Overview
Convert technical explanations, data, and architecture information into a format that executives and policymakers can use for investment decisions and decision-making. Translate "what is possible" (technology) into "what changes" (business / social impact).
Abstraction Ladder (Abstraction Levels)
Convert the input technical information to the appropriate level according to the target audience:
| Level | Target Audience | Focus of Expression | Example |
|---|---|---|---|
| Level 1 | Engineers | Implementation details, API, architecture | "Build a GPU cluster on Kubernetes and operate an inference server with vLLM" |
| Level 2 | Manager | Features, benefits, timeline | "Establish an AI model operation foundation, making it immediately available to researchers" |
| Level 3 | Executives (CxO) | Business impact, ROI, competitive advantage | "Achieve scalability of the research computing foundation and optimization of operational costs" |
| Level 4 | Policymakers | National strategic significance, social impact | "Build a research infrastructure foundation with international competitiveness" |
Conversion Pattern Collection
AI for Science Domain
| Technical Expression (Level 1) | Executive/Policy-oriented Expression (Level 3-4) |
|---|---|
| Turn ARIM data into RAG | Transform Japan's research data assets into a national-level knowledge foundation usable for AI |
| ML model accuracy of 92% | Achieve a 40% reduction in materials exploration costs compared with conventional methods |
| Build a GPU cluster on Kubernetes | Scalability of the research computing foundation and optimization of operational costs |
| Predict molecular properties with Graph Neural Network | Shorten the new material discovery cycle from years to weeks |
| Implement MCP+A2A protocol | Dramatically improve research productivity with a secure collaboration foundation between heterogeneous AIs |
Microsoft Technology Domain
| Technical Expression (Level 1) | Executive-oriented Expression (Level 3) |
|---|---|
| Azure OpenAI Service の Fine-tuning | AI optimized with company data improves operational efficiency by 30% |
| GitHub Copilot code completion | 40% improvement in developer productivity (GitHub survey data) |
| Data integration with Microsoft Fabric | Accelerate decision-making speed through unified management of siloed data |
Execution Procedure
Step 1: Input Analysis
- Read the input materials (technical documents, research-notes.md, storyline.md)
- Determine the current Abstraction Level of each piece of information
- Identify the target audience's Level from
requirements.mdfrom - List the points that need conversion
Step 2: Audience Level Assessment
Select a conversion pattern according to the target audience:
| Audience Type | Conversion Target Level | Emphasized Elements |
|---|---|---|
| CTO / Technical Executive | Level 2-3 | Technical Accuracy + Business Benefits |
| CFO / Finance Executive | Level 3 | ROI, cost reduction, payback period |
| CEO / President | Level 3 | Competitive advantage, market position |
| MEXT / Policy Officials | Level 4 | Alignment with national strategy, social impact |
| Board of Directors | Level 3-4 | Balance of risk and return |
Step 3: Execute Conversion
Apply the following conversion process to each piece of technical information:
1. 技術要素の特定: 「何の技術か」
2. 価値の抽出: 「その技術で何が改善されるか」
3. 定量化: 「どれだけ改善されるか」(可能な場合)
4. ビジネス言語化: 「経営指標でどう表現するか」
5. 文脈付与: 「なぜ今重要か」
Step 4: Generate Executive Summary
Output template: executive-summary.md
# Executive Summary: {タイトル}
## 提言の要旨
{3行以内で「何を」「なぜ」「どうするか」を記述}
## ビジネスインパクト
| 指標 | 現状 | 提案後 | 改善幅 |
|------|------|--------|--------|
| {KPI名} | {現在値} | {期待値} | {改善率} |
## 投資概要
| 項目 | 内容 |
|------|------|
| 初期投資 | {金額 or 規模感} |
| 運用コスト | {月額/年額} |
| 投資回収期間 | {N ヶ月/年} |
| 期待 ROI | {N% or 定性表現} |
## 競争環境
{競合との比較 — 自社/提案のポジショニング}
## リスクと対策
| リスク | 影響度 | 対策 |
|--------|--------|------|
| {risk} | 高/中/低 | {mitigation} |
## 推奨アクション
{意思決定者に求める具体的なアクション — 動詞で記述}
## タイムライン
{主要マイルストーン — 3〜5 項目}
## 参考資料
{詳細資料への参照リンク}
Rules for Handling Numerical Values During Conversion
When numerical values should be retained
- Quantitative data exists: “Accuracy 92%” → “40% cost reduction compared with conventional methods” (numeric conversion)
- Official statistics are available: “40% productivity improvement according to a GitHub survey” (citation with source)
- ROI can be calculated: Specific investment return figures
When expressing numerical values as estimates
- Exact figures are unknown: Instead of “significant improvement,” use “estimated 20-30% improvement” (express as a range)
- Conversion to a different metric: Note the original numerical value and the conversion logic
When not using numerical values
- Not comparable: For entirely new value, use qualitative expression
- Confidential data: When specific amounts cannot be disclosed, express as “on the order of XX billion yen”
Validation Loop
Conversion Quality Check
1. executive-summary.md の各セクションを検証:
a. 提言の要旨が 3 行以内か
b. ビジネスインパクト表に定量指標があるか
c. 推奨アクションが動詞で記述されているか
2. 技術的正確性の検証:
a. 変換後の表現が技術的に不正確でないか
b. 元の技術情報と変換後のビジネス表現が矛盾しないか
c. 数値の変換ロジックが妥当か
3. 聴衆適合性の検証:
a. Abstraction Level がターゲットに適合しているか
b. 専門用語が聴衆の知識レベルに合っているか
c. CTO 向けと CFO 向けで適切に使い分けているか
4. 全チェック通過 → ⏸️ ユーザー承認 → Phase 3 へ引き渡し
Failure Recovery
| Failure Pattern | Cause | Recovery Procedure |
|---|---|---|
| Excessive abstraction | Raised the Level too much | Add at least one specific numerical value or example |
| Technical inaccuracy | Meaning changed during conversion | List alongside the original technical expression and request review from an engineer |
| ROI cannot be calculated | Insufficient data | Explicitly state “additional data is required for quantification” and substitute a qualitative expression |
| Audience level misjudgment | Insufficient audience information in requirements.md | Create it at an intermediate Level 2-3 and confirm with the user |
| Basis for numerical values unknown | Numerical values without sources | Delete the numerical values and change to qualitative expression. Propose source investigation |
Quality Gates
- The key points of the recommendation are completed within 3 lines
- There is at least one quantitative metric in the business impact table
- Recommended actions are described with specific verbs
- The converted expression is technically accurate (does not contradict the original information)
- The Abstraction Level matches the target audience
- Numerical values have their source or conversion logic explicitly stated
- At least one risk and countermeasure is listed
- The timeline has at least 3 major milestones
- Confidential information is handled appropriately (redacted or removed)
Gotchas (pitfalls)
1. Excessive abstraction — loss of concreteness
If you raise the Abstraction Level too much, the material becomes "vague" and can no longer be used for decision-making.Always leave at least one concrete numerical value even after conversion. Maintain quantitative expressions such as "estimated 30% cost reduction" rather than "significant improvement." The key is not to lose "verifiability" even when abstracting.
2. Technical lies — semantic drift during conversion
Converting "ML model accuracy 92%" into "100% accurate predictions" is atechnical lie. Always verify that the converted expression does not contradict the original technical facts. Especially with AI-related content, pay attention to expressions of "accuracy" and "performance." Exaggeration destroys credibility.
3. Misjudging the audience level — the difference between CTO and CFO
For a CTO, Level 2-3 (business benefits while maintaining technical accuracy) is appropriate, while for a CFO, Level 3 (focused on ROI and financial metrics) is appropriate. Even within the same "executive team," the optimal conversion pattern differs by role. Always refer to the audience analysis in requirements.md.
4. Omission of "the cost of not investing"
Including not only ROI but also "the risk/cost of maintaining the status quo" greatly increases persuasiveness. Executives tend to respond more to "the risk of not doing it" than to "the reason to do it." This is especially essential in areas with intense competitive environments.
5. Unique requirements for policymakers
For MEXT / Cabinet Office,consistency with national strategy documents (such as the Science, Technology, and Innovation Basic Plan)must be made explicit. General business ROI will not resonate. Be mindful of the keywords "international competitiveness," "strengthening research capabilities," and "human resource development."
Memory Persistence
Patterns learned from past conversions:
- Successful pattern for MEXT: Discuss technology in the context of "strengthening research capabilities" and "international competitiveness." For budget requests, presenting "the risk of declining international standing if we do not invest" is effective
- For Microsoft HQ: Present "Market Opportunity Size" first. The TAM/SAM/SOM framework is preferred
- For CFO: A 3-year TCO (Total Cost of Ownership) comparison is the most persuasive. Present not only ROI but also the "Break-even point"
- For CTOs: Do not sacrifice technical accuracy. Level 2 is sufficient. Keep the architecture diagram, and omit only the implementation details
- Examples of failed transformations: When "Transformer architecture" was transformed into "innovative AI", trust was lost with engineers. Keep technical names while adding value
Prohibited items
- No technical lies — The transformed expression must not contradict the original technical facts
- No numbers without evidence — Do not include numbers without sources or transformation logic
- No excessive abstraction — Maintain at least one concrete number after transformation
- No ignoring audience level — Do not transform without referring to the audience analysis in requirements.md
- No unauthorized inclusion of confidential information — Confirm publication permission for internal financial data and unpublished plans
- No disparaging competitors — Do not unfairly belittle competitors in descriptions of the competitive landscape
- No fabricating sources — Do not cite reports or statistics that do not exist