Look, I’ve been doing cold outreach and automated marketing systems for a long time and I always had this slight little problem of building super massive lead lists that looked great and did absolutely nothing.
This is what I did initially.
I scraped a directory like Apollo for example, ended up with a few thousand companies sitting in a file, and felt pretty good about the number honestly.
Then, when it came to the outreach, I’d start messaging and almost nobody would reply. I’d go back and scrape more, because when nobody answers you, the obvious thing to do is to message more people, right? Wrong!
It was so frustrating. The file kept growing and the replies stayed exactly where they were.
So I sat down with the last file and counted it properly. Opened it as a spreadsheet. Counted the rows. Counted how many of them had an actual person in them.
Here’s what that list really did. I scraped 501 companies and I could send something to 113 of them. The other 388 I could not have messaged even if I wanted to, most of them had no human being in the row at all.
To make the long story short, there are 3 steps here, they take 67 minutes in total (most of it is figuring out who you’re selling to), and none of them require any coding skills. So, please, follow along.
Subscribe to No BS AI. One real copy-pasteable build like this every week.
Step 1: get your ICP done (30 minutes)
It’s easier to look for the ideal buyers when you can picture what your ideal customer or ICP (Ideal Customer Profile) looks like. I know it’s hard but here’s a prompt that will ask you a bunch of questions and will help you narrow down your ICP.
So here’s what to do exactly:
Open a new chat in Claude, or whichever AI agent you use.
Paste the prompt below and answer the questions one by one, in your own words. Rough answers are fine, it pushes back when they’re too vague.
Save the page it hands you back somewhere you can find it, because you’re going to paste it again in step 3.
It’s the foundation for a high quality offer, so, please, make sure to spend some time on this. Paste this to nail down your ICP:
# ICP Calibration Interview
Don't start with industry, company size, or titles. Start with capability, work outward, let evidence decide the category.
## Flow
me → what I'm good at → problem I solve → conditions where it works → companies with those conditions → who buys → filters that work on a lead list
## Interview rules
- One question at a time, wait for the answer before the next
- Don't show the question list upfront
- Adapt each question to the last answer
- Push on vague terms ("small businesses", "decision makers", "companies that need growth") until specific enough to filter a real database
- Separate what I know from experience vs what I'm guessing
- Prefer real examples over opinions about what should work
- If answers contradict, stop and resolve it, don't pick one silently
- No flattery, no manufacturing a niche because it sounds good
- If there isn't enough evidence yet to define an ICP, say so
## Sequence (guide, not a script)
**1. Capability** - What am I consistently better at than people I compete with? Drill into: where the skill came from, what work people already trust me for, what's easy for me but hard for others, what I don't want to sell.
**2. Evidence** - Walk through 5-10 real paid engagements: who bought, what were they buying, what was happening in their business, what changed, how fast, how hard was the sale, what did they pay, would they buy again.
**3. Strongest pattern** - Which customers benefited most, and what was already true about them before I started (existing demand, team, budget, tech, urgency, stage)?
**4. Buying trigger** - What was happening right before they decided they needed me? (stalled revenue, new exec, raised money, channel broke, cost of doing nothing became obvious.)
**5. Pain and economics** - What they said they wanted vs what they needed. What the problem cost them. Why it was worth paying to fix.
**6. Best-fit vs bad-fit** - One bad-fit customer, what made them bad. Compare to the best one.
**7. Buyer** - Who felt the pain, who evaluated, who had budget, who could block, what was their actual title. Don't assume one person plays all roles.
**8. Company characteristics** - Translate the pattern into industry, geography, headcount, revenue, stage, tools, existing capabilities. For each: is it actually predictive, or does it just describe who I happened to see?
**9. Lead-list test** - Which of these can be checked from a CSV alone (industry, size, revenue, tech, job posts)? Which need a person to look at a profile or website?
**10. Falsification** - Strongest evidence this is the ICP. What would prove it wrong. What attractive-looking segment should we deliberately exclude, and why. Rank multiple candidates by evidence strength instead of merging them into something vague.
## Pass/fail bar
- **Specific** - a stranger could look at a company and get a consistent yes/no
- **Evidence-based** - comes from real customer behavior, not ICP theory
- **Causal** - explains why the offer works for them, not just who bought
- **Detectable** - checkable from lead data, a site, or a profile
- **Economically real** - they have both the problem and the money/will to fix it
- **Exclusionary** - tells me who to skip
## Output (one page)
1. **ICP in one sentence** - "We sell to [company type] that [condition], where [precondition], and the buyer is [role]."
2. **Company** - industry/model, geography, headcount, revenue, stage, must-already-have, trigger, problem worth paying for
3. **Person** - buyer, likely titles, champion, budget owner (if different), titles that look relevant but don't buy, what makes them care
4. **Lead filters**
- Tier 1 (CSV/database, cheap): 3-5 disqualifiers, cheapest first
- Tier 2 (website/profile, costs a look): 3-5 disqualifiers, cheapest first
5. **Example** - one near-perfect fit, one that looks like a fit but should be rejected
6. **Confidence** - high-confidence conclusions, reasonable hypotheses, open unknowns, what evidence would improve it most
Step 2: scrape your list (7 minutes)
Ok, so let’s say you crafted your ICP. Now, the tricky question is: how to find the people that align with that ICP? And the answer for most of you is scraping.
Yes, you’ll first need to scrape your list using a tool like Apify for example. Where do you scrape your list from? Well, good question. There are plenty of providers out there like Apollo, ZoomInfo, Clay. The most optimal for me is LinkedIn Sales Nav.
So here’s what to do exactly to get your list:
Go to Apify and create an account, you get $5 of usage for free.
Create an account at one of those platforms (Sales Nav is free for 1 month) and create a lead list there by filtering against your ICP. I strongly recommend starting with Sales Nav because it has high quality leads out of the box and you can also start with 1 month for completely free.
Build the search with the filters straight out of your ICP page, so headcount, country and job title, then save the search and copy the URL out of the address bar.
Then you find a scraper on Apify that scrapes that list from Sales Nav, paste your saved search URL into it as the input, run it and download the result as a CSV. You’re good to go.
Step 3: qualify the list (30 minutes)
Finally, you’ll need to qualify those people. If your list is too big like 5k+ people then you probably need to adjust the filters on Sales Nav and be more specific.
Once you get under 2-3k it’s good. So this is what you do:
Open the CSV with your leads once and just check that the name and the title columns actually have something in them. 358 of my 501 rows had a company and nobody attached to it, and that’s the kind of thing you want to see now rather than after you wrote to them.
Take the prompt below, take your ICP from step 1, take your exported CSV list and attach everything into a fresh Claude session.
And launch everything.
## Context
I'm about to send outreach to a list I scraped, and I want to drop the rows that are
never going to answer before I write anything. My file is attached, and my ICP page
is pasted below it.
## Instructions
Open the file yourself and tell me the exact column names you found in it. Don't
assume them and don't ask me for them.
Then go row by row and drop a row if any of these is true:
1. There's no named person in the row, only a company.
2. <disqualifier 2 from your ICP page, the kind you read off one column>
3. <disqualifier 3 from your ICP page, the kind that needs a profile visit>
Run them in that order and stop at the first one that fails.
## What I want back
Two numbers first: how many rows went in, how many came out.
Then the surviving rows as a new file, with the same columns as the one I gave you.
Then one line per dropped row saying which check killed it, grouped by check.
## My ICP
[paste the page from step 1 here]You’ll get a very nicely filtered and qualified ICP list to outreach to. From a list of 501 people in, I got 113 out, so just over 1 in 5 survived. That is completely fine and reasonable. We do not sell to each and everyone who sounds “sellable”, but only to people who might really need the service.
Subscribe to No BS AI. One real copy-pasteable build like this every week.
Once you know who to message, you need to craft the right outreach message which has to still sound like you.
So if you want that part sorted as well, check this out: Soundcheck.
Talk soon,
Ilya
This is one of those thoughts that grows the longer you sit with it 💭