Evidence-Synthesized Ideal Customer Profile
Combine current internal evidence with external signals to build an actionable ICP
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 98%
This method treats an ideal customer profile as a research synthesis rather than a fictional persona drafted from intuition. The process searches recent internal materials—such as customer documents, case studies, persona files, and win-rate analyses—and compares them with external signals from analyst reports, job advertisements, technology stacks, funding events, reviews, and recent community discussions. The evidence is normalized into a structured profile covering the company and product, core buyer roles, known success metrics, recurring pains, competitors, differentiators, and exclusion flags. Recency filters prevent stale internal assumptions from dominating, while source thresholds improve external evidence quality. The resulting document becomes a reusable input for micro-audience discovery, messaging, qualification, and future profile refinement.
Origin
Extracted from Marketing Against The Grain during a demonstration of building HubSpot's ICP with Claude, Google Drive data, and external research.
Core principles
- 01Ground customer profiles in evidence rather than intuition alone.
- 02Prioritize recent internal information so the profile reflects the current business.
- 03Cross-check internal assumptions against external market behavior.
- 04Define exclusions as explicitly as inclusions.
- 05Structure the output for downstream segmentation and campaigns.
How to run it
- 1
Seed the internal search
Identify file names and topics associated with customers, personas, win rates, and case studies, then retrieve the most relevant internal documents.
Pro tip Use both direct customer evidence and performance evidence such as wins and losses.
Watch out Do not expose confidential internal material to an AI system without appropriate access controls.
- 2
Apply a recency filter
Prioritize recently edited sources so the profile reflects current positioning, customer behavior, and product capabilities.
Pro tip Keep older sources only when they provide durable evidence unavailable elsewhere.
Watch out Recent does not automatically mean accurate; preserve source traceability.
- 3
Collect external signals
Research analyst reports, recent job listings, technology stacks, funding signals, reviews, and current community discussions.
Pro tip Specify time windows and minimum evidence thresholds in the research prompt.
Watch out Unfiltered web evidence can introduce stale, duplicated, or low-quality claims.
- 4
Synthesize the profile
Organize the evidence into a company and product snapshot, buyer roles, success metrics, pains, competitors, differentiators, and exclusions.
Pro tip Keep fields structured so another assistant can parse them reliably.
Watch out Do not let polished prose obscure contradictions between sources.
- 5
Review and correct
Inspect the generated profile, challenge dubious exclusions or claims, and rerun focused research where needed.
Pro tip Ask sales, customer success, and product stakeholders to review fields they directly observe.
Watch out Treat the generated ICP as a versioned hypothesis, not unquestionable truth.
In the wild
The host asks Claude to search recent internal files and external sources, then produce a HubSpot CRM profile containing buyer personas, success metrics, pains, competitors, differentiators, and exclusion flags. The first result includes a questionable exclusion involving very small companies, which the host flags for refinement.
→ The team receives a structured ICP while retaining human judgment over disputed conclusions.
Common mistakes
Using stale persona documents
Old profiles may encode former product capabilities, buyers, or market conditions and should not outweigh current evidence.
Omitting exclusion flags
A profile that describes only desirable traits encourages wasted spend on prospects the product cannot serve.
Copying the first output unchanged
Generated conclusions can be wrong; the host explicitly identifies an exclusion that should be refined.
Is it for you?
Best for
B2B organizations with scattered internal customer evidence and accessible external market signals.
Not ideal for
A brand-new business with no customer evidence and no sufficiently specific market hypothesis.
From the transcript
“prioritize the 10 most recently edited files because you want them to be most up to date.”
“Then I look at external resources, things like analyst reports, things like LinkedIn jobs, things like tech stacks, things like funding signals, things like community…”
“what competitors do they use, what are differentiators, what are exclusion flags, like you think people that you definitely do not want to market because…”
From the episode
How I Turned Perplexity Labs into a Marketing Machine