customer-research works in three modes: analyze research material you already have, mine public sources such as Reddit and review sites, or run interviews and surveys to ask customers directly. Most projects combine them, starting with what is already public.
Each insight carries a high, medium or low confidence label, and personas are built only from enough independent data points. Deliverables include a synthesis report, a quote bank, personas and a jobs-to-be-done map.
What You Get
- A research synthesis with ranked themes, representative quotes and implications
- A quote bank of verbatim customer language organized by theme
- Personas with a profile, trigger events, pains, outcomes, objections and vocabulary
- A jobs-to-be-done map covering functional, emotional and social jobs
- A research gap list that says what is still unknown and how to find out
Common Use Cases
- Summarizing a folder of call transcripts into themes
- Mining G2 reviews and Reddit threads for customer language
- Planning customer interviews and a PMF survey
- Building provisional personas for a product with no reviews yet
What Makes It Unique
Every insight is labeled high, medium or low confidence by how many independent sources support it, with recency and sample bias checks and a minimum of five data points per segment before a persona is built. For new products it walks through proxy sources in order and tags each persona with its proxy.
Under The Hood
The SKILL.md and two references are loaded as context. It ships no scripts. When it mines online sources it uses web search and page fetches on Reddit, review sites, forums and similar places, and it treats fetched reviews and threads as untrusted data, ignoring any instructions inside them. It reads the product marketing context file in your project when present. Deliverables are written by the agent in chat or to files you request.