Use Case Discovery

FlashFusion consultant-led workshop that turns a client's pain points into a ranked backlog of 10–15 AI/automation use cases — each scored for feasibility, estimated for impact, and prioritized by WSJF. You run this as the facilitator: collect, score, estimate, rank, hand back a defensible backlog.

Evals

Should trigger:

1. "We're meeting the ops team at Northwind next week to figure out where AI could help — can you help me run the session?"
2. "I've got a list of 12 things the service desk wants to automate. Help me rank them so we know what to build first."
3. "Run an AI opportunity assessment for a 40-person finance department."

Should NOT trigger:

1. "Write the proposal and statement of work for the Northwind engagement." → that's scoping-ai-engagements (proposal/PRD/phase plan), not discovery.
2. "Build the ticket-to-article KB automation we scoped." → that's building-kb-pipelines (implementation of one chosen use case), not opportunity discovery.

Decision Tree

What does the client have?
  ├─ Nothing yet — "where could we use AI?" ──────── FULL WORKSHOP. Run all 5 steps (intake → ranked backlog).
  ├─ A vague pain area / one department ──────────── SCOPED WORKSHOP. Run steps 1–5 bounded to that department.
  ├─ An existing list of candidate use cases ─────── SCORE-ONLY. Skip step 2 collection; start at step 3 (feasibility) on their list.
  └─ One specific use case they've already chosen ── EXIT. They're past discovery → hand to scoping-ai-engagements.

If you cannot tell which: ask "Are we discovering opportunities from scratch, or ranking a list you already have?"

Intake (Step 1)

Collect these before scoring anything. Do not proceed on blanks — ask.

Field	Why it matters
Client / org name	Labels the backlog; sets tone
Industry / vertical	Calibrates feasibility (regulated data lowers data-availability scores)
Team / department size	Scales the hours-saved math
Top pain point (the one that triggered this)	Anchors the session; becomes use case #1 candidate
Workshop scope	Whole org, one department, or score-only
Data environment (optional)	Where the data lives, sensitivity, access — feeds feasibility

Capture verbatim. The "top pain point" is the thread you pull — most other use cases surface as adjacent friction.

Golden Path

1. Intake — collect the table above. Confirm scope via the decision tree.
2. Pain Point Collection — elicit 12–20 raw pains. Prompt by category: manual/repetitive work, slow handoffs, error-prone steps, knowledge that lives in one person's head, things that don't scale. One line each: who hits it, how often, what it blocks. Deduplicate into distinct candidate use cases.
3. Feasibility Scoring — score each candidate 1–5 on three axes (rubric below). Compute a feasibility composite.
4. Impact Estimation — for each candidate, estimate hours/week recovered and a revenue/cost delta (rubric below).
5. WSJF Prioritization — compute WSJF per candidate, sort descending, trim to the top 10–15. Produce the backlog in the output schema.

Feasibility Scoring (Step 3)

Score each use case 1–5 on each axis. 5 is always the most favorable.

Axis	1 (worst)	3	5 (best)
Data availability	Data doesn't exist / inaccessible / highly sensitive with no clearance	Exists but messy, partial, or needs cleanup	Clean, accessible, already structured
Implementation complexity	Multi-system, custom integration, new infra	Moderate — one integration, known pattern	Off-the-shelf / prompt-level / single tool
Stakeholder readiness	No sponsor, change-resistant, no time	Interested but unproven, needs convincing	Named sponsor, eager, willing to pilot

Feasibility composite = average of the three axes (1.0–5.0). Anything scoring data availability ≤ 2 is flagged DATA-BLOCKED and noted regardless of impact — it cannot start until data is sorted.

Impact Estimation (Step 4)

For each use case estimate two things. Show your assumptions inline.

• Hours/week recovered = (people affected) × (times performed/week) × (minutes each ÷ 60) × (automatable fraction). State each factor.
• Revenue/cost delta = annualized money effect. Pick the dominant lever and label it:
  • Cost-out: hours/week × 52 × loaded hourly rate → annual labor recovered. [FILL: client loaded hourly rate] if unknown.
  • Revenue-enabling: faster cycle time → throughput or conversion lift. Estimate conservatively; label [GENERATED].
  • Risk/error reduction: cost of current error rate × reduction. Use only if the client can quantify the error cost.

Never invent precise dollar figures. If an input is missing, insert [FILL: …] and carry the use case with hours-only impact.

WSJF Prioritization (Step 5)

Cost of Delay over Job Size — consistent with scoping-ai-engagements.

WSJF = (Impact Value + Time Criticality + Risk Reduction) / Job Size

• Impact Value — map the impact estimate to Fibonacci 1-2-3-5-8-13-20 (bigger hours/$ → higher).
• Time Criticality — Fibonacci. How much does delaying erode the value? (seasonal, contractual, competitive).
• Risk Reduction — Fibonacci. Does it de-risk compliance, key-person dependency, or error exposure?
• Job Size — Fibonacci, derived from feasibility: Job Size ≈ 6 − round(feasibility composite) (high feasibility → small job). Floor at 1.

Sort descending. Trim to top 10–15. Anything WSJF < 2.0 drops to a "parking lot" note, not the backlog.

Output Schema

Return a single document with these named sections:

1. Header — Client, industry, team size, workshop scope, date, facilitator.

2. Discovery Summary — 2–3 sentences: the top pain point and the headline opportunity.

3. Ranked Backlog — a table, sorted by WSJF descending, 10–15 rows. Required columns:

   #	Use Case	Pain Addressed	Feasibility (avg)	Hours/wk	Revenue/Cost Δ	WSJF	Flags

   Flags carries DATA-BLOCKED, [FILL: …], or parking-lot. Each $ figure labeled [SOURCED] or [GENERATED].

4. Top 3 Recommendations — for the highest-WSJF items: one paragraph each on why it's first, what's needed to start, and the obvious quick win.

5. Parking Lot — use cases scored but below the WSJF cut, with the one-line reason.

6. Open Data Needs — every [FILL: …] collected in one list so the client knows exactly what to provide.

Optional fields (include when known): per-use-case sponsor name, target start phase, dependency on another use case.

Self-Verification Checklist

Before returning the backlog:

1. Every use case scored on all three feasibility axes (no blanks).
2. Every use case has hours/week AND a revenue/cost line (or an explicit [FILL: …]).
3. Every WSJF has all four inputs shown — no bare final numbers.
4. Backlog is sorted by WSJF descending and trimmed to 10–15.
5. Every dollar figure labeled [SOURCED] or [GENERATED].
6. DATA-BLOCKED flag applied wherever data availability ≤ 2.
7. Open Data Needs lists every [FILL: …] in the document.
8. No invented client facts, rates, or benchmarks presented as known.

CLAIMS

• WSJF (Cost of Delay ÷ Job Size, Fibonacci) is an established SAFe prioritization method. [SOURCED — Scaled Agile Framework]
• The 1–5 three-axis feasibility rubric (data / complexity / readiness) is a FlashFusion convention defined in this skill. [GENERATED]
• Deriving Job Size from the feasibility composite (6 − round(composite)) is a convenience mapping defined here, not an external standard. [GENERATED]

COUNTEREXAMPLE

• Do NOT use for scoping/proposing a chosen engagement (→ scoping-ai-engagements) or implementing one use case (→ building-kb-pipelines).
• A high-WSJF use case that is DATA-BLOCKED should not be recommended as "start now" — sequence the data work first.

CONTRADICTIONS

• ! Impact Estimation wants precise ROI, but the no-fabrication rule forbids inventing rates — resolve by carrying [FILL: …] and reporting hours-only until the client supplies the figure.