Skip to content This library is free, offered as dana. If it helps your work, donate here →
penguin tree ai
← back to libraryDonate here

This is one of 3,622 free prompts in the penguin tree ai library. Sign up once and the whole library opens, with new prompts by email.

open the library →

AI use case evaluation

Evaluates a proposed AI use case across five dimensions — feasibility, ROI, risk, organizational readiness, and strategic alignment — producing scored criteria, a risk register with mitigations, an investment-grade…

download .md

Purpose

Convert a proposed AI use case into a structured, defensible evaluation that leadership can act on — scoring it across feasibility, ROI, risk, organizational readiness, and strategic alignment, then issuing an explicit go / iterate / no-go recommendation with the reasoning shown.

When to Use

Invoke this skill when the user asks to:

  • Evaluate or vet a proposed AI use case
  • Prioritize among multiple competing AI initiatives
  • Build a business case for an AI investment
  • Make a go/no-go decision on an AI project
  • Pressure-test an AI proposal before funding it
  • Identify the risks and readiness gaps in a planned AI rollout

Inputs Required

Before starting, collect:

  • Use case description: 3–5 sentences on what the AI will do, who uses it, and what decision or task it supports
  • Expected business outcome: The metric this is supposed to move (revenue, cost, cycle time, quality, risk reduction) with rough magnitude
  • Org context: Company size, industry, regulatory exposure, AI maturity (none / experimenting / production / scaled)
  • Data situation: What data exists, where it lives, quality level, and access constraints
  • Build approach (if known): Off-the-shelf / API + light build / custom model / agentic system / unknown
  • Budget and timeline: Investment ceiling and target deployment window
  • Stakeholders: Executive sponsor, end users, and groups that could block (legal, compliance, security, IT, labor)

If build approach or stakeholders are missing, infer reasonable assumptions and flag them in the output.

Instructions

Step 1: Restate the Use Case in One Paragraph

Before scoring, restate the proposal in plain language: who does what differently because of this AI, and what changes for the business. If the restatement reveals ambiguity (unclear user, unclear decision being supported, unclear success metric), surface it before scoring — an unscoreable proposal is itself a finding.

Step 2: Score Across Five Dimensions

Each dimension is scored 1–5 with a one-sentence justification. Anchors below.

Dimension 1 — Feasibility (technical) Can this actually be built and run reliably given current AI capabilities, data, and integration realities?

  • 5: Proven pattern, mature tooling, data is ready, clear integration path
  • 3: Achievable but requires meaningful data work, model tuning, or new infrastructure
  • 1: Pushes the frontier of current capability; data or integration blockers are severe

Consider: model capability fit, data availability and quality, latency requirements, integration complexity, evaluation/measurement design.

Dimension 2 — ROI (economic) Does the value clearly exceed the all-in cost over a reasonable horizon?

  • 5: >5x return within 12 months, with a credible measurement plan
  • 3: 2–3x return within 18 months, with some measurement uncertainty
  • 1: ROI is speculative, lagging, or dependent on adoption that isn't likely

Build a quick value calc: (users × frequency × time/error savings × loaded cost) − (build + license + run + change-management cost). Note assumptions explicitly.

Dimension 3 — Risk (downside exposure) What's the worst credible outcome and how likely is it?

  • 5: Low-stakes domain, reversible errors, no regulated data, no customer-facing exposure
  • 3: Meaningful but containable risk — internal use with sensitive data, or external with low decision authority
  • 1: High-stakes decisions, regulated industry, customer-facing autonomous action, or brand-critical output

Cover: hallucination/accuracy risk, data privacy and IP, regulatory exposure (GDPR, HIPAA, EU AI Act, sector-specific), security (prompt injection, data exfiltration), bias and fairness, vendor concentration, reputational risk.

Dimension 4 — Organizational Readiness Will the organization actually adopt and sustain this?

  • 5: Executive sponsor active, users want it, change-management capacity exists, governance is in place
  • 3: Sponsor exists but adoption requires workflow change; governance is forming
  • 1: No clear sponsor, users skeptical or threatened, no AI governance, no MLOps capability

Consider: sponsor strength, user pull vs. push, workflow disruption, training needs, AI governance maturity, MLOps/monitoring capability, labor or cultural friction.

Dimension 5 — Strategic Alignment Does this advance the company's stated strategy, or is it opportunistic?

  • 5: Directly advances a top-3 strategic priority; builds durable capability
  • 3: Supports a secondary priority or improves an existing capability
  • 1: Interesting but orthogonal to strategy; one-off win with no compounding value

Consider: link to strategic priorities, capability-building vs. point solution, competitive necessity, optionality created for future use cases.

Step 3: Compute the Composite and Apply Decision Rules

Sum the five scores (range 5–25). Apply these decision rules:

  • Go (20–25): Strong on all dimensions, no dimension below 3. Proceed to business case and resourcing.
  • Iterate (14–19): Promising but has 1–2 weak dimensions. Recommend specific reshape (narrower scope, different build approach, pilot first, fix readiness gap) before re-evaluation.
  • No-Go (5–13): Fundamental weakness. Reject or shelve. State what would have to change to revisit.
  • Hard No-Go (any single dimension = 1, especially Risk or Readiness): Override the composite. A score of 1 on Risk or Readiness is disqualifying regardless of other strengths.

Step 4: Build the Risk Register

List the top 5–7 risks specific to this use case. For each:

  • Risk: One-sentence description
  • Likelihood: Low / Medium / High
  • Impact: Low / Medium / High
  • Mitigation: Specific action, owner type (legal, security, product, ML), and timing (pre-launch / launch / post-launch)

Avoid generic risks ("model might be wrong"). Make them specific to this use case's domain, data, and exposure.

Step 5: Identify Readiness Gaps and Prerequisites

List what must be in place before this can ship, separated into:

  • Pre-build prerequisites: Data access, governance approval, legal review, sponsor confirmation
  • Pre-launch prerequisites: Evaluation harness, human-in-the-loop design, monitoring, rollback plan, user training
  • Post-launch prerequisites: Success metric tracking, feedback loop, model refresh cadence

Step 6: Recommend the Build Path

Given feasibility and ROI scores, recommend one of:

  • Off-the-shelf / SaaS: When the capability is commoditized and differentiation isn't in the AI itself
  • API + light orchestration: When foundation models suffice and value is in the workflow integration
  • Custom fine-tune or RAG system: When proprietary data or domain specificity is the moat
  • Agentic system: When the task requires multi-step reasoning and tool use — but only if monitoring and guardrails are mature
  • Pilot first, decide later: When uncertainty on feasibility or adoption is too high to commit

Name the recommended path and give the reasoning in 2–3 sentences.

Step 7: Issue the Final Recommendation

State the decision in one paragraph: Go / Iterate / No-Go, the composite score, the strongest reason for the decision, and the single most important next action. Be opinionated — don't hedge.

Output Format

  • Structure: Restatement → 5-dimension scorecard (table) → Composite + decision rule applied → Risk register (table) → Readiness gaps (3 lists) → Build path recommendation → Final recommendation paragraph
  • Length: 1,000–1,400 words
  • Format: Mix of prose, scorecard table, and risk register table — scannable for an exec, defensible for a reviewer
  • Sections: Use Case Restatement → Dimension Scorecard → Composite & Decision → Risk Register → Readiness Gaps & Prerequisites → Recommended Build Path → Final Recommendation
  • Tone: Senior strategy reviewer — analytical, direct, willing to say no; no AI hype, no hedging

Quality Checklist

Before delivering output, verify:

  • [ ] All five dimensions are scored 1–5 with a specific justification — not generic praise
  • [ ] Composite score is calculated and the correct decision rule is applied (including the single-dimension-of-1 override)
  • [ ] Risk register has 5–7 risks, each with likelihood, impact, and a specific mitigation owner — no generic risks
  • [ ] ROI dimension shows the value calc with assumptions stated, not just a qualitative claim
  • [ ] Readiness gaps are split into pre-build, pre-launch, and post-launch — not one undifferentiated list
  • [ ] Final recommendation is explicit (Go / Iterate / No-Go), not equivocal
  • [ ] Any inferred inputs (build approach, stakeholders) are flagged as assumptions

Notes

  • The most common failure mode in AI proposals is high feasibility + high ROI + low readiness — the tech works, the math works, but adoption never happens. Weight readiness heavily.
  • "Iterate" is the most useful recommendation in most real cases — outright Go and No-Go are rarer than they appear. Use Iterate to reshape scope rather than killing momentum on a viable idea.
  • A single-dimension score of 1 on Risk should almost always override a high composite. Regulated industries and customer-facing autonomous actions are where AI projects most often fail publicly.
  • If the user is comparing multiple use cases, run this skill once per use case and rank by composite score, breaking ties on Strategic Alignment.
  • Push back on vague success metrics during Step 1. A use case without a measurable outcome cannot be scored on ROI and should be sent back for reshaping before evaluation.

How to use this skill

This file is a Claude Code skill in Anthropic's standard format — markdown with YAML frontmatter at the top.

Installing in Claude Code:

  • ~/.claude/skills/<folder-name>/SKILL.md — auto-loads in every session anywhere on your machine (recommended for skills you'll use across multiple projects)
  • <your-project>/.claude/skills/<folder-name>/SKILL.md — auto-loads only when working in that project (recommended for project-specific skills)

Restart Claude Code or open a new session to register.

Using in Claude.ai, ChatGPT, Gemini, or other AI tools:

  1. Drag this file into a new chat, or copy and paste its contents
  2. Send your task or request as your next message — the AI will apply the skill where it fits

For Claude Code skill details: https://docs.anthropic.com/en/docs/claude-code/skills