# 山崎事務所 | Yamazaki Office

> AIが決めていいこと。人間が決め続けるべきこと。その境界線を設計する。複数の上場企業・HD企業のCxO直下で、AI・DX・データ戦略の意思決定とガバナンス設計を支援するFractional CxO。

This file is generated by `scripts/build-dist.sh` from the pages listed in `sitemap.xml`.
Do not edit it by hand — edit the page, and this file follows.

## Pages

- [山崎事務所](https://www.yamazaki-office.com/): AIが決めていいこと。人間が決め続けるべきこと。その境界線を設計する。複数の上場企業・HD企業のCxO直下で、AI・DX・データ戦略の意思決定とガバナンス設計を支援するFractional CxO。
- [About](https://www.yamazaki-office.com/about.html): ヤフー、電通、ストライプ、そして独立。山崎茂樹の経歴を事実ベースで記す。どこで何を任され、何を決めてきたか。資格・登壇歴も含め、経歴は判断の来歴として記録する。
- [関与のかたち（Engagement Model）](https://www.yamazaki-office.com/engagement.html): 顧問でも、コンサルでも、役員採用でもない。「経営の隣に座る」という第三の関わり方がある。肩書ではなく判断で関わるための、外部顧問としての作法を7つの型に整理した実践知。
- [判断の系譜（Judgment Lineage）](https://www.yamazaki-office.com/judgment-lineage.html): 実行者から、意思決定の設計者へ。ヤフー・電通・ストライプを経た20年のキャリアを、「担う判断の範囲と責任が広がり続けた軌跡」として語り直す。役職でなく判断の変遷を追う経歴。
- [Evidence](https://www.yamazaki-office.com/evidence.html): 守秘の壁の内側で行われた判断を、匿名化して6件記録した。AIガバナンス・DX組織・データと経営判断・複数部門をまたぐ変革の4テーマ。何をしたかではなく、何を決めたか。
- [Case 01｜流通・小売——承認ゲートの段階設計｜Evidence](https://www.yamazaki-office.com/evidence-case-01.html): 承認基準が存在しないまま要件定義が進む大型連携プロジェクト。全部を決めようとせず、スコープを最小の一点に絞り、3層の承認フレームワークを起案した。決めない範囲を先に決めるのも、設計である。
- [Case 02｜通信キャリア——AIモデル評価の独立監査｜Evidence](https://www.yamazaki-office.com/evidence-case-02.html): 解約予兆AIモデルの評価で、良い数字の解釈をまず疑い、母集団の偏りを指摘して拙速な成果認定を保留させた。AIガバナンスの実務は、都合の良い結果への懐疑から始まる。
- [Case 03｜DX開発部門——「忙しさ」の誤診断を退ける｜Evidence](https://www.yamazaki-office.com/evidence-case-03.html): DX組織の生産性低下の原因は、人手不足ではなく意思決定構造にあった。増員という処方を退け、仕事の仕分けを優先させる。誤診断のまま人を足せば、組織はさらに重くなる。
- [Case 04｜小売・EC——KPI変更承認の階層設計｜Evidence](https://www.yamazaki-office.com/evidence-case-04.html): 一見軽微なKPI変更が、収益に直結するリスクを持っていた。収益に影響する判断を現場だけに任せず、承認階層の設計に線を引いた。どこまでが現場の裁量かを決めるのが、ガバナンスである。
- [Case 05｜データ分析専門組織——採用戦略の再定義｜Evidence](https://www.yamazaki-office.com/evidence-case-05.html): 離職対応の採用を、短期の人繰りと中長期の組織防衛という二つの問いに切り分け、採用ターゲットを再定義した。頭数を足しても、勝ち筋は増えない。何のための採用かを、先に決める。
- [Case 06｜メディア企業——AI導入の統治ゲート設計｜Evidence](https://www.yamazaki-office.com/evidence-case-06.html): 経費申請ベースで既成事実化しつつあった生成AIツール。データ範囲・セキュリティ・契約の3ゲートを通すまで、実データに触れさせない統治を設計した。経費申請は、同意ではない。
- [Thinking](https://www.yamazaki-office.com/thinking.html): 隠すほうが楽だ。それでも公開する。AIガバナンスと組織の権限設計という、まだ誰も名前を付けていない領域で、実践者として発信していく。結論でなく、結論に至る手つきを残す場所。
- [Contact](https://www.yamazaki-office.com/contact.html): 経営課題の相談窓口。きれいに整理された依頼でなくていい。課題がまだ言葉になっていない段階の相談から、判断の設計は始まる。会社名・お役職・課題の概要だけで十分、山崎本人がすべて目を通す。
- [プライバシーポリシー](https://www.yamazaki-office.com/privacy.html): 株式会社山崎事務所（山崎事務所）のプライバシーポリシー。当サイトが取得する情報の範囲、取得方法、第三者提供の有無について記載しています。
- [Yamazaki Office](https://www.yamazaki-office.com/en/): What AI may decide. What people must continue to decide. We design the boundary. A Fractional CxO supporting decision-making and governance design for AI, DX, and data strategy directly under the CxOs of multiple listed companies and holding companies.
- [About](https://www.yamazaki-office.com/en/about.html): Yahoo!, Dentsu, Stripe, then independence. A factual record of Shigeki Yamazaki’s career: what he was entrusted with and what he decided. Qualifications and speaking history included, the career is recorded as a lineage of judgment.
- [Engagement Model](https://www.yamazaki-office.com/en/engagement.html): Neither advisor, consultant, nor executive hire. There is a third way of engaging: sit beside management. Practical knowledge organized into seven practices for external advisors who engage through judgment rather than title.
- [Judgment Lineage](https://www.yamazaki-office.com/en/judgment-lineage.html): From executor to designer of decision-making. A 20-year career through Yahoo!, Dentsu, and Stripe, retold as a path of expanding scope and responsibility for the decisions carried. A career traced through changing judgments, not titles.
- [Evidence](https://www.yamazaki-office.com/en/evidence.html): Six anonymized records of decisions made behind the wall of confidentiality. Four themes: AI governance, DX organizations, data and management decisions, and transformation across functions. Not what was done, but what was decided.
- [Case 01 | Retail and Distribution — A phased approval-gate design | Evidence](https://www.yamazaki-office.com/en/evidence-case-01.html): In a large integration project whose requirements advanced without approval criteria, the scope was narrowed to the smallest point that had to be decided now and a three-layer approval framework was drafted. Designing also means deciding first what will not be decided.
- [Case 02 | Telecommunications Carrier — Independent audit of an AI-model evaluation | Evidence](https://www.yamazaki-office.com/en/evidence-case-02.html): In evaluating a churn-prediction AI model, a favorable result was questioned first. Population bias was identified and a premature declaration of success was put on hold. AI governance in practice begins with skepticism toward convenient results.
- [Case 03 | DX Development Division — Rejecting a misdiagnosis of “busyness” | Evidence](https://www.yamazaki-office.com/en/evidence-case-03.html): The decline in a DX organization’s productivity was caused not by understaffing but by its decision-making structure. Rather than prescribe additional hiring, triage of work was prioritized. Add people under a misdiagnosis and the organization only grows heavier.
- [Case 04 | Retail and E-commerce — Designing the approval hierarchy for KPI changes | Evidence](https://www.yamazaki-office.com/en/evidence-case-04.html): An apparently minor KPI change carried risk directly tied to revenue. Rather than leave a revenue-affecting decision to the field alone, a line was drawn in the approval hierarchy. Governance decides how far field discretion extends.
- [Case 05 | Data Analytics Specialist Organization — Redefining recruitment strategy | Evidence](https://www.yamazaki-office.com/en/evidence-case-05.html): Recruitment in response to turnover was separated into two questions: short-term staffing and longer-term organizational defense. The hiring target was redefined. Adding headcount does not create more ways to win. Decide first what hiring is for.
- [Case 06 | Media Company — Designing governance gates for AI adoption | Evidence](https://www.yamazaki-office.com/en/evidence-case-06.html): A generative-AI tool was becoming a fait accompli through expense claims. Governance was designed to keep it from touching live data until it passed three gates: data scope, security, and contract. An expense claim is not consent.
- [Thinking](https://www.yamazaki-office.com/en/thinking.html): It is easier to keep things hidden. Still, we publish. A place to share as a practitioner in the still-unnamed territory of AI governance and organizational authority design: not conclusions, but the work that led to them.
- [Contact](https://www.yamazaki-office.com/en/contact.html): A point of contact for management challenges. Your request need not be neatly organized. Designing decisions begins while a challenge is still not fully articulated. A company name, title, and outline are enough; Shigeki Yamazaki reads every message.
- [Privacy Policy](https://www.yamazaki-office.com/en/privacy.html): How YAMAZAKI OFFICE INC. handles information, email inquiries, cookies, and access analytics on this website.
- [忘れない、をつくる。｜ObsidianとAI](https://www.yamazaki-office.com/thinking-obsidian-ai.html): 記録は手元に。判断はあなたに。経営と暮らしの記録を、次の判断につなげるObsidianとAIの入門特集。今日の3行、AIへの共有範囲、バックアップを学べます。日本語28ページのPDFも配布。
- [Make room to remember. | Obsidian and AI](https://www.yamazaki-office.com/en/thinking-obsidian-ai.html): Keep your records. Own your judgment. A practical introduction to Obsidian and AI: start with three lines, choose what to share, and protect your notes. Includes a 28-page English PDF.