For those who struggle with WBS creation: Tips for organizing by “deliverables” using AI

The reason WBS creation is difficult is not because of the workload of listing tasks, but because you need to take information scattered across plans, meeting minutes, scopes, and quality requirements, and break it down into Process → Task → Deliverable → Acceptance Criteria to align everyone's understanding.

AI is not useful for “guessing the correct WBS.”
Create a draft quickly, visualize omissions and unconfirmed items, and bring confirmation and agreement forward.
Using it this way can reliably reduce rework in implementation projects.

Decide first: Have AI create the “structure,” not the “schedule”

The first thing that falls apart in a WBS is not the dates or man-hours, but the “structure.” In this article, we will narrow down what we have the AI create from the start.

• Format: Process > Task > Deliverable > Acceptance Criteria (Draft)

• Prohibited: Do not write dates, names, utilization rates, or amounts

• Addition: Finally, generate 20 unconfirmed points/confirmation questions

Fixing these three points makes the WBS a “draft that can be used in meetings.”

Two ways to create: 1. Conversational type, 2. Document reference type

Even with the same goal (WBS draft), there are two entry points.

• 1. Conversational type: Summarize the key points yourself and provide them (safest for actual projects)

• 2. Document reference type: Have it reference “materials on hand” such as a project plan PDF (strong when materials are well-organized)

In this article, we will use a sample plan (PDF attached) to create a format that can be reproduced with either method.

1. Conversational type: Provide key points to create a WBS draft (safest for actual projects)

Instead of pasting the original plan, provide only the key points to the AI in a bulleted list.
This is the operation least likely to cause accidents in the field of implementation projects, as it is easy to omit proper nouns, contracts, amounts, and names.

Here, we will organize and provide the key points of the sample plan briefly (the main text contains excerpts; for details, refer to the sample project plan.pdf above).

Copy-paste box: Conversational prompt

あなたはシステム導入プロジェクトのPM支援者です。
以下は機密を除いた“要点のみ”です（固有名詞・金額・契約条件・人名・稼働率・内部URLは含めません）。

【目的（Why）】
- {例：発注業務の電子化、リードタイム短縮、法対応、KPI など}

【スコープ（What）】
- 対象業務：{例：注文、納期回答、受領、検収、支払通知}
- システム範囲：{例：ERP連携IF、WEBフロント}
- 対象外：{例：海外仕入先は次フェーズ など}

【進め方（How）】
- 方式：{例：ウォーターフォール＋ゲート}
- 品質条件：{例：結合テスト重大バグゼロで次工程}
- 変更管理：{例：変更要求→影響評価→承認→反映}

【出力要件（必須）】
- 形式：工程 > タスク > 成果物 > 受入条件（Draft）
- 依存関係（前提タスク）があれば括弧で付記
- 日付・工数・人名・稼働率・金額は書かない
- 最後に「未確認点（Unknown）」と「確認質問」を20個出す
- “それっぽい一般論”に逃げず、上の要点に基づいて作る

まず成果物ベースでWBS叩き台を作ってください。

The WBS generated by this prompt is a “draft,” not a finished product.
Its value lies in bringing omissions in deliverables and unconfirmed points (= points that will explode later) to the forefront.

[Example: WBS output from Gemini using the above prompt]

2. Document-reference type: Creating it by referencing a project plan PDF (NotebookLM, etc.)

If your project plan is well-organized in a PDF, the method of “having the AI reference the document” to generate a WBS is also powerful.
RAG: Retrieval-Augmented Generation
technology can be used to suppress the AI's tendency to provide plausible-sounding but generic answers, allowing you to get responses tailored to your company and specific project rather than generalities.

My point here is not to use difficult terminology, but that what you are doing is simple.

• Conversational type: Depends on your key points (relies on your ability to organize points)

• Document-reference type: Have it extract the basis from the document (reduces misunderstandings of premises)

Google's NotebookLM is a tool for importing sources like PDFs and proceeding with summarization, organization, and question generation based on those materials.
It is well-suited for the purpose of “generating a WBS after reading the project plan.”

Copy-paste box: Request text for NotebookLM

このノートブックに取り込んだ「プロジェクト計画書PDF」に基づいて、WBSの叩き台を作ってください。
（資料に書かれていないことは推測で断定せず、未確認として質問化してください）

【出力要件（必須）】
- 形式：工程 > タスク > 成果物 > 受入条件（Draft）
- 日付・工数・人名・稼働率・金額は書かない
- “計画書のどの記述に基づくか”が分かるように、各工程の末尾に根拠箇所（章番号や該当セクション名）を短く添える
- 最後に「未確認点（Unknown）」と「確認質問」を20個出す
- 可能なら、質問ごとに「誰に聞けば解決しやすいか（業務/IT/仕入先窓口など）」も付ける

まずは成果物ベースでWBS叩き台を提示してください。

【Attachment: WBS output using NotebookLM】

The advantage of the document-reference type is that the WBS is less likely to be “general theory” and more likely to take shape in line with the premises of that project plan.
In review meetings, being able to see “which description is the basis” makes reaching an agreement faster.

Common: Humans do the finishing touches (this is the core of “reducing rework”)

Whether it is the conversational type or the document-reference type, the WBS produced by AI is just a “placeholder.”
What reduces rework is this finishing step of turning the output into an “agreement.”

1) Inspect deliverables and acceptance criteria

• Where are the deliverables thin? (Requirements/Migration/Operations/Deployment are areas to watch)

• Do the acceptance criteria end with just “review complete”?
  → Rephrase to “what must be satisfied to proceed to the next step”

2) Shape unconfirmed items into “meeting questions”

• Are these questions structured so that you know who to ask to resolve them?

• Are they in a format that can be answered with Yes/No or multiple-choice options?
  Once unconfirmed items are organized as questions, the WBS becomes a script for building consensus rather than just a “management table.”

3) Add tasks that are often overlooked first

Projects like WEB-EDI implementation usually have the same blind spots.
Supplier onboarding (breaking down 90% participation into 'tasks'), legal compliance (Electronic Book Preservation Act, Invoice System), operational support desks, and establishing change management. Strengthening these areas first reduces the probability of project failure later on.

Summary: AI is not a 'WBS creator,' but a generator of drafts and questions

To proceed while organizing, create a draft quickly, bring unconfirmed items to the forefront, and secure agreements early.
Start safely with a conversational approach, and if necessary, gather evidence using document-referencing tools (like NotebookLM). Finally, have a human finalize it to make it an operational WBS.

Next time, we will connect the WBS created here directly to meeting minutes, issue management, and risk management to turn it into a template that works in the field.

※This article provides general information. For actual projects, please prioritize your organization's regulations, contract terms, and security policies.

New Series: AI Patterns You Can Use at Work

Read next:

For first-time readers