# AI Transformation Workshop > Facilitates a structured, multi-stage AI use-case discovery workshop for enterprises. Use this skill whenever a user asks for help with AI transformation planning, AI adoption strategy, finding/scoping AI use cases for their business, prioritizing AI initiatives, building an AI roadmap, or running an AI opportunity assessment. Trigger phrases include: "AI transformation", "where can AI help my business", "AI use case discovery", "AI roadmap", "AI readiness", "AI opportunity assessment", "prioritize AI projects", "数字化转型", "AI转型", "AI落地场景". The skill interviews the user stage by stage (business context, pain points, use-case ideation, value x feasibility scoring, roadmap), translates every recommendation into money (revenue upside, cost savings, probability-weighted risk) and produces an interactive HTML report with a full ROI analysis. - Skill: `laurenceshan/ai-transformation-workshop` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add laurenceshan/ai-transformation-workshop` - Raw SKILL.md: https://api.skillmd.com/api/skills/laurenceshan/ai-transformation-workshop/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Product & Planning - Author: laurenceshan (https://skillmd.com/u/laurenceshan) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/laurenceshan/ai-transformation-workshop --- # AI Transformation Workshop Facilitate a five-stage interview that turns vague "we should do AI" ambition into a scored, prioritized portfolio of concrete AI use cases and a pragmatic roadmap. ## Audience Three user profiles — adapt depth and vocabulary accordingly: - **Executives / business owners**: talk outcomes, money, risk. Minimal jargon. - **Consultants / FDE-style engineers**: expect structure, frameworks, artifacts they can reuse with clients. - **IT / digitalization teams**: care about data readiness, integration, effort. ## Core Principles 1. **Business first, technology second.** Every use case must attach to a named business process, a pain owner, and a measurable outcome. Reject "use LLM somewhere" answers. 2. **Numbers come from the customer.** Never invent revenue, conversion, pricing, or headcount figures. Critical numbers are stated by the user; secondary parameters may be proposed but must be confirmed. Every figure in the final report carries a source tag: user-confirmed or pending. 3. **One stage at a time.** Do not dump all questions at once. Finish each stage, summarize what was learned, get confirmation, then move on. 4. **Language adaptive.** Conduct the interview and write all deliverables in the user's own language, even though this skill is written in English. ## Workflow Run Stage 0 first, then the five stages in order. Read `references/financial-baseline.md` before Stage 0 and `references/interview-playbook.md` before Stage 1. ### Stage 0 — Financial Baseline (P&L first) Three setup confirmations: role & scope (whole company vs single department), precision mode (rigorous vs rough — confirmation rules apply in both), and attachments (financial statements, annual reports, ledgers; for returning users, also the previous report). Then decompose the user's P&L: revenue lines vs cost categories (labor / external spend / capital). The deliverable is a confirmed baseline table plus one verdict: is the business revenue-constrained or cost-heavy? Adapt all later questioning to the dominant cost category (see `financial-baseline.md`). If no attachment is available, build the simplified baseline through Stage 1 Q&A instead. ### Stage 1 — Business Context Establish: industry, company size, core business model, main revenue drivers, current data/IT maturity, any AI attempts so far (and why they failed or stalled). If the user's industry matches a file in `references/` (e.g. `industry-retail.md`), load it now — it shapes Stages 2-3. Confirm the currency for all figures (default USD, user may override). If Stage 0 did not already fix hourly rates of key people, ask for them here — never assume silently. ### Stage 2 — Pain Point Collection Map the value chain with the user and collect pain points: repetitive manual work, information bottlenecks, quality/consistency issues, slow decisions, knowledge locked in people's heads. Aim for 5-10 raw pain points, each tied to a process and a rough cost (hours/week, error rate, revenue at risk). ### Stage 3 — Use-Case Ideation Convert pain points into candidate AI use cases. Use the industry scenario library (if loaded) as seeds — propose library candidates the user has not mentioned and let them accept, reject, or adapt. Target 6-12 candidates. Each candidate gets a one-sentence definition: *who* does *what* with AI assistance to achieve *which outcome*. ### Stage 4 — Value x Feasibility Scoring Score every candidate on the two axes in `references/scoring-rubric.md` (read it before this stage). Value is scored ONLY AFTER the financial estimate for the candidate is complete and user-confirmed: the 1-5 anchors are relative to the user's own business (share of revenue or cost affected), not absolute thresholds. Feasibility: data availability, technical maturity, integration effort, organizational readiness. Produce a 2x2 classification: Quick Wins / Big Bets / Fill-ins / Money Pits. ### Stage 5 — Roadmap & Report Recommend a sequence: 1-3 Quick Wins to start within 90 days, 1-2 Big Bets to scope next, explicit "not now" list. Then produce the two deliverables. ## Deliverables Produce ONE interactive HTML report (self-contained single file), in the user's language. The report's six sections form the **SCORER framework**: | Letter | Section | Content | |---|---|---| | **S** | Summary | one-paragraph verdict + top 3 recommendations | | **C** | Context & Pains | business context recap merged into the pain point map | | **O** | Opportunities | the scored opportunity scorecard | | **R** | Returns | the ROI / money analysis | | **E** | Execution | the 90-day action plan | | **R** | Risks | risks and open questions | - **Layout**: fixed top navigation listing the six SCORER sections; clicking a nav item switches the content panel below. The page itself must never show a vertical scrollbar (only a panel may scroll internally if its content overflows). The layout must be responsive: desktop first, with tablet and phone breakpoints (nav may scroll horizontally, tables may scroll inside their own container, multi-column blocks stack to a single column) — the no-page-scroll rule holds at every viewport. The footer must credit the source repository (github.com/laurenceshan/ai-transformation-workshop), its license, and the line "Built with the SCORER framework". - **Summary section**: quote the headline ROI figures (annual net benefit, ROI, payback period) with a one-line interpretation, a visible disclaimer that all figures derive from estimates agreed during the workshop and are not guarantees, and a link that jumps to the Returns section. If the report recalculates ROI from adjustable inputs, the Summary figures must update in sync. - **Context & Pains section**: open with a compact business-context recap (industry, model, size, data maturity, prior AI attempts), then the pain point map. Assign every pain point an urgency level (e.g. Critical / High / Medium / Low — define the criteria: rate of time or money bleed x proximity to revenue), color-code on a red-to-green scale (red = most urgent), and sort the list from most to least urgent. - **Opportunities section**: table with use case, value score (1-5), feasibility score (1-5), quadrant, expected impact in money, first step. Sorted by priority. Include the 1-5 anchor definitions for both axes and the quadrant definitions, and render the quadrants as a 2x2 grid showing where each candidate lands. - **Returns section** (the signature feature): follow `references/money-model.md`. Show revenue, cost, and risk as separate blocks with their formulas; make currency, hourly rate, hours saved, and risk probability user-adjustable inputs with live recalculation where the medium allows. Present ROI as three tiers (conservative / expected / optimistic), defaulting to the conservative tier — a single optimistic point estimate is a known failure mode. State ROI and payback period per Quick Win and for the portfolio. Every input figure carries a source tag: user-confirmed or pending confirmation. If the environment cannot render files, fall back to inline markdown with the same section structure — but the money quantification is not optional. ## Resources - `references/financial-baseline.md` — Stage 0 guide: role & scope, precision mode, attachment intake, P&L decomposition, cost-structure-adaptive questioning, returning-user verification. - `references/interview-playbook.md` — per-stage question scripts, facilitation rules, and how to handle stuck or over-enthusiastic users. - `references/scoring-rubric.md` — scoring anchors for value (relative to business size) and feasibility, quadrant thresholds, worked example. - `references/money-model.md` — the ROI methodology: tiered number confirmation, revenue/cost/risk formulas, three-tier presentation, worked example. - `references/industry-template.md` — template for adding a new industry scenario library. - `references/industry-retail.md` — example library: retail & e-commerce.