How CloneICP Finds Your Best Customers' Clones in 60 Seconds

CloneICP uses semantic search to convert a plain-English description of your ideal customer into 20-50 ranked matches, each scored 0-100% with reasoning, in under 60 seconds. Here is how we built it, and what we learned.

The Problem: Boolean Search Is Broken

Every B2B sales team knows the drill. Open LinkedIn Sales Navigator. Build a Boolean string. Add filters for title, company size, industry, location. Scroll through hundreds of profiles. Copy the promising ones into a spreadsheet. Repeat tomorrow.

The average sales rep spends 4+ hours per week on this. The results are mediocre. Boolean search only finds people who match exact keywords. If someone calls themselves "Head of Revenue" instead of "VP Sales," they're invisible. If they work at a company you've never heard of, they don't exist in your filter set.

The deeper problem is structural. Boolean search matches strings. It doesn't understand meaning. You're searching for keywords when what you actually need is to find people who look like your best customers.

The Insight: Search by Meaning, Not Keywords

We kept hearing the same frustration from sales teams: "I know what my ideal customer looks like. I just can't describe it in a way that LinkedIn understands."

That's the gap. Sales reps have rich mental models of their ideal customer profile. They know it when they see it. But translating that intuition into Boolean operators loses all the nuance.

What if you could just... describe the person you're looking for? In plain English?

That was the starting point for CloneICP.

How Semantic People Search Works

Traditional search matches keywords. Semantic search matches meaning. The difference is everything.

When you type "VP of Revenue Operations at a B2B SaaS company who has led sales automation initiatives" into CloneICP, here's what happens:

1. Your description gets processed as a semantic query
2. Neural search scans the open web for people matching the meaning of your description
3. Each match comes back with profile data, context, and source URLs
4. Claude scores every match from 0 to 100%, with written reasoning for each score
5. Results display ranked by relevance, with match explanations

The whole thing takes less than 60 seconds. You get 20-50 scored matches per search.

The scoring is the key. It's not just "here are some people." Each result includes a percentage score and a plain-English explanation of why they matched. "94% match. VP of Revenue Operations at a B2B SaaS company, 200-500 employees. Strong signals: led sales automation initiative, published on outbound strategy, active in RevOps community."

That context turns a list of names into actionable intelligence.

The Stack

We built CloneICP as a standalone product. The technical choices were deliberate:

• Next.js for the frontend and API routes
• Supabase for auth, user data, and credit tracking
• Semantic Search for web-wide people matching
• Claude for scoring and match reasoning
• Stripe for credit-based billing

The breakthrough was neural search — search that understands what you mean, not just what you typed. Most AI search tools are built on traditional search APIs with an LLM wrapper. Neural search works from the ground up on meaning. That's the difference between finding 10 obvious matches and finding 50 matches including people you never would have thought to search for.

What We Learned Building It

Scoring needs transparency. Early versions returned matches without explanations. Users didn't trust the results. The moment we added match reasoning ("Here's why this person scored 87%"), trust went up and so did usage. People need to understand why the AI made a decision, not just see the decision.

Speed is a feature. Our first prototype took 3-4 minutes per search. That felt slow even though it was faster than manual prospecting. We optimized the pipeline to under 60 seconds and engagement doubled. When something feels instant, people use it more.

Credit-based pricing works for search. We considered monthly subscriptions but landed on credits because search usage is spiky. A sales rep might run 20 searches in one day and none for a week. Credits match the actual usage pattern.

The Numbers

The metrics are straightforward:

• 20-50 ranked matches per search
• Under 60 seconds from query to results
• 0-100% scoring with written reasoning for every match
• Hours saved: 4+ hours per week of manual prospecting replaced by a few minutes of searches

For a sales team running 10 searches per week, that's roughly 200-500 scored prospects replacing what used to be a few dozen hand-picked LinkedIn profiles from hours of filtering.

Why This Matters

The shift from keyword search to semantic search changes what's possible. Sales teams don't just find more leads. They find different leads. People they never would have discovered through Boolean operators because they don't match the keywords, but they match the meaning.

One user described it as "finding the customers you didn't know you were looking for." That captures it well.

CloneICP is live at cloneicp.com. The first search is free.

See What This Could Look Like for Your Team

CloneICP solves prospecting. But maybe your sales team's bottleneck is somewhere else. Proposal generation. Competitive analysis. Pipeline reporting. Whatever the repetitive work is, there's probably a pattern worth automating.

Describe your bottleneck and get a free Automation Blueprint that maps the path from manual work to AI-powered workflow. Takes 60 seconds.