problem and suggest you might have a perspective worth hearing. The goal is a reply, not a meeting.
Email two relates to the first email but approaches it from a different angle. Maybe it’s a different consequence of the same problem, or a specific example that makes the problem feel more concrete. Still short. Still focused on them, not you.
Email three introduces a different problem entirely. If email one was about not booking enough meetings, email three might be about not closing the meetings they do book. This covers the scenario where problem one didn’t resonate but problem two might.
A few copy principles that matter: never open with “I noticed you stepped into a new role” or anything that sounds like every other AI-generated cold email. Never ask for 30 minutes in the first email. Never use fake personalisation (referencing their university, local restaurants, or hobbies). Make the message relevant to the problem they’re facing, and the personalisation takes care of itself.
Yes, AI can generate your emails now. And they’ll be 70-80% there. But that last 20% (making it sound like you, cutting the filler, sharpening the opening line) is the difference between an email that gets a reply and one that gets archived. Use AI to get past the blank page, then edit like a human.
Here’s the thing, though. If you’ve done steps one through six properly, AI becomes incredibly powerful at this stage. Because you’ve already built the context layer. You can load your ICP definition, your persona profiles, your mapped problems, your campaign hypothesis, and your list criteria into Claude and say “write me a three-email sequence for this persona based on this campaign plan.” The output will be dramatically better than if you’d just opened a chat and said “write me a cold email for a Head of Legal.”
The seven steps aren’t just a planning exercise. They’re the context that makes AI actually useful. Without them, AI is guessing. With them, AI has everything it needs to generate relevant, specific copy that you can then refine in 30 seconds.
Most people open ChatGPT and say “write me a cold email.” That’s step seven without steps one through six. The output is generic because the input is generic.
After you launch: the diagnostic framework
You’ve launched the campaign. Now what?
Before you rewrite anything, figure out which problem you actually have. The fix is different for each one.
No opens? That’s a deliverability problem. Your emails aren’t reaching the inbox. Check your domain reputation, your sending volume, your warmup settings, and whether you’re landing in spam.
Opens but no replies? Your subject line worked but the message didn’t give them a reason to respond. Look at the copy. Is the problem you’re leading with actually relevant to this persona? Is the email too long? Is the CTA too aggressive?
Replies but negative or neutral? You’re either saying the wrong thing or saying it to the wrong people. Go back to the campaign plan. Check whether the persona is right. Check whether the problem is real. Sometimes the issue isn’t the message, it’s the audience.
Give each campaign enough time to prove itself before making changes. You need a reasonable sample size. Changing your subject line after 30 sends and two opens isn’t testing. It’s guessing. Let a campaign run through at least a few hundred contacts before drawing conclusions.
The order matters
I keep coming back to this because it’s the single biggest point.
Most teams rush to the copy. They think cold email is about writing good emails. It’s not. Cold email is about sending the right message to the right person at the right time. The copy is the last piece of that puzzle, not the first.
And here’s what’s changed with AI. The teams that do these seven steps in order now have a massive advantage. Because every step creates context. Your TAM estimate, your ICP definition, your persona profiles, your problem mapping, your campaign hypothesis. All of that context, when loaded into AI, turns a generic tool into something that genuinely understands your business and your buyers.
The teams that skip the foundations and go straight to “AI, write me a cold email” will keep getting generic output and wonder why AI isn’t working for them. The teams that build the context layer first will use AI to execute faster and better than they ever could manually.
Do these seven steps in order. Each one feeds the next. If you skip the foundational work, no amount of clever copywriting (or clever AI) will save you.
And if you’ve done the foundational work properly, the copy doesn’t need to be clever at all. It just needs to be relevant.
That’s what makes it work.