Internal-External ICP Builder
Build an ICP by combining company records, customer calls, and outside research
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 91%
The Internal-External ICP Builder starts by giving an AI model access to relevant company-owned evidence, such as documents, emails, CRM information, and recorded customer calls. The model infers a rough ideal customer profile from the patterns in that evidence. It then pairs those findings with external research to add market characteristics, likely priorities, terminology, and contextual details that may not exist internally. The resulting ICP is not treated as a finished answer; it is embellished, reviewed, and checked against actual customers. This creates a practical profile much faster than starting from a blank template while retaining a connection to observed behavior. The method is especially useful for teams that have accumulated customer knowledge but have never organized it into a formal persona.
Origin
Extracted from Marketing Against The Grain during a discussion of using Claude, ChatGPT, Google Drive, email, HubSpot data, and customer calls to construct an ICP.
Core principles
- 01Internal evidence reveals who already engages and buys
- 02External evidence fills gaps in company-owned data
- 03Customer calls add language and motivations that records miss
- 04An initial ICP should be refined rather than treated as final
How to run it
- 1
Connect internal evidence
Gather the documents, emails, CRM records, and customer materials that reveal who engages with the company and why.
Pro tip Prioritize sources tied to actual customer behavior rather than internal opinions.
Watch out Exclude data the model is not authorized to process.
- 2
Add customer conversations
Record and transcribe representative customer calls so the profile includes customer language, goals, and objections.
Pro tip Sample across customer types instead of relying only on enthusiastic accounts.
Watch out Obtain appropriate consent before recording or processing conversations.
- 3
Generate the rough ICP
Ask the model to infer the likely ideal customer, including attributes, pains, desired outcomes, and buying context.
Pro tip Require the model to tie conclusions back to supplied evidence.
Watch out Do not confuse frequent attributes with causal buying criteria.
- 4
Enrich externally
Pair the rough ICP with outside data and research to fill information gaps and understand the wider market.
Pro tip Separate verified external facts from model inference.
Watch out External data can reinforce stereotypes if accepted uncritically.
- 5
Validate and revise
Compare the profile with real customer experiences, then update weak or unsupported sections.
Pro tip Use the ICP as a living hypothesis.
Watch out Do not deploy highly personalized campaigns before checking that the profile is accurate.
In the wild
A founder supplies relevant Drive documents, CRM notes, email patterns, and transcripts from recent customer calls. An AI model identifies a recurring buyer type, then enriches it with external company and market information. The founder checks the result with several customers and corrects two unsupported assumptions.
→ The company gains a usable ICP for positioning and prospecting without hiring a full product marketing team.
Common mistakes
Treating the first profile as truth
The initial output is a hypothesis assembled from incomplete evidence and requires validation.
Using only external demographics
External attributes without internal behavior or interview evidence rarely explain why customers buy.
Is it for you?
Best for
Founders and small teams with scattered customer information but no formal ICP or persona research process.
Not ideal for
Businesses with too little customer evidence to distinguish actual demand from speculative assumptions.
From the transcript
“you can literally just ask Claude to build you an ideal customer profile.”
“And then once you have a rough ICP, you can ask the LLMs to go and embellish it with more information.”
“Just record some of your customer calls, put that with the data you have, and you will have a pretty robust, like ideal customer profile…”
From the episode
The AI Stack That Makes Our Product Marketing 10x Faster