icp-refiner

name	icp-refiner
description	Interactively refine your Ideal Customer Profile through guided questions, web research, and iterative Amplemarket searches until your targeting criteria are dialed in.
metadata	{"author":"amplemarket","version":"1.0.5","category":"Configuration"}
compatibility	Requires Amplemarket MCP server

ICP Refiner

Interactively refine your Ideal Customer Profile through guided questions, web research, and iterative Amplemarket searches until your targeting criteria are dialed in.

Instructions

When a user wants help defining, sharpening, or narrowing their Ideal Customer Profile, guide them through a structured conversation that combines their domain knowledge with real market data from Amplemarket.

Steps

1. Ask the user foundational questions to establish a starting ICP hypothesis. Gather:

   • What product/service do you sell? Understanding the value prop determines which roles and industries to target.
   • Who are your best customers today? Ask for 2-3 company names or domains. These anchor the firmographic profile.
   • What role buys your product? Target titles, seniority, and department (e.g., "VP of Engineering" or "marketing leaders").
   • What is your typical deal size? This guides company size targeting (smaller deals align with SMB, larger with enterprise).
   • Who should be excluded? Industries, company types, or geographies that are not a fit (e.g., "no government or non-profits").
   • Who are your competitors? Ask for competitor company names and/or domains. These will be excluded from all search results to ensure the ICP only contains potential buyers, not competitors or companies that sell similar products.

   If the user already has a partial ICP, skip questions they have answered and focus on gaps. Present questions conversationally, do not dump all five at once. Start with the most important (product and best customers) and follow up based on responses.

2. Research the user's company by calling WebSearch with the user's company name or domain to understand their positioning, product category, and competitive landscape. Use the findings to validate and enrich the ICP hypothesis. For example, if the company sells developer tools, prioritize engineering and IT departments.

3. Enrich best customer examples if the user provided company names or domains. Call mcp__claude_ai_Amplemarket__enrich_company for each example company (up to 3). Extract firmographic attributes:

   • Industry and sub-industry
   • Company size (employee count)
   • Company type (privately held, public, etc.)
   • Location and headquarters
   • Technology stack (if relevant to the user's product)

   Look for patterns across the examples. Shared industries, similar sizes, or common company types form the foundation of the ICP.

4. Map descriptions to valid enum values by calling:

   • mcp__claude_ai_Amplemarket__get_industries to match the user's industry descriptions and the enriched company industries to valid API values.
   • mcp__claude_ai_Amplemarket__get_job_functions to match target departments and roles to valid API values.

   Present the mapped values to the user for confirmation: "Based on your input, I'm mapping your target industries to [X, Y, Z]. Does that look right?"

5. Run the first search with the initial ICP hypothesis by calling mcp__claude_ai_Amplemarket__search_people with:

   • person_titles: derived from target role
   • person_seniorities: mapped seniority levels
   • person_departments: mapped department values
   • person_locations: target geographies
   • company_industries: resolved industry values
   • company_sizes: inferred from deal size and best customer patterns
   • company_types: if exclusions specified
   • full_output: true
   • page_size: 20

   Competitor exclusion: After retrieving results, filter out any prospects who work at companies the user identified as competitors. Also use WebSearch to identify additional competitors in the same product category (e.g., search for "[user's product category] competitors" or "[user's company] alternatives") and exclude those as well. When presenting results, flag and remove any company that appears to sell a similar product, as these are competitors, not prospects. If uncertain whether a company is a competitor, check their website description from enrichment data and ask the user: "I found [company] in the results. They seem to sell [similar product]. Should I exclude them?"

   Report the total result count to the user: "Your initial ICP returns 8,421 prospects (after excluding X competitors). Let's look at the distribution to see if we need to narrow or adjust."

6. Present a distribution breakdown of the search results. Analyze the returned prospects and summarize:

   • By title: Top 5 most common titles and their percentage of results
   • By industry: Top 5 industries represented
   • By company size: Distribution across size ranges
   • By location: Top geographies represented

   Present the distribution as a formatted summary:

   Title Distribution:

   • VP of Sales: 18% (2,156)
   • Director of Sales: 29% (3,468)
   • Sales Manager: 38% (4,548)
   • Other: 15% (1,793)

   Industry Distribution:

   • Computer Software: 41%
   • Internet: 22%
   • Staffing & Recruiting: 14%
   • Other: 23%

   Ask the user targeted questions based on the distribution:

   • "42% of results are Managers - do you want to keep them or focus on Director+ only?"
   • "I see a lot of results from the Staffing & Recruiting industry - should we exclude that?"
   • "Most prospects are in the US - do you want to add or restrict geographies?"

7. Iterate 2-3 times based on feedback. For each round of refinement:

   • Adjust the search filters based on user responses (add seniority filters, exclude industries, narrow geographies, change company size ranges).
   • Call mcp__claude_ai_Amplemarket__search_people with the updated parameters.
   • Report the new result count and how it changed: "Narrowed from 8,421 to 3,247 by restricting to Director+ seniority and excluding Staffing & Recruiting."
   • Show the updated distribution if the user wants to keep refining.
   • Continue until the user is satisfied or the count is in a reasonable range for their use case.

8. Present the final ICP definition in a structured format:

   Your Refined ICP

   Criteria	Value
   Target Titles	VP of Sales, Director of Sales, Head of Revenue
   Seniority	VP, Director, Head
   Department	Revenue
   Industries	Computer Software, Internet, SaaS
   Company Size	201-500 employees, 501-1000 employees
   Company Type	Privately Held
   Locations	United States, United Kingdom
   Exclusions	Non Profit, Government Agency; Staffing & Recruiting industry
   Competitors Excluded	Gong, Outreach, Salesloft, Apollo, Groove
   Total Matching Prospects	3,247

   Include a brief refinement summary: "Started with 12,400 prospects. Narrowed to 3,247 by adding VP/Director seniority filter, excluding Staffing & Recruiting industry, and focusing on 201-1000 employee companies."

9. Offer to create a lead list from the final ICP. Ask: "Would you like me to create a lead list from these 3,247 prospects? I can add them to Amplemarket with enrichment enabled so you have emails and phone numbers ready to go."

   If the user agrees, call mcp__claude_ai_Amplemarket__create_lead_list with:

   • type: "linkedin"
   • name: descriptive name based on the ICP (e.g., "ICP - VP Sales - SaaS - US/UK - Mar 2026")
   • leads: array of {"linkedin_url": "..."} objects from the search results
   • options: {"enrich": true, "validate_email": true}

   If the total exceeds one page, paginate through all results using mcp__claude_ai_Amplemarket__search_people with incrementing page values, then use mcp__claude_ai_Amplemarket__add_leads_to_lead_list for subsequent batches.

   Confirm creation with the user by showing the list ID, total leads added, and enrichment settings enabled. Inform the user about credit usage: "This list of [N] leads will use [N] enrichment credits. The list is now processing. You can check its status anytime."

Important Notes

• Always validate industries and job functions against the API before using them in searches. Never guess enum values.
• Show the user the result count after every search iteration so they can see the impact of each change.
• If the user does not know their best customers or target role, use WebSearch to research their company and suggest an initial hypothesis based on their product category and competitive positioning.
• Keep refinement to 2-3 iterations to avoid fatigue. If the user is satisfied early, move to the final summary.
• When presenting distributions, focus on the most actionable insights. Highlight filters that would have the biggest impact on narrowing or broadening the results.
• Always exclude competitors from results. Use the user's competitor list plus WebSearch to identify companies that sell similar products. If a company in the results looks like it could be a competitor (similar product description, same category on G2/Capterra), flag it and ask the user before including it.

Dynamic Fields Generated

Field	Example Value	Source
{{icp_title_pattern}}	VP/Director of Sales	search_people analysis
{{icp_seniority}}	VP, Director	User input + validation
{{icp_industry}}	Computer Software, SaaS	get_industries
{{icp_company_size}}	201-1000 employees	User input + search validation
{{icp_location}}	United States, United Kingdom	User input
{{icp_total_addressable}}	3,247	search_people
{{icp_department}}	Revenue, Marketing	get_job_functions
{{icp_exclusions}}	Non Profit, Government Agency	User input
{{icp_competitors_excluded}}	Gong, Outreach, Salesloft, Apollo	User input + WebSearch
{{icp_company_type}}	Privately Held	User input
{{icp_refinement_summary}}	Narrowed from 12K to 3.2K by adding seniority filter	Analysis

Examples

Example 1: Full Refinement Workflow

User prompt: "Help me define my ICP - I sell sales engagement software but my targeting feels too broad"

What the skill does:

1. Asks foundational questions: product, best customers, target role, deal size, exclusions.
2. User responds: "We sell to sales leaders, our best customers are Gong, Outreach, and Salesloft. Average deal is $25K/year. No government or education."
3. Calls WebSearch to research the user's company positioning.
4. Calls mcp__claude_ai_Amplemarket__enrich_company for gong.io, outreach.io, and salesloft.com.
5. Discovers pattern: all three are 201-1000 employee, privately held, Computer Software / SaaS companies in the US.
6. Calls mcp__claude_ai_Amplemarket__get_industries and mcp__claude_ai_Amplemarket__get_job_functions to resolve enum values.
7. Runs mcp__claude_ai_Amplemarket__search_people with initial hypothesis. Returns 12,400 results.
8. Shows distribution: 38% Managers, 29% Directors, 18% VPs, 15% other. Top industries: Computer Software (41%), Internet (22%), Staffing (14%).
9. User says: "Drop Managers and exclude Staffing." Reruns search. Now 4,812 results.
10. Shows updated distribution. User says: "Focus on 201-1000 employees only." Reruns. Now 3,247 results.
11. Presents final ICP summary and offers to create a lead list.

Example output:

---

Your Refined ICP

Criteria	Value
Target Titles	VP of Sales, Director of Sales, Head of Sales, Director of Revenue Operations
Seniority	VP, Director, Head
Department	Revenue
Industries	Computer Software, Internet, Information Technology and Services
Company Size	201-500 employees, 501-1000 employees
Company Type	Privately Held
Locations	United States
Exclusions	Non Profit, Government Agency, Educational; Staffing & Recruiting industry
Total Matching Prospects	3,247

Refinement Journey:

• Round 1: Initial search with broad criteria returned 12,400 prospects.
• Round 2: Removed Manager-level seniority, excluded Staffing & Recruiting industry. Result: 4,812 prospects.
• Round 3: Narrowed company size to 201-1000 employees. Result: 3,247 prospects.

Key Insight: Your best customers (Gong, Outreach, Salesloft) are all mid-market SaaS companies. Focusing on this segment removed 74% of noise from your initial search.

Would you like me to create a lead list from these 3,247 prospects with email enrichment enabled?

---

Example 2: Starting from a Best Customer Example

User prompt: "Acme Corp is our best customer - find me more companies like them"

What the skill does:

1. Calls mcp__claude_ai_Amplemarket__enrich_company with domain "acmecorp.com" to extract firmographic attributes.
2. Discovers: 500 employees, Computer Software industry, Privately Held, headquartered in San Francisco.
3. Asks follow-up: "Acme Corp is a 500-person software company in SF. What role at Acme buys from you, and are there geographies or sizes I should exclude?"
4. User responds: "CTO and VP Engineering. US only, 200-1000 employees."
5. Calls mcp__claude_ai_Amplemarket__get_industries and mcp__claude_ai_Amplemarket__get_job_functions to resolve values.
6. Runs mcp__claude_ai_Amplemarket__search_people with mapped criteria. Returns 5,100 results.
7. Shows distribution: titles skew heavily toward Engineering Manager (45%). User narrows to C-Suite and VP only.
8. Reruns. 1,890 results. User approves and requests a lead list.

Example output:

I enriched Acme Corp and found the following profile:

Attribute	Value
Industry	Computer Software
Employees	500
Type	Privately Held
HQ	San Francisco, CA

Based on this, I searched for CTOs and VPs of Engineering at similar companies. Here is your refined ICP:

Criteria	Value
Target Titles	CTO, VP of Engineering, Chief Technology Officer
Seniority	C-Suite, VP
Department	Engineering & Technical
Industries	Computer Software, Information Technology and Services
Company Size	201-500 employees, 501-1000 employees
Locations	United States
Total Matching Prospects	1,890

Would you like me to create a lead list from these 1,890 prospects?

Example 3: Narrowing an Overly Broad ICP

User prompt: "My ICP is too broad - I'm targeting all marketing people at tech companies and getting 50,000+ results"

What the skill does:

1. Asks clarifying questions: "What seniority level actually buys? What company size closes fastest? Any geographies that perform better?"
2. User responds: "Directors and above buy. Companies with 100-500 employees close fastest. US and UK are our main markets."
3. Calls mcp__claude_ai_Amplemarket__get_industries to resolve "tech companies" to specific industry values.
4. Runs mcp__claude_ai_Amplemarket__search_people with narrowed criteria. Returns 9,200 results.
5. Shows distribution: "Marketing" department dominates, but "Director of Demand Gen" and "VP of Growth" are the highest-converting titles based on user feedback.
6. User narrows titles to demand generation and growth roles specifically. Reruns. 2,100 results.
7. Presents final ICP with the path from 50,000+ to 2,100 clearly documented.
8. Offers to create a lead list from the refined set.

Example output:

Your original ICP was returning 50,000+ results. Here is how we narrowed it:

Refinement Journey:

• Start: All marketing people at tech companies - 50,000+ prospects
• Round 1: Restricted to Director+ seniority, 51-500 employee companies, US and UK only - 9,200 prospects
• Round 2: Focused on demand generation and growth titles specifically - 2,100 prospects

Your Refined ICP

Criteria	Value
Target Titles	Director of Demand Generation, VP of Growth, Head of Demand Gen, Director of Marketing Operations
Seniority	Director, VP, Head
Department	Marketing
Industries	Computer Software, Internet, Information Technology and Services
Company Size	51-200 employees, 201-500 employees
Locations	United States, United Kingdom
Total Matching Prospects	2,100

That is a 96% reduction from your original search. These 2,100 prospects are much more likely to be actual buyers. Want me to create a lead list?

Troubleshooting

Problem	Solution
Zero results returned	Relax filters one at a time in this order: 1) Remove geography restrictions. 2) Broaden company size by one tier in each direction. 3) Add adjacent seniority levels (e.g., add Manager if only targeting Director+). 4) Try related industry values from mcp__claude_ai_Amplemarket__get_industries. Report each change and its impact on result count.
Too many results (10,000+)	Add filters progressively: 1) Restrict seniority to Director+ or VP+. 2) Narrow company size to 1-2 tiers. 3) Add geography constraints. 4) Exclude low-fit industries. Show the user the count after each change so they can decide when to stop.
Industry values don't match user's description	Call mcp__claude_ai_Amplemarket__get_industries and present the closest 5-10 options. Let the user pick which ones apply. Common mismatches: "SaaS" maps to "Computer Software" or "Internet"; "fintech" maps to "Financial Services" or "Banking".
User doesn't know their ICP at all	Start with WebSearch to research their company and product. Identify the product category, then suggest a default ICP based on common buyer personas for that category (e.g., developer tools typically sell to VP Engineering / CTO at mid-market software companies). Run a broad initial search and let the data guide refinement.
Conflicting criteria from user	Surface the conflict explicitly: "You mentioned targeting startups (1-50 employees) but also said your average deal is $50K/year. That price point typically fits 200+ employee companies. Which should we prioritize?" Let the user resolve the tension, then adjust.
search_people returns unexpected results	Verify enum values by rechecking mcp__claude_ai_Amplemarket__get_industries and mcp__claude_ai_Amplemarket__get_job_functions. Common issues: 1) Industry values are outdated, so re-fetch. 2) Title keywords match unrelated roles, so add department filter to constrain. 3) Location strings are ambiguous, so use "City, Country" format. 4) Company size enum doesn't match, so use exact values like "201-500 employees".
Enriched best customer has sparse data	Try enriching with the company's LinkedIn URL or alternate domains. If firmographic data is limited, fall back to asking the user to describe the company profile manually and use that as the basis for the ICP hypothesis.