Skip to content
Beta Multi-model support (OpenAI, Gemini, and more) launching soon.
RBA RBA Skills

discovery-interview-synthesis

Installable Ships in discovery-synthesizer — or copy the raw skill below.
Open Claude.ai

Paste into Claude Code, Claude.ai, the desktop app, or the browser extension — ⌘/Ctrl+V.

ExperimentalNo security review

Discovery Interview Synthesis

Transform raw qualitative discovery material into a structured, evidence-grounded insights report. The output is a markdown document ready for internal review and iteration before client sharing.

When this skill applies

  • User pastes interview transcripts, workshop notes, or session summaries
  • User provides a file path to a transcript, notes document, or survey export
  • User asks to "synthesize," "theme," "analyze," or "report on" discovery material
  • Post-workshop cleanup: turning a messy Miro board export or FigJam dump into structured findings

Skip when the user only wants a single artifact type: stakeholder map → stakeholder-map; root cause → five-whys-root-cause; persona → persona-draft; requirements register → requirements-workshop.

Workflow

Step 1 — Load and orient

Read all provided files or pasted text. Extract:

  • Number of participants (deduplicate by role, not by session)
  • Session types present (interview / workshop / survey / observation)
  • RBA engagement context if stated (platform-selection, pre-envision, experience design, managed services)
  • Target platform(s) mentioned — flag Sitecore, Optimizely, Umbraco, Azure, or composable signals
  • Date range of sessions

Summarize what you found and confirm with the user before coding.

Step 2 — Clean and structure the raw material

For each source document:

  1. Identify the participant's role (e.g., Marketing Director, IT Lead, Content Author, End User)
  2. Separate facilitator prompts from participant responses
  3. Flag direct quotes — mark with [QUOTE] tag for retrieval in later steps
  4. Mark observations vs. opinions vs. explicit requirements statements
  5. Note emotional tone where it is legible (frustrated, enthusiastic, uncertain)

Do not interpret yet. This step is transcription hygiene only.

Step 3 — Apply thematic coding

Code the cleaned material against the standard RBA discovery coding framework (see reference.md § Coding Framework). Apply codes line by line or segment by segment:

  • Descriptive codes — what the participant described (e.g., current-workflow, pain-point, workaround)
  • Interpretive codes — what it means for the engagement (e.g., content-governance-gap, platform-capability-mismatch, user-trust-deficit)
  • Evaluative codes — significance or frequency (e.g., high-frequency, single-source, outlier)

Every quote tagged [QUOTE] must receive at least one code. Outliers — single-participant observations that contradict the emerging pattern — get their own [OUTLIER] flag rather than being subsumed into a theme.

Step 4 — Build the affinity cluster

Group codes into themes. Minimum viable cluster: at least two participants or two independent data sources contribute to the same code for it to become a theme. Single-participant codes stay flagged as outliers unless they carry strategic weight (e.g., a C-suite constraint).

Name each theme in plain language, not jargon. Bad: "Multi-channel content orchestration deficiency." Good: "Authors cannot publish consistently across channels without IT involvement."

For each theme record:

  • Theme name (plain-language statement)
  • Supporting evidence count (N participants, N quotes)
  • Confidence level: High (3+ independent sources), Medium (2 sources or 1 strong source), Low (1 source, insufficient corroboration)
  • Representative quotes (2–3 best quotes; always include participant role, never personal name)
  • Design/platform implications

Step 5 — Identify opportunity areas

Map each high-confidence theme to an opportunity area. An opportunity area is a direction for intervention — not a feature, not a solution. It answers: "What does this theme tell us we need to create room for?"

Use the framing: "People need a way to [verb] [object] [context]" — this is a JTBD-compatible framing that connects to persona drafting and requirements work downstream.

Step 6 — Draft the insights report

Use the structure in templates.md. Sections in order:

  1. Executive summary — 3–5 bullets, outcome-leading
  2. Methodology — session types, participant count, date range, recruitment approach if known
  3. Key findings — one section per theme, with confidence rating, supporting evidence, representative quotes, and implications
  4. Opportunity areas — JTBD framing, cross-referenced to findings
  5. Outliers and weak signals — do not bury; these are often the most strategically important signals
  6. Recommended next steps — scoped to the RBA engagement phase (pre-envision, envision, design, build)
  7. Open questions — questions this research raised that require further investigation

Step 7 — Self-check before returning

  • Every claim traces to at least one coded piece of evidence
  • No participant is identified by name in any client-facing section
  • Outliers are surfaced, not suppressed
  • Platform implications use hedged language ("may require," "signals a need for") unless the evidence is conclusive
  • Executive summary could stand alone if separated from the body
  • The word count for the report body is reasonable for the data volume provided

Return the report as a markdown block. Remind the user to run a brand voice pass (rba-brand-voice) before client distribution.

Confidence rating guide

Rating Meaning
High 3 or more independent participants or sources corroborate the finding
Medium 2 independent sources, or 1 source with strong corroborating observation
Low Single source; include but flag explicitly — not suitable as a standalone design driver

Platform adapter flags

When these signals appear in transcripts, add a [PLATFORM NOTE] annotation in the relevant finding:

  • Sitecore — references to Experience Editor, Content Hub, Horizon, xConnect, or "our Sitecore" → flag content-editing experience and integration surface implications
  • Optimizely — references to CMS 12, Content Graph, or experimentation → flag headless-readiness and A/B testing governance implications
  • Umbraco — references to backoffice, Grid editor, or Heartcore → flag editor experience and headless-vs-MVC choice implications
  • Composable / headless — references to "API-first," "decoupled," or specific frontend frameworks → flag authoring experience and preview complexity implications

Platform notes do not change the finding; they flag what the platform team should read first.

Reference and template files

  • reference.md — full coding framework, evidence standards, outlier-handling rules
  • templates.md — report section templates with placeholder guidance
Technical details
Plugin
discovery-synthesizer
Author
Dakota Smith
Skill name
discovery-interview-synthesis
Maturity
experimental
Security review
No security review
Owners
  • strategy
Argument hint
[paste transcript text or @file-path] [--participants N] [--platform sitecore|optimizely|umbraco|composable] [--deliverable report|summary|slides-notes]
Allowed tools
  • Read
  • Write
  • Glob
Source
plugins/discovery-synthesizer/skills/discovery-interview-synthesis/SKILL.md

Edit this skill on GitHub — opens GitHub's web editor. No git knowledge needed.