If you want to know how to build a cold calling list with Claude, this is the exact process we run at Outbounda. Claude builds every list our reps dial. Recently we booked 22 meetings in three days off a list it built from start to finish.
I’ve been doing outbound for four years and I’ve never been this time efficient. A list that used to take one to two full days now takes a couple of hours. That speed is only worth having because of the checks around it, so those are in here too.
What list building looked like before
You’d open the CRM, click into a company, find the contacts, then go and look each of them up on LinkedIn. An hour of that got you five to ten people ready to contact.
Stretch that across a full list and it was a day or two of someone’s week. Usually a rep’s, which is time they should have spent on the phone.
The tools Claude works with
Claude doesn’t do this alone. It sits on top of tools we were already using:
- TheirStack for the companies. It gives us hiring signals and the technology a company runs.
- Apollo and AI Ark for the contacts inside those companies, with mobile numbers and emails.
- Supabase as the database. It holds our do-not-contact lists and every prospect who’s already in a campaign.
- Salesfinity as the dialer, where an approved list ends up.
Claude pulls the companies, finds the people, checks them against the database and hands back a clean file with everyone who’s left.
Three things Claude never does
Limits first. There are three jobs I don’t hand over:
- Deciding who we target. The ICP is a human call.
- Approving the list. Nothing goes into the dialer until a person has verified the file.
- The call itself.
Everything below sits inside those three rules.
Step 1: Start with a trigger, not headcount
You could build a list off headcount. Every software company with around 200 staff, say. But headcount only tells you how big a company is.
What we filter on is something specific you can check about a company, like the tech they use or who they’re hiring. We call that a trigger, or a signal. Every time we’ve tested it, a trigger has worked better than filtering by company size.
Here’s a real one. We work with a B2B SaaS company that sells into e-commerce stores. Their trigger is any store offering buy now pay later alongside at least one other payment option. Every extra way to pay is another set of payouts, fees and refunds a finance team reconciles by hand at month end. That’s a reason to pick up the phone.
For that client the market came to around 31,000 e-commerce stores. Just over half, 16,000 to 17,000, were running buy now pay later. That’s the part of the total addressable market we call first.
What goes in the prompt
I give Claude three things:
- The ICP. Industry, revenue, the platforms they run on and the job titles we want.
- The trigger. What it is and why it matters to the buyer.
- The rules. The non-negotiables for this list.
The rules are short and blunt. A few from that build: mobile only, never a switchboard or main line. Check the live database, never an export. And every row gets a one-line reason for being on the list, written in plain English.
From there Claude pulls every company with the trigger, matches each one to the ICP, scores it and cleans up the data.
A trigger is only as good as where you look it up
I learned this on an earlier build. I had Claude sort a list into tiers and not a single company made the top two, because my Apollo search hadn’t brought back any hiring data. There was nothing to rank on. That’s why the company pull now comes from TheirStack, and Apollo handles the people.
Step 2: Find three or four people and a mobile for each
Once Claude has every company with the trigger, it pulls three to four people at each one from Apollo or AI Ark.
Every one of them needs a mobile number. For cold calling, that number matters more than anything else on the list. The company’s main line will usually just get you a receptionist.
To get the mobiles we use a waterfall enrichment:
- Claude checks Apollo first.
- If Apollo doesn’t have the number, it tries Wiza.
- If Wiza doesn’t have it, AI Ark or Firmable.
With our current stack we find a number for 80 to 90% of the contacts we go looking for. You’ll need to play around with the order for your own region to get the most coverage.
The benchmark to judge it on is connect rate. Anything over 10% is pretty decent. Across our clients we’re seeing north of 15% on mobiles, and sometimes over 20%.
Step 3: Take off everyone you shouldn’t be calling
A lot of people skip this step. Before a single contact goes into the dialer, four groups come off the list:
- Anyone on the client’s do-not-contact list. Current customers, active deals and pipeline, and any company they don’t want us reaching out to for whatever reason.
- Any company where we’ve already booked a meeting.
- Anyone already in one of our active campaigns.
- Duplicates. The same person on the same list twice.
We also check that nobody going into a new campaign has been touched in the last 60 to 90 days.
All of it lives in Supabase. Every person and company we’ve already touched, the reason they’re suppressed (already a customer, already called, on the client’s list) and the source it came from. It’s a living file, which is why Claude checks it every time it builds a list. We don’t rebuild the whole market every day, because the market doesn’t move that fast. The suppression list does.
Step 4: Check Claude’s work before a rep dials
Claude does all the work I give it. I still never blindly trust what it tells me. When the file comes back, every row has the company, the person, their direct number and why they’re on the list. I check three places where it can go wrong.
The numbers
Claude will sound very confident about a total it didn’t even count. If it says 800 fresh leads, I want the file and the row count that proves it. I recount the rows myself and read through them.
Stale data
It can work off last week’s export, and people you’ve already called end up back on the list. Every build has to check the live database for who’s been called.
The reporting
It might tell you a campaign is performing well because emails got opened, or because LinkedIn connection requests were accepted at a decent rate. The only results that count are replies from real people and meetings that actually happened.
Once all three are checked, the list is approved and goes into Salesfinity. Each client has their own list in there, with new people added every month.
What it adds up to
Claude builds the list. A person checks it before it goes live. What used to take a couple of days takes a couple of hours, and the reps always have good leads ready to dial.
The time we get back goes into the three jobs Claude doesn’t do: choosing who to target, approving what goes out, and making the calls.