Writing the email is the smallest job AI can do in an outbound programme. The bigger wins sit in the work around it: mapping who to talk to, finding them across several databases, deciding which job titles actually buy, and reading the results afterwards. This is how we use Claude Code for cold email at Danish Lead Co., across work for 110 B2B teams that has produced more than 10,000 sales conversations and over $30M in attributed revenue.
Everything below comes from live client campaigns shown in the video, with the actual numbers. If you only use AI to write first lines, this is the rest of the opportunity.
What Is Claude Code Actually Useful for in Cold Email?
Claude Code is most useful for the manual research and data work that used to take a team days, not for writing the copy itself. It is an AI coding agent that runs on your own machine, reads and writes files, and connects to the tools you already use through MCP servers or APIs. That last part is what makes it different from a chat window: it can pull a list from a database, verify the emails, score every contact and write the results to your own database in one conversation.
Mapping a client's market. Turning onboarding answers and a client's website into a structured table of verticals, proof points and owners.
Building lists across several databases. Running a waterfall rather than trusting one provider.
Scoring who to contact. Judging thousands of job titles for how likely each person is to make the buying decision.
Research-based personalisation. Writing specific ideas for each company from its own website.
Analysing results. Answering questions about which campaigns, lists and email variants produced meetings and deals.
How Do You Personalise One Campaign Across Dozens of Verticals?
You map the verticals once in a table, attach real proof to each, and let Claude Code join that table to every contact as custom variables. It replaces building a separate small campaign for every niche, which is slow and tells you little about what works.
The example is TobinLeff, an M&A advisor to marketing agencies. Their market splits into many agency types (PR, digital marketing, PPC, Meta ads and more), and they had strict criteria on size, revenue and EBITDA. Building a separate campaign per agency type would have been a large job before we knew which angle performed.
So we built one Google Sheets table instead:
Map the verticals. Start from the client's onboarding form and meeting, then verify each vertical against what their own website actually says.
Attach proof to each vertical. For an M&A firm that means past transactions: who was sold, merged or acquired, and by whom.
Attach an owner. TobinLeff has partners with different sector experience, so each vertical got the right partner name.
Join it to every contact. Claude Code matched each lead to its vertical, transaction and partner as custom variables. That step used to take a day or two. It took about half an hour.
Clean the company names. "Nike LLC" becomes "Nike", because a raw legal name in an email reads as automated.
The email then names the prospect's vertical, cites a relevant transaction and introduces the partner who would actually run the deal. It is longer than most cold email, and some longer variants performed even better, because it is written in the language this buyer uses. In about 45 days the campaign produced more than 99 positive replies, from agencies that were actually growing and profitable, and the booking rate reached 37%.
If you do not know your buyer's language yet, that is fine, it just takes more sends to find it. We shorten that by researching it up front, scraping forums such as Reddit and review sites for how buyers describe the problem in their own words. Our note on why personalisation beats volume covers the trade-off in more depth.
How Do You Build a Cold Email List With a Database Waterfall?
You find the companies first using whichever source covers that market best, then find contacts at those companies across several providers, then verify every email before it is sent. No single database covers every niche, and the people who are easiest to find are the ones receiving the most cold email.
Step
Tools we use
What it solves
Find companies (common markets)
AI Ark, GetLeads
Broad coverage with firmographic filters
Find companies (hard niches, e.g. hotels)
DiscoLike
Lookalike discovery where keyword filters fail
Find companies (local businesses)
Outscraper (Google Maps)
Businesses missing from B2B databases
Find contacts and emails
GetLeads, Smartlead's database
As many contacts per company as possible
Verify emails
MillionVerifier
Removes invalid addresses before sending
Verify catch-all domains
BounceBan
Rescues addresses other verifiers cannot judge
Fill the gaps
Additional email finders
The 20 to 30 percent left unverified
Most of these tools have an MCP server or an API, so you connect them to Claude Code directly. Search for the tool's name plus "MCP" and the setup instructions usually come up first. Once connected, building a list is a conversation: describe the market, the size criteria and which tools to use, check the output, and push back where it is wrong.
A recent example: a commercial cleaning company list, filtered to roughly $5M to $100M in revenue. The first pass came back thin, so we asked Claude Code to try DiscoLike, and the list doubled. Expect that back and forth.
Two practical points:
Watch the spend. Per-record databases get expensive fast if the agent pulls a huge list you did not need. GetLeads is a sensible place to start because it offers an unlimited plan, which caps the downside.
Verify in between. Typically 70 to 80 percent of found emails verify as valid. The remainder go to other finders rather than into the campaign.
We log every step to Supabase, so every contact, verification result and send is a row we can query. If the agent makes a mistake, the data shows it. Our 2026 cold email data explains why list quality now matters more than anything downstream of it.
How Do You Find the Right Decision Maker When Job Titles Are Inconsistent?
Pull every contact at each qualified company, then have AI score each title for how likely that person is to make the purchase, and contact only the top of the range. This works when titles are too varied for keyword filters to catch.
Deltex sells duvets and bedding to hotels across several European countries. Titles vary by country, language and hotel size, and there is no standard title for "the person who buys the duvets". Guessing titles up front would have missed most of the right people.
What we did instead:
Build the company list first using DiscoLike, AI Ark and GetLeads.
Run every company through an ICP checker to confirm it was genuinely a hotel or hospitality group of the right size.
Pull every contact at those companies through GetLeads and Smartlead, even at very large groups.
Score every title from 0 to 100 for how likely that person is to decide on buying bedding, given the hotel's size and description and what we were selling.
Contact the top-scoring slice first. Above roughly 85 the scores held up well, so we started with the best 15 percent.
That took the list from around 500 contacts to more than 2,000, and the contacts it found are the ones other senders miss, because they cannot be reached with keyword filters. You can see what that programme produced in the Deltex case study.
What Does Research-Based AI Personalisation Look Like?
It gives the prospect something specific and useful about their own business, generated from their own website. A first line about someone's university does nothing for an email that is about their hotel.
For a client that designs content for hotel screens, the system scraped each hotel's website for things like its restaurant name, spa or happy hour offer, and wrote three concrete ideas for what that hotel could put on its screens. Emails per positive reply halved, so the same volume produced roughly twice the positive replies, and the booking rate rose as well. People are more willing to talk when it is obvious you did the work.
Our copywriting rule is simple: make it as relevant as you can, name one specific problem they will recognise immediately, and keep the call to action easy to say yes to. For how to test variants against each other, see A/B testing cold email copy with AI insights.
How Do You Use Claude Code to Analyse Campaign Results?
You track every send, reply and meeting in your own database, then ask Claude Code questions about it in plain English. The tracking is the hard part. Without it there is nothing to analyse.
We store every email sent, every reply (categorised by AI) and every meeting booked in Supabase, and for most clients the CRM syncs back so we can see how deals progress. Then we ask things like "break down this client's last three campaigns: which variants produced positive replies, which produced meetings, and why?" The metrics we optimise, in order:
Emails per positive reply, the first signal that a list and message work.
Emails per meeting booked, because a positive reply that never books is not worth much.
Pipeline progress and closed deals, the only number that actually pays.
Optimising on positive replies alone is a trap. It is easy to write an email that gets lots of interested replies and never gets anywhere. Our B2B outbound benchmarks give ranges to compare against.
What Happens After the Reply?
Speed and booking mechanics decide whether a positive reply becomes a meeting. With AI drafting replies we can respond within five minutes, often before the prospect has left their inbox, and that alone lifts booking rates noticeably. For TobinLeff the reply also routes to the right partner's own calendar, which is part of how that campaign reached a 37% booking rate.
The hard part is not the speed, it is dialling in how the replies sound, what they say to book the meeting and how they follow up. We have also spent two months testing whether to offer specific meeting times or a calendar link. The difference is small so far, around 5 percent, but specific times linked to the calendar are performing best, which matches what we have seen before. Follow-up still matters too: our data on how many follow-up emails to send shows where replies actually land.
Conclusion
Claude Code for cold email pays off in the work around the copy. It lets a small team map dozens of verticals in an afternoon, run multi-database list waterfalls, score thousands of job titles, write research-based ideas for each company, and interrogate results all the way to closed deals. None of it replaces knowing your buyer or measuring the right thing. It removes the manual hours that used to stop teams doing either properly.
Key Terms Glossary
Claude Code:Anthropic's AI coding agent, which runs locally, reads and writes files and connects to external tools, making it suited to data-heavy outbound work.
MCP (Model Context Protocol):A standard way to connect an AI agent to an external tool, giving it both access and instructions for how to use that tool.
Database waterfall:Querying several data providers in sequence so that contacts one provider misses are found by the next.
Catch-all domain:An email domain that accepts mail for any address, so standard verifiers cannot confirm whether a specific mailbox exists.
ICP checker:An automated check that each company on a list genuinely matches the ideal customer profile before anyone is contacted.
Emails per positive reply:The number of emails sent for each interested response, the primary efficiency metric for a cold email campaign.
Video Transcript
Most companies are still running cold email outreach the same way they did years ago, building huge lists and sending generic messages that get ignored. But AI coding tools like Claude Code are changing how outbound teams can research prospects, personalize campaigns, and build repeatable systems. The problem is that most people only really use AI for writing emails, which is really the smallest part of the opportunity. So in this video, I'll show you how we use Claude Code inside a cold email workflow to find better leads, automate manual tasks and create campaigns designed to generate real sales conversations. So for context, at Danish Lead Co, we've run outbound for 110 B2B teams, generated more than 10,000 real sales conversations, and that has led to more than $30 million in attributed revenue. So it's not exactly theory. I'm going to show you real campaigns on my screen exactly how they're built up and why they perform. So knowing that AI can improve your cold email workflow is one thing, but building a system that consistently finds the right prospects and creates qualified conversations requires a lot of testing and execution. So if you want to see how my team and I help B2B companies build predictable outbound systems, check out the first link in the description. And make sure to check out the next videos because we'll be breaking down more real campaigns, strategies, and systems that help companies generate a consistent pipeline. Most teams that we talk to just start by building a huge database or try to build a huge database. Then they pretty much just send one email out to everyone. Like now you can actually do this without having you know 50 people writing first liners anymore because of AI. An example of a client where we did this and it worked really well was or is TobinLeff They are a M&A advisor for agencies actually. So essentially they work with pretty much any type of agency or business within the marketing industry. So there's actually a ton of like verticals or types of agencies within that. So for example, it's like anything from PR to digital marketing or PPC agency. The problem with this was that they still had relatively strict criteria of like the size of these companies. So like the revenue and EBITDA and also just like the types. So it had to actually fit well and it had to actually be within the verticals that they also worked within. That meant that if we were to like build a small campaign for only PR agencies and another campaign for only PPC agencies or Meta ads agencies or whatever and like divided that small in the first go, it would be a like massive job and we wouldn't know what would actually perform. So what we did instead was we essentially built this Google Sheets table where first we started by mapping out all the verticals that they were working in. So we did that first of all by the information that they gave us in the onboarding form and on the onboarding meeting but second of all actually verifying all of that on their website and what was on their website. Then we mapped that to case studies on their website. So like for an M&A firm it's not really called case studies but it's essentially like transactions that they've helped facilitate. So then we got all the names of those and then we got all the like sort of the descriptions because some of them are a sale, some of them are a merger and then described or like writing which company uh bought them or which company acquired them and which company they merged with or whatever it might be. Then as well we also had this column which is partner name because TobinLeff has a lot of different partners with a lot of different sort of experiences in the different verticals. So we mapped those out into each vertical as well. And each of them then also had a partner. So this would have been a pretty massive job in the past without Claude. And the next step from here of actually mapping this to each lead so that we have these custom variables that we can then use in the campaign would have like taken a day or two in the past but took I don't know half an hour now with Claude Code. Right. So I'm just going to show one example here of the scripts where we actually use all of these custom variables. It's called essentially here. Then we input these variables. So I was researching vertical. So I was researching digital marketing firms and came across company name. All the company names we spent quite a bit of effort on actually cleaning the company names as well. So it doesn't say like Nike LLC, but it actually just says Nike. We recently helped case study company navigates its transaction description. So we recently helped X case study company. So we we recently helped Danish Lead Co navigate its transaction to balcon not a real thing by the way and that you know we've helped a bunch of others in the Malcomm space and we have a really strong team then sort of introducing the the specific partner here. So this has like a decent amount of verticals but at the same time we've spent the time actually writing the basis of the script in the language that we know these people use and that they will understand. And this is not exactly like the shortest email that we've ever written and some of the variance in this was actually a lot longer as well and performed even better. So this is just a good example as well of knowing your customer is actually one of the most important parts. And if you don't that's fair as well. It's just going to mean that it's going to take like longer and more emails to test and actually see what resonates with them like how do they describe their problems etc. So what we do for clients as well is we do a lot of uh research initially with AI like scraping forums like Reddit and review sites and things like this to actually get the language of the buyers as well or the the ICP. So this is just to show how this actually performed in I think just like 45 days or so. We've generated more than 99 positive replies and it's positive replies like this one from actual companies that like are actually growing and actually has positive EBITDA and all of these things meeting book rate and this view is not correct. Um I'll show you that later. Right. So the next thing that we use Claude Code for quite a lot is building lead lists and sort of doing waterfalls where we not only rely on one database to build a lead list or build a contact. So we'll use a few different tools for this. for actually first mapping the like the companies that we can actually reach out to and that fits the criteria etc. We will often times use AI Ark sometimes GetLeads oftent times if it's like a a niche that's not really well like hotels can actually be super hard to find then we use tools like DiscoLike if it's more like local businesses then we'll use Outscraper which essentially scrapes Google maps so when you've built the list of the companies the next step is then to find as many contacts from those companies as possible all of these tools will either have an MCP or an API that you can connect directly to Claude Code . So literally if you just search for the tool name and MCP then it'll come up and then it'll show you how to actually connect it into your Claude Code or Codex or whatever you use. When doing that and MCP essentially just gives all of the information and the access but all the information about how to use the tool and then like the access like the API access to actually use the tool as well. I think I would probably recommend starting out with tools like GetLeads because they have like a an unlimited plan so 497 so it can't go like completely crazy whereas most other databases like a they like it can get expensive if your Claude like goes crazy and like pulls out a huge list or whatever. When you've then done that then it's literally just chatting with your Claude Code. So to actually show you an example, I was building a list for a potential client that we essentially have our Claude Code connected to all of these different tools including then verification tools as well. And that's the the next thing when you you start to actually find the contacts as well is actually verifying the emails in between. So you essentially pull out all the relevant contacts from for example GetLeads verify all of them with MillionVerifier or like we use MillionVerifier and then all the catch-all emails from that we verify with a BounceBan which is like does a great job of actually verifying catch-all emails. So that will give you like typically it's like 70 to 80% of like all the emails that you find that are actually valid that you can use. But the rest of the 20% those contacts you will then like won't actually use other contact or email finders to try and find different uh emails for this is an example of the list that I built yesterday. So I essentially just explained to it the company what they wanted to get done that they currently have 300. We wanted to find commercial cleaning companies which are not exactly the easiest to find either. They wanted to have like a certain size uh which I think was like 5 to 100 million in revenue or something like that. And I just told it which of the tools I should use. So used yeah these different tools. Did that first saw the output from that was not exactly what I wanted it to be. So then I told it to try and find more companies that fit through DiscoLike and then we ended up with like twice as many. The process from that is it's like it can be a bit back and forth, but it's really thinking about okay, where are you first of all going to find the companies? Which databases are most likely going to have it? Then using those if you want to just max out the time, which I recommend unless you have like unless you're targeting every home services company in the US, then probably wouldn't cuz that's going to be quite expensive pretty quickly. Yeah. Then I would recommend using GetLeads, Smartlead. They have their own database as well. Works great for finding emails and contacts because Smartlead has a lot of data. But then it's essentially just building it in. So Claude runs this waterfall and finds as many contacts as you can. We've built it into our like platform so that it runs in here. We've connected it up to Supabase, which I would highly recommend as well. So you literally just track everything and you have all the data points. So if Claude does something wrong, then you have everything in Supabase as well. And it works extremely fast and extremely well, right? So then there's another example of like actually the hospitality industry for client of ours Deltex where it's actually very difficult to or it was very difficult to know from each of those hospitality groups or hotels which titles were relevant to reach out to both because it was in different like across different European countries. It was like in different languages but also because it very much depends on the size of the hotel and they literally just have the most random titles ever. So it's very hard to know like we needed to know who's likely going to be a decision maker for like buying duvets for the hotel. What we did there instead of trying to find the titles before building the lists. So like with keywords in the different databases, we first built the list of companies with DiscoLike and I think we used a and GetLeads for that as well. But first built out as many companies as we could. Then all of those companies we ran through an ICP checker to make sure all of them were actually like a fit in terms of size and that they were actually you know a hospitality group or a hotel. Then all of those we used GetLeads and Smartlead to find contacts and emails from those companies and we literally just pulled out all the contacts. So even if a you know company had like 10,000 employees we just pulled out all of them and then what we did was we had an AI ICP checker that we've built into the app but you can do like most of it just directly with Claude Code as well. We had that look through each and every one of those titles and determined is this title likely to be a decision maker for something like buying duvets for the hotel. And that then looked at like the hotel, the description of the hotel, how big the hotel was and then the title and you know what we were selling and then it gave it a score of like 0 to 100 of what like how likely it actually was. Then afterwards we could then go in and then we could see okay like I think it was after like score like 85 then it started to be a bit like weird. So we started with the first like 15% that scored the the best and that gave us a lot more contacts that were a lot more accurate and those contacts are also typically the contacts that get less emails because that list was a lot higher harder to build than you know just using the keyword filters in the uh databases. In this specific example, we went from like 500 contacts or something like that to like more than 2,000 contacts. If you are new to the channel, just keep watching. But if you've been watching my videos for a while and still haven't taken the first step toward an outbound system that brings in sales meetings every week, book a call with us below. We'll walk you through how it would work for you. Then another example of how we use Claude Code in a campaign where we actually had like the AI freestyle a little bit more. We didn't just do like variables and had it like match each variable to the company, but instead we actually had AI write three ideas like for the hotels where we're reaching out to hotels and what my screens did was they had these sold either like both the screens but at least the designs of what they would put on the screens because most hotels doesn't actually have anything on there or it's not really optimized. first like we built a list similar to the other hotel list and instead of writing you know hey I saw you went to Copenhagen business school or whatever that's not relevant for what we're reaching out with we actually reached out and gave them three ideas of you know how like the signs that we could add to their screens at the hotel so we did that by scraping their website. So getting the context from on there because typically they would write on there like the name of the restaurant or that they had you know they have happy hour uh two for one or whatever it might be. The AI took that specific information and then you used that to write three ideas which could for example be we made a design where we showed like the discount or like the happy hour for the restaurant to get more people in there. those campaigns we went from that not performing very well um to I think it's doubled or tripled the performance of that campaign. Yeah. So here you can see it actually doubled the performance. So this is emails per positive reply which is the like one of the primary metrics that we look at and the positive replies and actually increase the booking rate as well because people are you know they're just more likely to actually want to do anything if if they feel like you know you actually did some work. um in the email lab. So like our overall philosophy when it comes to copywriting and and like cold emails is like the more relevant you can make it the better. So show that you've done some sort of homework or that this is super relevant to their specific situation or pain points or whatever it might be. Then just have a specific problem that you're solving that they understand very easily than just a super simple CTA. Then the next thing that you can do with Claude Code as well is analyzing the data. But to be able to analyze the data, you have to first track all the data. So we do that in Supabase. Again, we've just connected Supabase to Claude Code and like to our app. So we track every email that we ever sent, every reply and we categorize every reply with AI. We track like every meeting that's been booked uh at least for most clients that clients essentially set up a or a HubSpot uh calendar and then we're able to see like when a lead has actually booked a meeting as well. and we're able to book it in for them as well. So, what we do look at is emails per positive reply. We look at emails per meeting booked and then we look at the like actual pipeline. So, the actual closed deals eventually, but at least how they're progressing through the pipeline. All of this we can do because we actually have Claude Code set up as well. So we can ask specific questions about like breaking down a specific campaign or the last three campaigns for our client, which email variants like which of the copies actually performed the best and and why do we think all the way down to meetings booked? So we're not just optimizing based on, you know, where we got the positive replies because it's very easy to like write an email that gets a lot of positive replies but never gets uh you know anywhere from the positive reply. So here's an example of TobinLeff where the booking rate is actually very high. at 37% and part of the reason for this is because we use we call it our Gia because Gia is on our team. So with this for TobinLeff because they had all of these different partners and depending on the vertical of the campaign we wanted to send it to a specific partner and book it into a specific calendar for that partner. That took a little bit of like building as well, but we were able to also do that so that you know we can mention a specific partner and the meeting was then booked for that specific partner as well and that we could track it as well. TobinLeff are doing a great job of actually updating their CRM but essentially they update their HubSpot and then that just syncs to to our CRM. So we have all of that data as well and we can see which leads are actually going somewhere. So when you use AI you are suddenly able to actually like reply within 5 minutes. So before the person or the the lead has actually left their email their inbox then they get the response back and that in itself increases booking rates quite drastically. The very important part and what has also taken a lot of time is actually dialing in the sort of the language and how it responds and uh how it books in the meetings and what it says to book in the meetings and how it follows up and all of these things. So, another thing that this has actually enabled us to do is test sending uh links and like specific times and dates where we could have the meeting. So, the AI does that or where we just send a calendar link. And we've actually been testing that for the last 2 months now. The difference is very low. It's only like 5% difference in the booking rate so far. But it looks like sending the specific times with like directly linked to the Calendly is what performs the best. um which is what we we had seen in the past and what logic um sort of points you toward as well. Right? So knowing that AI can improve your cold email workflow is one thing but actually building a system that consistently finds the right prospects and creates qualified conversations requires a lot of testing and execution. So, if you want to see how my team and I help B2B companies build predictable outbound systems, check out the first link in the description. And make sure to check out the next videos because we'll be breaking down more real campaigns, strategies, and systems that help companies generate consistent pipeline.
Want an Outbound System Built This Way?
If you want this kind of research, list building and analysis behind your own outbound, book a call with Danish Lead Co. We will look at your market, how your current lists are built and what you track today, and show you where a system like this would change the numbers. You can also see the results it produced in our case studies and how the whole system fits together in how we book 400+ meetings a month.
It can, but that is the least valuable use. The bigger gains come from research and data work: mapping a market, building and verifying lists, scoring which contacts can buy, and analysing what produced meetings and deals.
What is the best way to use Claude Code for cold email?
Connect it to your list, verification and tracking tools through MCP or API, then use it for the manual work around the copy: building database waterfalls, joining proof points to contacts as custom variables, scoring job titles and querying campaign results.
Which tools can you connect to Claude Code for outbound?
Most list and verification tools offer an MCP server or an API. In our workflow that includes AI Ark, GetLeads, DiscoLike, Outscraper, Smartlead, MillionVerifier, BounceBan and Supabase.
How do you stop AI list building from getting expensive?
Start with a database that has an unlimited or flat-rate plan, tell the agent the exact size criteria before it pulls anything, and review a sample before it builds the full list. Per-record pricing gets expensive quickly if an agent pulls more than you need.
How many found emails are usually valid?
In our experience 70 to 80 percent of found emails verify as valid after running them through a standard verifier and then a catch-all verifier. The rest go to other email finders rather than into the campaign.
How do you find decision makers when job titles vary by country?
Pull every contact at each qualified company, then use AI to score every title from 0 to 100 for buying authority given the company and the offer. Contact the highest-scoring slice first and widen from there.
Does AI personalisation improve cold email reply rates?
When it gives the prospect something specific and useful from their own business, yes. On one hotel campaign, writing three tailored ideas per hotel halved emails per positive reply. Generic first lines about someone's background do little.
Is it better to send meeting times or a calendar link?
In two months of testing the difference has been small, around 5 percent, with specific times linked to the calendar performing slightly better. Replying fast matters more: answering within five minutes lifts booking rates noticeably.
