Research

Overview

research is the diagnostic entry point for this toolkit. Given a description of a business and its audience, it proposes candidate segments, scores them on eight dimensions, and recommends which funnel stage — awareness, engagement, or integration — is the actual bottleneck for the top segment. Everything else in this toolkit either consumes that diagnosis (the router's Full-Flow Mode) or exists to pressure-test one segment on its own (thought-partner mode).

When to Use

• An operator doesn't know which segment to prioritize and wants a from-scratch diagnosis.
• An operator already has a segment in mind and wants it pressure-tested, not just validated.
• Run automatically first, always, in the router's Full-Flow Mode — its output is the weighting input for awareness, engagement, and integration.
• Not for refining an already-agreed single-stage program with no segment question attached — go directly to awareness, engagement, or integration for that.

Read First

Before anything else, read and apply using-fan-development-skills/SKILL.md — it defines the Core Operating Behaviors, detects Single-Skill vs. Full-Flow Mode, and defines Session Continuity & File Output. Check for an existing research.md for this business before proceeding, and read it in as context if one exists. Everything below assumes those behaviors are already active.

Resolve the Intake

Two things get resolved before any proposing or scoring happens.

1. Mode.

• Build-from-scratch — the operator doesn't have a segment yet, or wants a full diagnosis from zero.
• Thought-partner/refine — the operator already has a segment or in-flight program and wants it pressure-tested. Skip Propose Candidates below; score only the segment(s) they name.

2. Starting point, regardless of mode — what the operator has to work from:

• Share data — they paste or describe real data about their audience.
• Identify softening — they name where a metric is declining.
• Name a target metric — they state what they want to move, with no current baseline.

Build-from-scratch only — establish what the business is:

1. Ask for the company or product name.
2. Ask whether to research it (search the web to draft a description) or take the operator's own description instead.
3. If web-researched: draft the description, show it back, and confirm it's accurate before continuing. Do not proceed on an unconfirmed description.
4. If operator-provided: use it as given — it's already theirs, no confirmation loop needed.

Confirm Coverage (build-from-scratch mode only)

Before proposing candidates, ask the operator directly: "Are there any key metrics, data sources, resource considerations, priorities or anything else about the business you'd like to add before we generate segment candidates?" Wait for their answer and fold anything they share into the business/audience understanding before moving to Propose Candidates below.

Propose Candidates (build-from-scratch mode only)

1. Start from the default list: new/first-touch users, lapsed users, power users not yet advocates, adjacent-market prospects, internal/employee advocates. This list is a floor, never a ceiling.
2. Add candidates the business/audience description or shared data actually suggests. A run that scores only the five defaults, unmodified, regardless of what business it's given is a bug, not a valid outcome — pressure yourself to find at least one segment specific to what the operator described.

Source Discernment

Before searching or asking for material, reason out loud about what kind of external evidence actually matters for this business — a consumer app implies app-store reviews and social listening; a B2B/enterprise product implies analyst reports and community benchmarks; a local/physical business implies local market data. State this reasoning; never apply a fixed hierarchy regardless of business type.

Search or Ask

If a web search or fetch capability is available in this session, use it — both to fill in the business description above and to source real evidence for scoring below. This is the default path, not a hypothetical to describe and then skip. If no such capability is available, say so plainly and ask the operator to paste source material instead — see Graceful Degradation.

Weight the Dimensions

The eight dimensions — strategic fit, evidence of affinity, segment size, conversion likelihood, repeat/retention potential, ease of activation, operational feasibility, evidence confidence — are never weighted the same way twice. Before scoring:

1. State a summary of the business/audience understanding built so far.
2. Propose a specific weighting — which dimensions matter more for this business and why — with the reasoning tied to that context.
3. Pose it back to the operator as a question, guess attached — summary, then a question, then a guess for how they'd weight it based on the context built so far.
4. Wait for their reaction before scoring. Fold in any correction.

Score

Score every candidate segment (defaults plus additions, or the named segment(s) in thought-partner mode) against the weighted dimensions. Any claim without solid evidence — a size estimate, a conversion assumption, a "this segment behaves like X" comparison — gets LOW_EVIDENCE explicitly, not smoothed into a confident-sounding number.

Graceful Degradation

No search capability, or a search that comes back thin: ask the operator to paste source material directly rather than inventing evidence to fill the gap. This is a first-class fallback, not an edge case — state plainly when it's happening, don't silently degrade quality without saying so.

Output the Brief

• Ranked candidate segments — or, in thought-partner mode, the named segment's score with a note on how it compares to what a from-scratch scan would likely surface.
• Evidence per dimension, with LOW_EVIDENCE flags visible, not buried in prose.
• A recommended stage emphasis for the top segment(s) — is the bottleneck awareness, engagement, or integration? This becomes the weighting input if the operator continues into the full flow.
• Close per Mode Detection in the router: Single-Skill Mode ends with a stated next-step recommendation; the Full-Flow check-in stop is the router's job, not this skill's.
• Save to fan-development/<business-slug>/research.md per the router's Session Continuity & File Output, and state the path in the close.

Common Rationalizations

Red Flags

• Every run proposing the same five default segments with no business-specific additions.
• Weighting applied without ever stating the reasoning behind it, or applied identically across unrelated business types.
• A LOW_EVIDENCE-worthy claim presented without the flag.
• A web-researched business description used without ever being shown back for confirmation.
• Build-from-scratch mode moving straight from the business description into Propose Candidates without asking the operator if there's anything else to add first.
• Thought-partner mode running a full candidate-proposal sweep the operator didn't ask for.
• A search capability available in the session but never actually invoked.

Verification

Before treating a run as complete, confirm:

• Mode (build-from-scratch vs. thought-partner) and starting point were resolved before any proposing or scoring
• Build-from-scratch: the business description is either operator-provided or web-researched-and-confirmed before continuing
• Build-from-scratch: the operator was explicitly asked for any key metrics, data sources, resource considerations, or priorities to add, before candidates were proposed
• At least one candidate segment beyond the default five, when the business description supports one
• Source-discernment reasoning was stated out loud, not applied as a fixed hierarchy
• A web search capability was actually invoked when available in the session
• Dimension weighting was reasoned per-business and confirmed with the operator via the summary/question/guess pattern before scoring
• Every LOW_EVIDENCE-worthy claim is flagged, not smoothed over
• Output includes ranked segments (or the thought-partner score), evidence, confidence flags, and a stage-emphasis recommendation
• An existing research.md for this business was checked for and read first if one existed; the brief was saved to the correct path per Session Continuity & File Output