Most Ideal Customer Profile (ICP) workshops produce documents nobody uses. You spend three hours in a room with whiteboards, you list firmographics, you agree on revenue range and employee count, and then the document sits in a Google Drive folder while your SDRs keep targeting whoever replies.

The problem isn't that the workshop was wrong. The problem is that you started with assumptions instead of data. Your best customers are already in your CRM. You just haven't asked it the right questions yet.

Claude Code can analyse your HubSpot or Salesforce export and surface patterns you didn't know existed. No pivot tables, no waiting on your data team, no guessing. This is how you turn a messy CRM export into a working ICP in 15 minutes.

Why your current ICP process doesn't hold up

ICP documents fail because they're built on what you think should work, not what actually works. You make logical guesses about company size, industry, and tech stack. Then you build targeting around those guesses. Then you wonder why conversion rates stay flat.

The real ICP lives in your closed-won data. It's the pattern that emerges when you look at who actually bought, how long the deal took, and what triggered the conversation in the first place. That data is already sitting in your CRM. You've just never isolated it properly.

Claude Code doesn't replace your ICP thinking. It accelerates the evidence-gathering step that most teams skip. Instead of debating whether your ICP is mid-market or enterprise, you export your deals, run the analysis, and see what the last 50 closed-won accounts actually look like.

What you need before you start

This isn't a theoretical tutorial. You need three things before Claude Code can run the analysis:

If your CRM is missing employee count or revenue, add a data enrichment step before you export. Clay can backfill those fields from ZoomInfo or Clearbit in under an hour. Without firmographic data, the analysis will tell you who bought but not why they fit.

Export the last 12 months of closed-won deals if you have them. If you're early stage and don't have 30 deals yet, include closed-lost deals and tag them in a separate column. The comparison will show you what differentiates buyers from tire-kickers.

Step 1: Upload the export and prompt Claude Code

Open Claude and attach your CSV. Start with this exact prompt:

```

Analyse this CRM export of closed-won deals and identify the dominant ICP patterns. Focus on:

• Industry distribution

• Employee count ranges

• Revenue or deal size clustering

• Geographic concentration

• Deal source patterns

Output a summary table with the top 3 ICP segments ranked by frequency, plus the average deal size and sales cycle for each segment.

```

Claude Code will load the CSV, run descriptive statistics, and return a segmented breakdown. The key insight isn't the most common industry. It's whether your closed-won deals cluster around one segment or spread across three.

If you see one dominant segment (for example, 60% of deals are HR Tech companies with 200 to 500 employees), that's your primary ICP. If you see three segments with roughly equal distribution, you either have three ICPs or you haven't found the real pattern yet.

Step 2: Isolate the highest-value segment

Revenue per deal matters more than deal count. If 40% of your deals come from enterprise accounts but they close at 3x the contract value of mid-market deals, enterprise is your ICP even if mid-market has more logos.

Prompt Claude Code to rank segments by total revenue contribution:

```

Rank the ICP segments by total revenue contribution. Show me the segment that generated the most revenue in the last 12 months, even if it wasn't the highest deal count.

```

This is where most teams get their targeting wrong. They optimise for volume instead of value. Your SDRs chase 200-person companies because there are more of them, but your best economics come from 1,000-person companies that close faster and churn less.

Partner UP ran this analysis for an HR Tech client and found that 70% of their revenue came from companies with 500+ employees, even though those deals represented only 30% of total opportunities. The ICP wasn't wrong. The targeting was just weighted toward the wrong segment.

Step 3: Find the deal source pattern

ICP analysis isn't just firmographics. It's how those accounts entered your pipeline. Inbound, outbound, referral, and event-sourced deals behave differently. The segment that converts best from outbound might not be the segment that converts best from inbound.

Prompt Claude Code to break down deal source by segment:

```

Group the top ICP segment by deal source (inbound, outbound, referral, event). Show conversion rate and average sales cycle for each source within this segment.

```

If your best segment converts at 40% from referrals but 12% from outbound, you have a channel-fit problem, not an ICP problem. You're targeting the right companies through the wrong motion.

This is where Clay comes in. Once you know your highest-converting segment and source, you can build a Clay workflow that enriches net-new accounts in that segment and triggers outbound only when they match both the firmographic profile and the intent signal that worked for past deals.

Step 4: Test the edge cases

The accounts that almost fit your ICP but didn't close tell you where the boundaries are. If you filtered for 200 to 500 employees but your best deals all came from the 400 to 500 range, your lower bound is too low.

Export your closed-lost deals from the same period and upload them as a second CSV. Prompt Claude Code to compare:

```

Compare the closed-won ICP segment to the closed-lost deals. Highlight the firmographic differences between accounts that closed and accounts that didn't. Focus on employee count, revenue, and industry.

```

Claude Code will surface the delta. Maybe closed-lost deals skew toward smaller employee counts. Maybe they're in adjacent industries that look similar but have different buying cycles. This is the negative signal that sharpens your targeting.

At Partner UP, we ran this for a B2B marketplace client and found that deals under $50k annual contract value (ACV) had a 60-day longer sales cycle and a 25% lower close rate, even within the same industry and employee range. The ICP wasn't just firmographics. It was deal size.

Step 5: Turn the analysis into targeting criteria

Claude Code gives you the pattern. You still have to turn it into targeting logic. Take the output from step 2 and step 3 and build a table that maps to your outbound tools.

This table is your targeting brief. It's what you hand to your SDR, your Clay operator, or your outbound agency. It's specific enough to filter but broad enough to keep your list above 500 accounts.

If your list drops below 500 accounts, your ICP is too narrow. Loosen one variable (usually employee count or revenue range) and re-run the analysis to see if the conversion rate holds.

When to re-run the analysis

ICP isn't static. As you move upmarket or launch a new product, the pattern changes. Re-run this analysis every quarter if you're pre-Series A. Every six months if you're post-Series A.

Watch for two signals that mean it's time to refresh:

• Your close rate drops by more than 10% quarter-over-quarter

• Your average deal size shifts by more than 20% in either direction

Both signals mean your targeting criteria no longer match the market or your product has outgrown the original ICP. Export the last 90 days of closed-won deals and repeat the Claude Code workflow.

What this doesn't replace

Claude Code is a pattern-finder, not a strategy tool. It tells you who bought, not why they bought or what pain you solved. You still need customer interviews. You still need to talk to your CS team about retention and expansion patterns.

But it removes the guesswork from the first step. Instead of starting your ICP workshop with a blank whiteboard and assumptions, you start with a data-backed segmentation that already reflects reality. The workshop becomes about refinement, not invention.

FAQ

Can I use this if I don't have 30 closed-won deals yet?

Yes, but include closed-lost deals and tag them separately. The comparison will still show patterns, even if the sample size is small. Below 15 total deals, wait until you have more data.

What if my CRM export is missing employee count or revenue?

Run the export through Clay first. Use a ZoomInfo enrichment step to backfill missing firmographics. Without those fields, the analysis will default to industry and geography only, which isn't enough for targeting.

Do I need to clean the CSV before uploading it to Claude Code?

Not usually. Claude Code handles messy data well. If you have duplicate rows or formatting issues, it'll flag them. The only manual step is making sure your column headers are clear (for example, "Employee Count" not "EMP_CT").

Partner UP works with GTM and RevOps teams on ICP development, data enrichment, and outbound targeting. If you're running CRM exports manually and guessing at patterns, reach out at hello@partneruphq.com or book a call at calendly.com/eleilademir.