Getting Started

Step 1: Install the Plugin

Open any Claude Code session and run:

claude plugin install ai-cdo-office

Verify the installation:

claude plugin list    # Should show ai-cdo-office

Type /cdaio: to see all available workflows in the autocomplete.

Step 2: Initialize Your Project

Navigate to your project folder (or create a new one) and run:

/cdaio:init

This creates the working structure:

your-project/
├── context/
│   ├── client-context.md       ← your organization profile
│   └── (supporting docs)       ← PPTX, PDF, DOCX, XLSX
├── deliverables/
│   └── (auto-created per run)  ← strategy-2026-03-09/, board-prep-2026-03-10/, ...
└── .gitignore

What each folder does:

Step 3: Fill In Your Context

Edit context/client-context.md with your organization details. Here’s what to include:

# Client Context

## Organization
- **Company name:** Acme Financial Services
- **Industry:** Financial Services
- **Revenue / size:** $4.2B revenue, 12,000 employees
- **Geographic scope:** North America + EMEA

## Data & AI Maturity
- **Current maturity level:** 2-Emerging
- **Data team size:** 45 people
- **Key platforms:** Snowflake, Databricks, Azure
- **AI models in production:** 3 (fraud detection, credit scoring, churn)

## Current Priorities
- **Top 3 initiatives:**
  1. Enterprise AI strategy for the board
  2. Data governance framework (regulatory pressure)
  3. GenAI use case evaluation
- **Key challenges:** Siloed data, no governance council, shadow AI
- **Upcoming deadlines:** Board meeting in 3 weeks, regulatory audit in Q2

## Stakeholders
- **CEO/Board expectations:** AI ROI proof, risk mitigation plan
- **Key executive sponsors:** CFO (data monetization), COO (operational AI)
- **Business units engaged:** Retail banking, risk management

## Regulatory Environment
- **Applicable regulations:** GDPR, DORA, EU AI Act
- **Compliance status:** GDPR compliant, DORA gap assessment in progress

Tip: The more context you provide, the more tailored every deliverable becomes. Agents will ask for missing information if needed.

Supporting Documents

Place any existing materials alongside your context file:

context/
├── client-context.md
├── current-board-deck.pptx      ← agents will reference your existing deck style
├── last-audit-report.pdf        ← compliance officer uses this for gap analysis
├── data-governance-policy.docx  ← governance team builds on existing policies
└── ai-budget-2025.xlsx          ← analyst uses this for investment case

Agents extract relevant information from these documents when building your deliverables.

Step 4: Run Your First Command

Try a maturity assessment to see the team in action:

/cdaio:assess

The Head of AI & Analytics will:

1. Read your context/client-context.md
2. Ask clarifying questions if any details are missing
3. Score your organization across 6 dimensions and 24 subdimensions
4. Produce a detailed report with gap analysis and recommendations
5. Save everything to deliverables/assess-2026-03-09/

All 20 Commands

All commands accept inline context to skip the Q&A phase:

/cdaio:strategy Allianz SE, €150B insurer, 150k employees, GDPR + EU AI Act scope

Agents also read everything in your context/ folder automatically — existing board decks, audit reports, governance policies, org charts, and budget spreadsheets.

How to Talk to the Agents

The agents are conversational. Point them at real companies, real documents, and real deadlines:

Reference Your Company and Documents

Agents read your context/ folder and use it as the foundation for every deliverable:

"I'm the new CDAIO at Société Générale. I've put our current data strategy
deck in context/sg-data-strategy-2025.pptx and last quarter's board minutes
in context/board-minutes-q4.pdf. Run the maturity assessment using these
as your baseline."

The Head of AI & Analytics will parse your existing documents, extract the current state, and score your maturity against what’s in them — not generic assumptions.

Use Real Company Context

Give the agents a real company and they’ll research public information to ground their work:

"Build an AI strategy for Roche. They're a $65B pharma company with
strong R&D data capabilities but fragmented commercial analytics.
Look up their latest annual report for data investment numbers.
They need to present to the board in 4 weeks."

"Run a vendor evaluation for BBVA. They're comparing Databricks vs
Snowflake for their new data lakehouse. They currently run on-prem
Teradata with 500TB. Budget is €8M over 3 years."

Ask Follow-Up Questions

After any deliverable, iterate with specifics:

"The CFO at Unilever will push back on the €12M ask — prepare a phased
investment option starting at €3M with clear stage gates."
"Add a slide benchmarking our AI maturity against JPMorgan and Goldman Sachs."
"Rewrite the executive summary for a non-technical board — our chair
is a former CFO who thinks in ROI and payback periods."

Activate Specific Team Members

Talk directly to any of the 17 agents with real-world tasks:

"Talk to the Compliance Officer — we're a German bank subject to
BaFin, DORA, GDPR, and EU AI Act. Assess our gaps using the policy
documents I've put in context/compliance/."

"Have the Data Architect design a target-state architecture for
L'Oréal's consumer data platform. They have 35 brands, 3 CDPs,
and want to consolidate onto one cloud-native stack."

"Get the Use Case Lead to score 20 GenAI use cases for a $4B
hospital system. Read context/ai-use-case-longlist.xlsx for the
current list. Prioritize by clinical impact and regulatory risk."

Provide Additional Context Mid-Conversation

Drop new information anytime — documents, links, or facts:

"We just acquired a fintech with 80 data engineers and a Databricks
stack. I've added their architecture doc to context/acquisition-arch.pdf.
Factor this into the target-state architecture."

"Look up the latest Gartner Magic Quadrant for Data Integration Tools.
Our vendor evaluation needs to reflect the 2025 positioning."

The agents will incorporate new context into their ongoing work — whether it’s from files you drop into context/, facts you share in conversation, or public information they research.

The 17 Agents

Leadership

Governance

AI & Analytics

Operations

Quality & Oversight

Playbook Walkthroughs

Playbooks are multi-agent workflows where the CDO coordinates handoffs between team members. Each produces a coordinated set of deliverables.

AI Strategy Sprint

Trigger: /cdaio:strategy or describe the need in natural language

/cdaio:strategy Nestlé, CHF 94B food & beverage, 270k employees, need AI strategy for supply chain and consumer analytics. Read context/ for existing data landscape docs.

What happens:

1. Head of AI & Analytics runs a maturity assessment → XLSX
2. Use Case Lead identifies and scores AI opportunities
3. Data & AI Architect designs the target-state platform → PPTX (13 slides)
4. AI/ML Lead evaluates build-vs-buy for model capabilities
5. Data Analyst builds the investment case → XLSX (6 tabs: NPV, IRR, payback)
6. CDO synthesizes everything into a strategy deck → PPTX (17 slides)
7. Quality Reviewer validates MBB standards compliance

Output: deliverables/strategy-<date>/ containing strategy PPTX, strategy DOCX, maturity XLSX, investment case XLSX, architecture PPTX

Board Meeting Prep

Trigger: /cdaio:board-prep or describe the need in natural language

/cdaio:board-prep AXA Group, €100B insurer, board in 3 weeks. CEO wants ROI proof on our data mesh investment. I've put last quarter's KPI dashboard in context/kpi-q4.xlsx and the previous board deck in context/board-deck-q3.pptx.

What happens:

1. Data Analyst compiles KPI scorecard with RAG status → XLSX
2. CDO sets the narrative arc (Situation-Complication-Resolution)
3. Exec Comms Lead builds the board deck → PPTX (17 slides with action titles)
4. Chief of Staff prepares talking points and anticipated questions → DOCX
5. Quality Reviewer runs the MBB standards diagnostic
6. Board of Directors performs the final strategic gate review

Output: deliverables/board-prep-<date>/ containing board deck PPTX, scorecard XLSX, talking points DOCX, meeting prep DOCX

Governance Framework

Trigger: /cdaio:governance or describe the need in natural language

/cdaio:governance Deutsche Bank, BaFin + ECB regulated, GDPR + DORA + EU AI Act. Current policies in context/data-governance-v2.docx. Need full framework before regulatory audit in Q2.

What happens:

1. Head of Data Governance defines the governance charter
2. Data Steward builds the data dictionary → XLSX (4 tabs)
3. Compliance Officer maps regulatory requirements → DOCX
4. Program Manager creates the RACI → XLSX (3 tabs)
5. Data Custodian defines technical controls and access policies

Output: deliverables/governance-<date>/ containing policy DOCX, data dictionary XLSX, RACI XLSX, compliance checklist DOCX

First 90 Days

Trigger: /cdaio:first-90-days or describe the need in natural language

/cdaio:first-90-days I'm starting as CDAIO at Siemens Energy in 3 weeks. €30B revenue, 92k employees, industrial IoT + energy trading data. Previous CDO left after 18 months. Read context/ for the org chart and existing strategy docs.

What happens:

1. CDO builds the 90-day phased plan (Listen → Quick Wins → Strategic Foundation)
2. Chief of Staff maps all stakeholders and designs the operating rhythm
3. Head of AI & Analytics runs an initial maturity assessment → XLSX
4. Head of Data Governance audits current governance state
5. Program Manager designs the target operating model and RACI → XLSX
6. Data Analyst sets KPI baselines and benchmarks → XLSX
7. All 17 agents contribute their domain-specific priorities and quick wins

Output: deliverables/first-90-days-<date>/ containing phased plan, stakeholder map, maturity baseline XLSX, operating model, KPI scorecard XLSX

Quarterly Operating Rhythm

Trigger: /cdaio:quarterly or describe the need in natural language

/cdaio:quarterly Philips, Q1 review for the data office. Read context/kpi-tracker.xlsx for current metrics. We launched 3 new ML models this quarter and need to report to the ExCo.

What happens:

1. Chief of Staff compiles the status report → PPTX
2. Data Analyst updates the KPI dashboard → XLSX
3. CDO sets priorities for the next quarter
4. Stakeholder Relations drafts communications for each audience

Output: deliverables/quarterly-<date>/ containing status PPTX, scorecard XLSX, meeting agendas DOCX, stakeholder comms DOCX

AI Governance Program

Trigger: /cdaio:ai-governance or describe the need in natural language

/cdaio:ai-governance BNP Paribas, 48 ML models in production (credit scoring, fraud, AML, pricing). EU AI Act compliance deadline approaching. Model inventory in context/model-registry.xlsx.

What happens:

1. AI/ML Lead builds the model inventory and risk classification framework
2. Compliance Officer maps AI regulatory requirements (EU AI Act, NIST AI RMF)
3. Head of Data Governance defines AI-specific data policies
4. Program Manager creates the AI governance RACI → XLSX

Output: deliverables/ai-governance-<date>/ containing AI governance framework, model inventory, risk classification matrix, AI policy DOCX

Data Organization Redesign

Trigger: /cdaio:org-design or describe the need in natural language

/cdaio:org-design Vodafone, 100k employees, 15 OpCos across Europe. Currently fully decentralized — each OpCo has its own data team. CEO wants a federated model. Current org chart in context/data-org-current.pdf.

What happens:

1. Program Manager assesses current-state org and identifies gaps
2. CDO defines the target operating model (centralized, federated, or hybrid)
3. Head of AI & Analytics maps required capabilities to roles
4. Stakeholder Relations plans the change communication strategy

Output: deliverables/org-design-<date>/ containing target org chart, role descriptions, transition plan, RACI XLSX, change comms DOCX

Data Quality Program

Trigger: /cdaio:data-quality or describe the need in natural language

/cdaio:data-quality Novartis, pharma, clinical trial data and commercial analytics have 23% duplicate rate. Data quality audit results in context/dq-audit-2025.xlsx. Need remediation plan before FDA inspection in Q3.

What happens:

1. Data Steward runs a data quality assessment across key domains
2. Data Engineer identifies root causes in pipelines and infrastructure
3. Data Custodian audits access controls and data lineage
4. Program Manager builds the remediation roadmap and RACI → XLSX

Output: deliverables/data-quality-<date>/ containing quality assessment, root cause analysis, remediation roadmap, data dictionary XLSX

Cost Optimization Sprint

Trigger: /cdaio:cost-optimization or describe the need in natural language

/cdaio:cost-optimization Shell, $18M annual Snowflake + Databricks spend. Cloud bills in context/cloud-costs-2025.xlsx. CFO wants 30% reduction without impacting production ML models. Look up their latest earnings call for context on cost pressure.

What happens:

1. Data Analyst audits current spend and benchmarks against industry → XLSX
2. Data Engineer identifies infrastructure waste and rightsizing opportunities
3. Data & AI Architect recommends architecture consolidation
4. CDO builds the business case with phased savings → XLSX

Output: deliverables/cost-optimization-<date>/ containing spend audit XLSX, waste analysis, savings roadmap, investment case XLSX

Vendor & Technology Evaluation

Trigger: /cdaio:vendor-eval or describe the need in natural language

/cdaio:vendor-eval BBVA, comparing Databricks vs Snowflake vs Google BigQuery for new lakehouse. Currently on-prem Teradata, 500TB. Requirements in context/platform-requirements.docx. Budget €8M over 3 years. Look up latest Gartner and Forrester positioning for each vendor.

What happens:

1. Data & AI Architect defines requirements and builds the scoring framework
2. Data Engineer evaluates technical capabilities and integration complexity
3. Data Analyst builds the TCO model → XLSX
4. Compliance Officer checks vendor compliance and data residency

Output: deliverables/vendor-eval-<date>/ containing requirements matrix, vendor scorecard, TCO analysis XLSX, recommendation DOCX

Generating PPTX Presentations

The AI CDAIO Office can produce boardroom-ready PowerPoint files. Every presentation follows MBB (McKinsey, BCG, Bain) standards built into the agents.

Setup

Install the generator dependencies:

# PPTX + DOCX generators (Node.js)
npm install pptxgenjs docx js-yaml

# XLSX generators (Python)
pip install openpyxl pyyaml

Agents detect installed generators automatically. Without them, you still get the same strategic output — just in markdown instead of formatted files.

PPTX Generators

Four generators produce PowerPoint presentations:

Example: Generate a Board Deck PPTX

/cdaio:board-prep Zurich Insurance, CHF 72B, board in 2 weeks. Previous deck in context/board-q3.pptx, KPIs in context/kpi-dashboard.xlsx. Chair wants to see AI ROI and DORA compliance progress.

The Exec Comms Lead generates a 17-slide PPTX following this structure:

1. Title slide — with your organization name and date
2. Executive Summary — Situation-Complication-Resolution format
3. Agenda — 3-5 sections max
4. KPI Dashboard — 5-7 metrics with RAG status (green/amber/red)
5. Initiative Portfolio — status tracker with Harvey Ball progress indicators
6. Risk Matrix — probability vs. impact
7. Budget Summary — waterfall chart (spent, committed, remaining)
8. Recommendations — 3 concrete actions
9. Next Steps — with owners, dates, and success criteria 10-17. Supporting slides — deep dives on each initiative, appendix data

Output saved to deliverables/board-prep-<date>/board-deck.pptx

Example: Generate an Architecture PPTX

/cdaio:architecture Maersk, migrating from on-prem Oracle + Teradata to Azure cloud-native. Current architecture diagram in context/arch-current.pdf. $12M budget over 2 years. Must support real-time container tracking and predictive logistics.

The Data & AI Architect produces a 13-slide PPTX covering all 8 architecture layers:

1. Current-state architecture assessment
2. Target-state architecture blueprint
3. Technology stack comparison matrix
4. Data ingestion and pipeline patterns
5. Storage and compute layer design
6. ML/AI platform architecture
7. Governance and security controls
8. Migration roadmap with phased approach 9-13. Cost estimates, vendor comparison, risk assessment

Output saved to deliverables/architecture-<date>/architecture-blueprint.pptx

MBB Presentation Standards

Every generated PPTX follows consulting-grade standards from the built-in MBB style guide:

Slide anatomy:

• Action titles — every slide title is a complete sentence stating the key takeaway (not topic labels). Passes the “Zelazny Test”: an unfamiliar reader understands the point from the title alone.
• One idea per slide — if you need “and” in the title, it’s two slides.
• Three bullet points maximum — never more per slide.
• Quantify everything — no “significant growth”, always “47% CAGR” or “$12M annual savings”.

Narrative structure:

• Pyramid Principle — conclusion first, then supporting arguments, then evidence. Top-down communication.
• SCR Framework — every executive summary uses Situation-Complication-Resolution.
• MECE — all categorizations are Mutually Exclusive, Collectively Exhaustive.
• Ghost Deck process — titles are written first as a standalone storyline, then bodies are built to prove each title.

Visual standards:

• RAG status colors — Green (#2C5F2D, on track), Amber (#D4A843, at risk), Red (#B85042, critical)
• Harvey Balls — for maturity indicators: empty (L0), quarter (L1), half (L2), three-quarter (L3), full (L4)
• Chart selection — comparison → bar, trend → line, proportion → stacked bar, positioning → 2x2 matrix
• Waterfall charts — green bars (positive), red (negative), with connector lines and +/- signs
• Layout rules — consistent positions, max 70% fill, 0.5-inch margins, grid-aligned, max 3 visual elements per slide

Quality gate: The Quality Reviewer (10 years McKinsey Visual Communications, 3,000+ decks reviewed) runs a diagnostic on every PPTX before delivery, checking action titles, data sourcing, visual consistency, and narrative coherence.

Other File Generators

Beyond PPTX, agents produce DOCX and XLSX deliverables:

Why Claude Code (Not LangGraph, CrewAI, etc.)

Most agent frameworks require Python glue code, state machines, and infrastructure. The AI CDAIO Office chose Claude Code because:

• Zero infrastructure — no servers, no Docker, no API keys. Install the plugin, type a command, done.
• Natural language agents — each agent is a markdown file with personality, workflow, and expertise. No Python classes or DAGs.
• Built-in tool use — Claude Code natively reads files, writes deliverables, and runs generators.
• Plugin distribution — users install with one command. No cloning repos or configuring environments.
• Context window as memory — project context lives in files that Claude reads naturally. No vector databases.

The trade-off: you need Claude Code (and an Anthropic API key or Claude subscription). But if you’re already using Claude, this gives you a full CDAIO office with zero additional setup.

Frequently Asked Questions

Do I need to fill in all fields in client-context.md? No. Fill in what you know. Agents will ask for missing information during conversations. But more context upfront means better output with fewer questions.

Can I run commands without /cdaio:init first? Yes. Commands work without the folder structure. But init ensures deliverables are organized and context is loaded automatically.

What if I want to change the output after it’s generated? Just tell the agent. “Make the executive summary shorter” or “Add a section on regulatory risk.” Agents iterate on their work.

How do I get PPTX/DOCX/XLSX output instead of markdown? Install the generator dependencies (npm install pptxgenjs docx js-yaml and pip install openpyxl pyyaml). Agents automatically detect installed generators and produce formatted files.

Can I customize the PPTX style or branding? Yes. The design system is defined in shared/design-system.md. Edit colors, fonts, and layout rules there. Generators read these settings when building files.

What are the MBB standards the agents follow? MBB refers to McKinsey, BCG, and Bain — the top-tier management consulting firms. The built-in style guide (shared/mbb-style-guide.md) enforces Pyramid Principle, action titles, MECE structuring, SCR narrative framework, and consulting-grade chart standards across all deliverables.

Can multiple people use the same project folder? Yes. The context/ folder is shared. Each person’s Claude Code session reads the same context and writes to deliverables/.

How do I customize an agent’s behavior? Each agent is a markdown file in office/. You can edit their personality, workflow, or expertise directly. Changes take effect in your next session.

How does the Board of Directors agent work? The Board is a simulated oversight layer (Finance, Risk, and Strategy committees). The CDO routes board-facing deliverables through it as the final strategic gate — challenging assumptions, stress-testing ROI, and reviewing from a shareholder-value lens.