Learn / Lead lists
Apollo + Claude Code: scored lead lists in one prompt
The last lead list took less than 10 seconds to build. If you've ever built one the old way, you know that shouldn't be possible, because normally it takes hours: log into a tool, set every filter by hand, export, re-import, enrich, score everything in a separate sheet. We send millions of cold emails for clients every month. There wouldn't be enough hours in the day to do that by hand.
So we wired Claude Code into Apollo and turned the whole thing into a system you set up once. After that, every list you ever build is one prompt. This page is the full written version of that setup, step by step.
Why hand-built lists stop scaling
Most people treat targeting as messing around with filters until it feels right, which is how you end up with a huge list of people who were never going to answer. When your targeting is a guess, every email you send is a guess too, and you pay for it in burned domains and dead reply rates. The fix isn't better guessing. It's basing the profile on the people who already bought from you, writing that down once, and making every list trace back to it.
Step 1: turn your ICP into a file Claude can read
You could list the traits yourself. The easier way is to have Claude interview you:
Interview me about my ideal customer profile one question at a time to help me build a targeted lead list for outbound. Once we're done, save everything into an MD file called icp.md.
It asks 4 or 5 questions (company types, exclusions, triggers, geography, the core decision-maker), compiles a definitive ICP, and saves it to icp.md. That file matters because everything after this step runs on it: it's the brain of the whole system, and you only write it once.
Step 2: connect Apollo and let Claude pull the list
Grab an API key from Apollo's developer portal (API keys, create new), drop it into Claude Code, and say "let's take a stab at it." Claude reads icp.md, translates it into an Apollo search, and iterates: testing queries, checking results against the profile, and adjusting.
In the training run, that produced a sample of 50 contacts across 25 mid-size US hospitals, two per hospital across 22 states, with keyword inclusions and name-based exclusion filters that stripped out vendors, IT firms, and insurance brokers on its own. Name, title, company, enriched work email, location, LinkedIn URL, domain. No tab juggling, no CSV shuffle.
Step 3: score every lead against the ICP
This is the step most people skip. They pull a list and treat every prospect as the same priority, so a perfect-fit buyer and a barely-relevant one get the same effort. Ask Claude to score the list against icp.md instead, 1 to 10, weighted by decision-making authority. In the hospital example, whoever owns the hiring plan scores highest, with clinical executives weighted as influencers.
The score isn't decoration; it routes channels. Going multi-channel, the 10s get email, LinkedIn, and phone. The 1s get email only. That's what turns a pile of contacts into a prioritized hit list you can actually work through.
Step 4: verify, then ship it to your sequencer
Most of the list comes back with emails; a waterfall enrichment can chase the stragglers. Then run everything through a verification tool like MillionVerifierbefore a single send, and upload straight to your sequencer. Claude Code can drive this step too. Skipping verification is how good lists burn good domains, so don't.
Step 5: turn the whole thing into a one-command skill
Everything above ran as separate prompts because it was being demonstrated. The last step collapses it: ask Claude, in plain language, to create a skill called Apollo Lead List Builder that reads icp.md, pulls matching leads from Apollo, scores each one against the ICP, and outputs a Google Sheet sorted by score.
A skill is just a saved playbook: the chain of events written down once, in order. From then on, a fresh scored list is a one-word command. That's the 10 seconds from the opening line.
The expensive part of list building was always the hours. This removes the hours.
A great list doesn't book calls on its own. It needs an offer a stranger can't ignore and a script that doesn't read like a pitch. If you're still choosing a database, start with Apollo vs Prospeo; if enrichment and personalization at scale is the gap, that's the Clay workflow.
tl;dr: ICP in a file, Apollo behind an API key, Claude doing the pulling and scoring, and a skill so it all runs on one command. Set up once, reuse forever.
FAQ
Do I need a paid Apollo plan to connect Claude Code?
You need an Apollo account with API access: the key comes from Apollo's developer portal under API keys. Plans and API limits change, so check Apollo's current pricing page for what your tier includes before you build the workflow around it.
What is an icp.md file?
A plain text file describing your ideal customer profile: company types, exclusions, triggers, geography, and the decision-makers you want. You build it once by having Claude interview you, and it becomes the brain every future list is pulled and scored against. MD is just Markdown; Claude can read it any time.
Can I use a database other than Apollo with Claude Code?
Yes. The pattern is the same with any database that exposes an API: hand Claude the key, let it translate your ICP file into a search, and iterate. Apollo is the most common starting point; if you're weighing alternatives, see our Apollo vs Prospeo comparison.
How does Claude score the leads?
Against the icp.md file, on a 1 to 10 scale, weighted by decision-making authority. In the hospital example from the training, the owner of the hiring plan scores highest and clinical executives score as influencers. The score then routes channels: 10s get email, LinkedIn, and phone; 1s get email only.
Is this the same thing as using Clay?
Different job. This system sources and scores a list from one prompt; Clay earns its keep on enrichment and personalization at scale once you have the list. Plenty of teams run both: Apollo plus Claude Code to build and rank, Clay to enrich. Our Clay workflow breakdown covers that half.
If you'd rather never touch any of this, we run the entire cold outbound system (infrastructure, lists, copy, and qualifying the replies) for our clients: book a call and see if it fits. Or keep it DIY: the full video is on YouTube, and every training by topic is on the trainings hub.
PS - verify before you send, even when the list looks clean. The waterfall finds emails; it doesn't promise they're deliverable.