customer-research

Customer Research

Uncover what customers actually think, say and struggle with, so positioning, product and copy rest on evidence instead of assumption. The bar: every insight carries at least one exact quote with its source, and nothing is invented to fill a gap.

When to use

• "What do customers say about this?" "Mine these reviews." "Go through these call transcripts."
• Build a persona, a jobs-to-be-done map, or a bank of customer quotes for copy.
• Find why customers buy, hesitate, switch or leave.

Not for building a prospect list (lead-scouting) or writing the copy itself (landing-copy, cold-email, ad-copy-testing). This skill feeds them.

Three modes

1. Analyse what exists. The Commander hands you transcripts, survey exports, reviews, support tickets, win and loss notes. You extract the signal.
2. Mine public sources. Review sites, forums and communities where customers speak unprompted. You know where to look and what to take.
3. Go and ask. No signal exists yet, or only a customer can answer. You prepare interviews and surveys; the Commander runs them. See references/interviews-and-surveys.md (included at the end of this skill).

Most jobs mix modes. Mine what is already public before asking: it tells you what to ask and in whose words. When first-hand interviews and public posts disagree, weight the interviews higher.

Method

1. Set the goal. Ask what the research is for (messaging, personas, product gaps, churn) and what material already exists. Then, only if unclear: which customer segment, what the product is, and which deliverable is wanted. Look for a product brief in the workspace with fs.search before asking.
2. Collect the material. Files the Commander gave: fs.list, then fs.read (it reads PDF text). There is no audio or video transcription: if a call exists only as a recording, ask for the transcript text. Public sources: web_search with the search patterns in references/source-guides.md (included at the end of this skill), then web_fetch each thread or review page; use browser.navigate and browser.get_text when a page only shows its reviews in a real browser. If a site sits behind a login or a verification wall, record it as not reachable and move on.
3. Log quotes as you read. Append each useful passage to research/quote-log.csv with fs.append, using the fields in the table below. Copy the words exactly at the moment you read them. Do this before any summarising; a quote rebuilt from memory is not a quote.
4. Extract. Run every asset through the six-part framework below.
5. Synthesise. Cluster similar pains, outcomes and triggers across assets. Score each theme by frequency (how many independent sources) and intensity (how strongly it is felt, judged from the language). Split by customer profile where patterns differ. Pick the five to ten quotes that carry each theme best. Flag contradictions, especially where customers say one thing and do another.
6. Label confidence on every insight using the guardrails below.
7. Check the quotes. For each quote in the deliverable, confirm the exact words are in the source: fs.search for file quotes, the logged line for web quotes. Fix or drop any that do not match.
8. Write the deliverable the Commander chose, save it with fs.write, and name it with deliverable_note.