Workflow · Research synthesis
From customer data to opportunities
Real work Rebuilt from a Claude plugin I built for a content team
Hundreds of survey comments, reviews, support conversations and interview notes become a short list of behaviours and opportunities, each with its evidence. An artificial intelligence (AI) model does the reading and the first tagging. A person checks the tagging, ranks the opportunities, and owns the brief.
Try it
- Select Run the synthesis.
- Read how the tagging was checked, and the behaviours it found.
- Confirm the priorities, and the workflow writes the brief.
Not run yet
What this workflow cannot tell you
- How many customers have a problem. Counts show what people said, and people who write in are not everyone.
- Why people behave as they do. That needs interviews and observation, which the brief asks for.
- Whether an opportunity is worth building. Impact and effort here are a first guess for the team to test.
Safeguards
What keeps the workflow safe to run on a real knowledge base.
- Personal data removed firstNames, emails, order numbers and card details are masked before any item is tagged or quoted.
- A codebook before the taggingEvery theme has a written definition and an example before the AI tags anything, so the tags mean the same thing every run.
- A person checks the taggingA person tags a sample by hand. If agreement is under 90%, the run stops until the codebook is fixed.
- Two sources or it is not an insightA behaviour needs evidence from at least two sources, so one loud channel cannot set the agenda.
- Quotes are word for wordEvery quote keeps its source and item number, so anyone can find the original.
- A person ranks the opportunitiesThe workflow suggests a ranking. A person decides it before anything is written.
- The brief is a draftIt goes to the team for review, with its confidence and what is still unknown.
Take it to the next level
How to get results that match the best teams, and what to build next.
- Tag with two modelsRun the tagging in Claude and Gemini and compare them. Where they disagree, the codebook needs work.
- Add what people doJoin the comments to product data, such as drop-off and repeat contacts, to find where words and actions differ.
- Keep a living libraryStore every tagged item and insight in one place, so the next study starts from what is known.
- Track themes over timeRe-run monthly and show which themes grow or fade after each release.
- Synthesise togetherUse the AI output as the starting wall for a team session, not as the answer.
- Tie each opportunity to a measureName the number that should move, such as contacts about payouts, before work starts.
Rebuilt from a Claude plugin I built for a real content team. Keystone is an invented payments product, and every name, figure and conversation here is made up.