Icp Research

icp-research

Builds rigorous, evidence-based personas using Jobs-to-be-Done. Writes to knowledge/icp/personas.md directly so the rest of the OS uses the updated context.

Where this method comes from

The struggling moment, the hiring and firing triggers, and the Four Forces are Clayton Christensen and Bob Moesta's Jobs to be Done, developed in Competing Against Luck (2016) and Moesta's switch-interview work. The unit of analysis is the moment of change, not the demographic: people do not buy because of who they are, they buy because something happened.

The Four Forces are push of the situation, pull of the new solution, anxiety about the new, and habit of the present. A segment where push and pull are strong but anxiety and habit are stronger is a segment that will not switch, however well it matches on firmographics.

Demographics and firmographics are context. The job the customer is hiring you for is the insight.

Christensen's milkshake study is the shape of the whole method: the buyers had nothing in common demographically, and everything in common situationally. That is why this skill treats a segment that shares firmographics but not a struggling moment as a failed segment, and says so.

Before you build: is the evidence strong enough?

Persona work lands in a file six other skills read as truth, so the first decision is whether to build at all.

How many personas to build, in priority order: 1 primary always, then a secondary only if its struggling moment is genuinely different from the primary's. Two personas with the same struggling moment and different job titles are one persona. Maximum 4 in total, because past that no downstream skill can hold them in context.

Hard rules on the numbers, so the judgement is not re-litigated every session:

Stop conditions. Do not proceed to writing personas at all if there has been no customer contact and no win/loss data. If the struggling moment for the primary persona cannot be written as a scene, stop and say so: that single gap invalidates every section below it.

Warning signs, in priority order:

1. Two personas share a struggling moment. Merge them, and say why.
2. The anxiety and habit forces are blank while push and pull are full. That means only won deals were studied. Lost deals and no-decision deals are where anxiety and habit live, and they are what explains the losses.
3. The "words they use" section reads like marketing copy. Real customer vocabulary is blunter and more specific than any brand's own language.
4. Firmographics are tight and the struggling moment is vague. That is an account list, not an ICP, and targeting on it will produce well-qualified prospects who do not move.

When to use

• "Build our ICP from scratch"
• "Refine our buyer personas"
• "Who are we actually selling to?"
• "Update the persona doc based on these customer interviews"
• "I uploaded customer research, build personas from it"

Inputs needed

• Source material (any combination):
  • Existing customer list (CSV in uploads/)
  • Customer interview notes
  • Win/loss data
  • Sales team input (paste in chat)
  • Existing case studies in knowledge/content-library/case-studies/
  • Pasted survey results
• Number of personas to build (default: 1 primary, 1-2 secondary; max 4)
• Use case: positioning, ABM targeting, content strategy, paid ads, sales enablement

Framework: Jobs-to-be-Done (JTBD)

People don't buy products. They hire them to do a job in their life. The job - not the demographic - is what drives the purchase.

Three job dimensions (all three required in every persona)

The struggling moment

The moment when the customer's current solution broke and they started looking. More predictive of buying behavior than any demographic. Every persona must have one.

Format: "What was happening in their life/job when they started looking? What was the specific situation?"

Hiring and firing triggers

• Hiring trigger: what made them actively seek a solution? (the push)
• Firing trigger: what were they using before, and why did it stop being enough? (the pull away)

Four Forces of Progress

Understanding all four forces explains why good products still lose deals - anxiety and habit beat push and pull.

JTBD statement format

"When [struggling situation], I want to [motivation], so I can [expected outcome]."

Example: "When the board asks for pipeline attribution and I can't answer, I want a marketing analytics tool I can trust, so I can show that marketing is driving revenue and keep my budget."

Process

1. Load existing knowledge. Read knowledge/icp/personas.md (if any), knowledge/markets/positioning.md, and case studies. If personas already exist, ask: "Replace or refine?"

2. Gather evidence. If uploads/ has customer data, run analysis:

   • Common industries
   • Common company sizes (employees, revenue)
   • Common roles (the buyer, the user, the influencer)
   • Common geographies
   • Common acquisition channels
   • Win-rate by segment if available

   If no data, interview the user with these questions (not generic persona questions - JTBD-specific):

   JTBD Interview Questions

   The struggling moment:

   • "Walk me through the specific day or week when you decided to start looking for a solution."
   • "What was happening that made the status quo no longer acceptable?"
   • "How long had that situation been going on before you started looking?"

   The firing trigger:

   • "What were you doing before? A spreadsheet, a different tool, a consultant, nothing?"
   • "What happened that made that stop being enough?"

   The hiring trigger:

   • "What made you choose this solution over doing nothing, or staying with the old approach?"
   • "What was the specific promise that made you move?"

   The four forces:

   • "What were you most anxious about when you decided to switch?" (anxiety)
   • "What almost made you not do it?" (habit + anxiety)
   • "Who else had to sign off? What were their concerns?" (committee anxiety)

   The three job dimensions:

   • "When you imagined this working perfectly, what did that feel like?" (emotional job)
   • "How did you explain this decision to your CEO or team?" (social job)
   • "What specific outcome were you hired to produce?" (functional job)

   Standard questions for firmographic baseline:

   • "Who are your top 5 customers by revenue or fit?"
   • "What pattern do you see across them?"
   • "Who do you NOT want as a customer?"
   • "Which deals close fastest?"

3. Write the persona doc. Use this structure (one block per persona):

   # ICP and Personas (DD-MM-YYYY)
   
   ## Primary persona: <Name + role>
   Example: "Maya, Head of Demand Gen at Series B SaaS"
   
   ### Demographics
   - Role: <title>
   - Seniority: <IC | Manager | Director | VP | C-level>
   - Team size: <range>
   - Reports to: <role>
   
   ### Firmographics
   - Company stage: <seed | A | B | C | growth | enterprise>
   - Employees: <range>
   - Revenue: <range>
   - Industries: <list>
   - Geographies: <list>
   
   ### The struggling moment
   <Specific narrative of the situation that triggered them to look for a solution.
   Must be a concrete scenario, not a general pain point.>
   Example: "It's the Thursday before the board meeting. Pipeline is flat.
   The CEO just asked 'what is marketing doing?' and Maya doesn't have an answer
   that will hold up to scrutiny."
   
   ### Jobs-to-be-Done
   
   **Functional job** (the practical task):
   <What they need done, specifically.>
   
   **Emotional job** (how they want to feel):
   <How they want to feel when the job is done.>
   
   **Social job** (how they want to be perceived):
   <How they want others to see them as a result.>
   
   **JTBD statement**:
   "When [struggling situation], I want to [motivation], so I can [expected outcome]."
   
   List 2-3 JTBD statements if the persona has multiple jobs.
   
   ### Four Forces of Progress
   
   | Force | What it is for this persona |
   |---|---|
   | Push (away from status quo) | <specific situation that's making them move> |
   | Pull (toward new solution) | <specific promise that attracted them> |
   | Anxiety (about the new) | <what they're afraid might go wrong> |
   | Habit (inertia) | <what makes doing nothing easier> |
   
   ### Goals (what they're measured on)
   1. ...
   2. ...
   3. ...
   
   ### Pains (what's broken today)
   1. ... (specific, not generic)
   2. ...
   3. ...
   
   ### Hiring / Firing triggers
   - **Hiring trigger**: <what caused them to actively seek a solution>
   - **Firing trigger**: <what they were using before and why it stopped working>
   
   ### How they buy
   - Discovery channels: <where they find tools>
   - Influencers: <who they trust>
   - Decision criteria: <ranked: outcomes, price, fit, brand, integrations>
   - Buying committee: <other roles involved>
   - Typical sales cycle: <weeks/months>
   - Common objections: <list - mapped to the four forces where possible>
   
   ### Where they hang out
   - Communities (Slack, Discord, Reddit subs)
   - Newsletters and podcasts
   - Conferences
   - LinkedIn groups, hashtags
   
   ### Voice (how to write to them)
   - Words they use: ...
   - Words they avoid: ...
   - Tone: <formal | conversational | technical>
   - Length tolerance: <skim | medium | deep dives>
   
   ### Anti-persona (NOT this person)
   Who looks similar but isn't a fit. Example: "Solopreneurs and pre-seed founders.
   They don't have budget or scale."
   
   ### Evidence
   - Source 1: <interview, survey, win-rate analysis>
   - Source 2: ...
   - Confidence level: high / medium / low (based on N data points)
   
   ---
   
   ## Secondary persona: <Name + role>
   Same structure, shorter
   

4. Cross-check against case studies. Open knowledge/content-library/case-studies/. Do the existing customers fit the personas? Specifically: do their struggling moments match? If not, flag the mismatch. Demographic fit without JTBD fit is a weak persona.

5. Check against knowledge/learnings.md. Past campaigns may have revealed which struggling moment converts fastest. Update the four forces accordingly.

6. Self-check. Each item is checkable by reading the persona document. Where an item says count, count it in the artifact rather than asserting the item passed.

   • Every persona's struggling moment describes a scene with a time, a person and an event in it. If it can be read as a job description, it fails
   • Every persona has all three job dimensions filled, and the emotional and social entries use different words from the functional one rather than restating it
   • Every JTBD statement matches the template exactly: "When ..., I want ..., so I can ...". Count the statements per persona and confirm 1 to 3
   • All four force cells are filled for every persona. A blank anxiety or habit cell is a research gap, so mark it [NEEDS INPUT] rather than leaving it empty
   • Hiring and firing triggers are both named, and the firing trigger names an actual prior solution
   • At least 3 pains, each containing a number, a frequency, or a named artifact. "Wants to grow" fails, "spends 30% of the week pulling reports" passes
   • Anti-persona names a specific group that looks like a fit and is not, with the reason
   • The evidence section lists the sources by name and the Confidence level matches the count of them against the table above. Compare the two, do not assert
   • Every "words they use" entry is traceable to a named source, or is written [NEEDS INPUT: real customer language]. Count the sourced entries
   • Each persona block has exactly one struggling moment, and no two blocks share it

7. Write to knowledge/icp/personas.md, only through the ownership rules below.

8. Save a version snapshot to output/icp-research/<DD-MM-YYYY>-personas.md so prior versions are preserved.

9. Offer follow-ups:

   • Update positioning to match (/brand-context or manual edit)
   • Build content briefs targeted at the struggling moment (not generic persona)
   • Run /competitor-analyst to map competitors against the same JTBD

Rules

• Never invent firmographics. If you have 3 customers, say "based on 3 interviews, low confidence" rather than overstating.
• The struggling moment is mandatory. A persona without one is just a demographic profile.
• The anti-persona is mandatory. If you don't know who is NOT your customer, you don't know who is.
• Pains must be specific. "Wants to grow revenue" is not a pain. "Spending 30% of week manually pulling reports that still don't answer the CEO's questions" is.
• All three job dimensions (functional, emotional, social) are required. Most personas have functional jobs only. Emotional and social jobs are where messaging differentiation lives.
• Voice notes must be concrete enough to use. "Professional" is useless. "Uses 'GTM' not 'go-to-market', avoids exclamation points, prefers numbered lists" is useful.
• Never invent the "words they use" vocabulary. Invented customer language is the worst thing this skill can emit, because /content-writer and /copy-review then treat it as observed speech and put it into public copy. Unsourced vocabulary is [NEEDS INPUT: real customer language].

Shared file ownership: knowledge/icp/personas.md

This skill owns knowledge/icp/personas.md. It owns the file's structure, the section order, the primary and secondary persona slots, the ICP firmographic summary, and the anti-persona. /customer-persona also writes to this file, as a contributor of single persona blocks into the structure this skill defines.

When a persona block already exists and carries a confidence value higher than the one you are about to write, do not silently replace it. Show both, name the difference in evidence, and let the user choose.

The write gate, which is not optional and is identical in both skills. Neither may relax it:

1. Write to this file only after showing the user the diff and getting an explicit yes. Six other skills read it as truth, so an ungated write turns one thin session into permanent canon nobody can trace back.
2. If confidence is low, meaning fewer than 5 distinct data points, write to output/ only and say plainly that the persona is not yet canon. Do not offer the shared-file write at all.
3. Carry the confidence: high | medium | low field into every block written, so a reader can tell canon from draft without opening the sources.
4. Never invent a firmographic, a struggling moment, a quote, an objection, or a piece of customer vocabulary to complete a block. [NEEDS INPUT: <what would fill it>] is the correct output.

What this skill cannot know

These are the limitations that bite in practice. Where one applies to the artifact, write it into the document as an open question rather than leaving the reader to assume it was verified.

1. Whether the struggling moment generalises. Interviews capture the people who bought and agreed to talk. The buyers who never reached you, and the ones who churned quietly, have struggling moments this method never sees.
2. Whether the segment is commercially worth serving. Nothing here sizes the segment, prices it, or estimates acquisition cost. A perfectly evidenced persona can still describe a segment too small or too cheap to build a business on.
3. What people will actually do, as opposed to what they said. Switch interviews are reconstructions, and reconstructions rationalise. Anxiety in particular is under-reported, because nobody enjoys describing what scared them.
4. Whether the four forces have shifted since the interviews. A competitor's launch, a price change, or a regulation can flip habit and anxiety within a quarter. Every persona block should carry the date its evidence was gathered so a reader can judge staleness.

Related skills

• /customer-research for running the interviews and surfacing the verbatim language, and it must run first when there are fewer than 5 distinct data points
• /customer-persona for a single deep buyer profile inside a segment this skill has already defined, and it contributes blocks into the file this skill owns
• /positioning-doc for turning the struggling moment and the pull force into a positioning statement
• /competitor-analyst for mapping who else is being hired for the same job, which is where the anxiety and habit forces usually come from
• /brand-context for creating the knowledge base when knowledge/icp/personas.md does not exist yet
• /content-writer for consuming the voice notes, which is why unsourced vocabulary is forbidden above
• /campaign-brief for targeting a campaign at the struggling moment rather than at the job title

Quick Reference: JTBD Framework