Rapid Topic Discovery & Synthesis Recipe

Quick overview

This is a 30–48 hour field recipe for rapidly surfacing cross-functional, operational, or service opportunities that usual dashboards miss. The approach blends human interviews, fast artifact harvesting, a small targeted data pull, pattern-clustering (optionally assisted by AI), and a light scoring step to produce one-page evidence-based opportunity briefs you can action in a 90‑day cycle.

When to use it

Use this recipe when you need focused, evidence-backed opportunities quickly—for example before a leadership prioritization meeting, to seed a 90‑day improvement backlog, or when you suspect cross-silo problems but lack a clear starting point.

What you need

• Team: 2–4 people (facilitator/analyst + 1–2 interviewers + optional data person)
• Timebox: 30–48 hours total (see phase breakdown below)
• Tools: video/notes tool, spreadsheet, lightweight analysis tools (e.g., Excel, Google Sheets, Python/R for small pulls), optional topic-clustering tool or AI assistant
• Artifacts: process maps, KPIs, incident logs, customer complaints, recent audit reports, meeting notes, contracts, product catalogs

Phase plan (timeboxed)

1. Plan & scope — 2–4 hours
   • Clarify scope (units, processes, customer segments, timeframe).
   • Identify 6–12 micro-interview stakeholders across functions (frontline, ops, sales, support, product, procurement, etc.).
   • Assign roles and confirm available artifacts and data owners.
2. Discover — 10–16 hours
   • Conduct 6–12 micro-interviews (10–20 minutes each). Capture short notes and the 1–2 most important examples or documents mentioned.
   • Scrape artifacts using the checklist below. Assemble a small, shareable artifact bundle.
3. Pull & prepare data — 4–8 hours
   • Run a lightweight data pull focused on a few high-value tables/logs identified in interviews (e.g., incident counts by product/business unit, lead times, complaint counts, rework events).
   • Obey privacy and access rules—use anonymized aggregates where appropriate.
4. Cluster & synthesize — 6–12 hours
   • Combine interview notes, artifact highlights, and data signals to cluster themes. Use simple manual coding in a spreadsheet or an automated topic-clustering tool to group similar issues.
   • Use safe, transparent AI prompts for suggestion (see safe-AI prompts below) but always validate AI outputs against source quotes and data.
5. Score & brief — 2–4 hours
   • Score top candidate opportunities using the rubric below (impact, effort, confidence, cross-functional reach).
   • Create 1-page opportunity briefs for the prioritized items and propose next steps (owners, 90‑day experiment idea, metrics).

Micro-interview script (10–20 minutes)

Use this short script to keep interviews focused and evidence-rich.

1. What process or outcome worries you most right now? (Ask for one specific recent example.)
2. Who or what is affected when that happens? (Customers, teams, costs, quality)
3. Do you have data or documents that show this happens? Where are they stored?
4. Have attempts been made to fix this before? What changed and what didn’t work?
5. If we could change one thing in the next 90 days, what would you try?
6. Who else should we talk to or which artifact should we see next?

Artifact scraping checklist

• Process maps, SOPs, and RACI charts
• Recent internal incident/defect logs and root-cause notes
• Customer complaints, NPS verbatims, and support tickets
• Audit findings, regulatory issues, and legal exceptions
• Performance dashboards and KPI extracts (time series of key metrics)
• Contracts, SLAs, handoffs, and external dependencies
• Recent project retrospectives and meeting notes

Lightweight data pull guidance

Keep pulls narrow and hypothesis-driven:

• Define 2–4 KPIs or event types to extract (e.g., rework count by plant, order lead time percentiles by product line).
• Pull a manageable timeframe (last 3–6 months) and a reasonable sample size.
• Anonymize personal data and aggregate by group when privacy is a concern.
• Prefer simple descriptive stats and visualizations (counts, rates, histograms, trend lines).

Clustering & safe AI-assisted discovery

AI can help surface themes from text (interview notes, ticket verbatims) but treat outputs as hypotheses:

• Prep: normalize text (remove names / PII), keep original source pointers for every automated label.
• Prompt pattern (example): "Given these anonymized interview excerpts and ticket texts, suggest 6–10 concise topic labels and for each label list 3 representative excerpts and an estimated signal strength."
• Validate: for each AI-suggested topic, confirm at least two human-sourced evidence items (quote, artifact, or data point).
• Document provenance: keep a simple log of which themes came from interviews, which from data, and which were AI-suggested.

Scoring rubric (simple)

Score each candidate 0–5 in the four dimensions and compute a weighted total (example weights in parentheses):

1. Impact (0–5) — potential benefit if solved (weight 40%)
2. Effort (0–5) — estimated effort/cost (inverse scoring; lower effort = higher score) (weight 20%)
3. Confidence (0–5) — evidence strength (interviews, artifacts, data) (weight 25%)
4. Cross-functional reach (0–5) — number of org areas affected (weight 15%)

Example calculation: Total = 0.4*Impact + 0.2*Effort + 0.25*Confidence + 0.15*Reach. Rank candidates by total and pick the top 2–4 for brief creation.

1‑page opportunity brief template (use for each prioritized item)

Deliverables

• Prioritized list of 6–10 candidate opportunities with scores
• 2–4 one-page opportunity briefs ready for prioritization
• Evidence bundle (interview notes, artifact index, data extracts) and a provenance log
• Suggested owners and 90‑day experiment proposals

Common pitfalls & how to avoid them

• Relying on AI outputs without source validation — always attach original quotes/artifacts to a theme.
• Scope creep — keep the initial scope tight and defer broader problems to later cycles.
• Data-induced false leads — prefer simple descriptive checks and flag correlations as hypotheses not facts.
• Privacy and access oversights — anonymize and get permission before sharing personal data.

How to make this reusable

Store interview templates, the artifact checklist, scoring sheet, and one-page brief as reusable templates in your team’s toolkit. After each run, log what changed and iterate the recipe.

Example mini-case (illustrative)

Interview signals and ticket verbatims pointed to delayed supplier confirmations. A small data pull confirmed a high proportion of orders waiting >48 hours. Clustered themes suggested a mismatch between order formats and supplier EDI capabilities. Scored high on impact, medium on effort, high confidence. Proposed 90‑day experiment: a single product line pilot with a simplified order template and dedicated supplier communications owner; expected metric: reduction in order confirmation lead time by 50%.

Safe-AI prompt examples (copy-and-adapt)

"Given these anonymized interview excerpts and ticket verbatims, propose up to 8 concise topic labels. For each label, list two representative excerpts and estimate whether the evidence is interview-based, artifact-based, or data-based."

"Summarize this cluster into a one-line problem statement suitable for a one-page brief and suggest a measurable hypothesis for a 90‑day test."

Next steps and handoff

After creating the briefs, present top items to decision-makers with the evidence bundle and recommend which items to take into a 90‑day experiment pipeline. Capture feedback and schedule owners for the experiments.

Image idea

Search phrase: rapid synthesis workshop