Skip to main content

Capstone: Building Your Finance Digital FTE

You've built the pieces. Now it's time to assemble the whole.

Throughout this chapter, you've created finance-specific Skills that encode your expertise. You've connected to Google Sheets through MCP. You've designed Subagents for specialized financial roles. You've established governance checkpoints that keep humans in control of critical decisions.

Each component is valuable on its own. Combined into a coherent system, they become something more: a Finance Digital FTE—an AI-powered system that handles end-to-end finance workflows with the consistency of automation and the judgment of expertise.

This isn't just a productivity tool. It's a sellable asset that packages your finance knowledge into a product other finance teams would pay for.

The Digital FTE Vision Recap​

Remember the Agent Factory thesis from Chapter 1: every domain expert can build AI systems that work 24/7, delivering consistent results at a fraction of human cost. The Digital FTE is the culmination—not a replacement for finance professionals, but an extension of their capabilities.

What You BuiltRole in Digital FTE
Skills (Lesson 3)Reusable intelligence: variance-analyzer, transaction-classifier
MCP Connections (Lessons 4, 7)External systems: Google Sheets, accounting platforms
Subagents (Lesson 8)Specialized roles: Modeler, Analyst, Assessor, Validator
Governance (Lesson 9)Safety rails: human checkpoints, audit logging, access controls

A Finance Digital FTE combines all four into a system that can handle recurring workflows—monthly close, quarterly reporting, budget variance analysis—with you directing strategy and approving outputs.

Finance Digital FTE Architecture​

Here's how all the pieces fit together:

Finance Digital FTE
├── Skills (Reusable Intelligence)
│   ├── variance-analyzer (from Lesson 3)
│   │   └── Identifies significant variances, traces root causes
│   ├── transaction-classifier (from Lesson 3)
│   │   └── Categorizes transactions with confidence levels
│   ├── reconciliation-assistant
│   │   └── Matches bank transactions to GL entries
│   └── narrative-generator
│       └── Converts analysis into stakeholder prose
│
├── MCP Connections (External Systems)
│   ├── Google Sheets MCP (from Lesson 4)
│   │   └── Read/write financial data in spreadsheets
│   └── Accounting Platform MCP (from Lesson 7)
│       └── Fetch transactions, trial balance, journal entries
│
├── Subagents (Specialized Roles)
│   ├── Financial Modeler (from Lesson 8)
│   │   └── Builds forecasts, updates projections
│   ├── Scenario Analyst (from Lesson 8)
│   │   └── Creates what-if comparisons
│   ├── Risk Assessor (from Lesson 8)
│   │   └── Flags exceptions, runs stress tests
│   └── Validator (from Lesson 8)
│       └── Checks data integrity before finalizing
│
└── Governance (Safety Rails)
    ├── Human-in-the-loop checkpoints (from Lesson 9)
    │   └── Approval required before posting/sending
    ├── Audit logging (from Lesson 9)
    │   └── Every AI action documented with reasoning
    └── Access controls (from Lesson 9)
        └── Role-based permissions for sensitive data

Each component contributes essential capability. Remove any layer and the system loses either intelligence, connectivity, specialization, or safety.

Capstone Project Options​

Choose one of these four workflows to build as your capstone. Each represents a real finance use case that combines all chapter components.

Option 1: Monthly Financial Health Checkup​

What it does: Automated monthly analysis that reviews key metrics, identifies variances, assesses risks, and generates an executive summary.

Workflow sequence:

  1. Validator verifies all transactions posted for the month
  2. Variance-analyzer compares actuals to budget/forecast
  3. Risk Assessor flags unusual items or concerning trends
  4. Narrative-generator creates CFO summary
  5. Human reviews and approves distribution

Key metrics tracked:

  • Revenue vs plan ($ and %)
  • Gross margin trend
  • Operating expense run rate
  • Cash position and runway
  • AR aging and collections

Output: One-page executive summary with variance table and risk flags.

Option 2: Accounts Receivable Aging Workflow​

What it does: Analyzes customer payment patterns, prioritizes collection efforts, and assesses credit risk.

Workflow sequence:

  1. MCP pulls AR aging report from accounting system
  2. Transaction-classifier categorizes customers by risk tier
  3. Risk Assessor identifies highest-risk accounts
  4. Scenario Analyst models collection scenarios (best/base/worst)
  5. Narrative-generator creates collection priority list with talking points
  6. Human reviews and assigns to collectors

Key metrics tracked:

  • Days Sales Outstanding (DSO)
  • Aging buckets (current, 30, 60, 90+ days)
  • Customer concentration risk
  • Collection probability by tier

Output: Prioritized collection list with customer-specific talking points.

Option 3: Budget vs Actual Reporting​

What it does: Comprehensive variance analysis with automated explanation generation and action recommendations.

Workflow sequence:

  1. MCP pulls budget and actual data from Google Sheets
  2. Validator checks data completeness and accuracy
  3. Variance-analyzer identifies and categorizes all variances
  4. Financial Modeler updates forecast based on trends
  5. Narrative-generator creates variance commentary for each department
  6. Human reviews and distributes to department heads

Key deliverables:

  • Summary variance table (total company)
  • Department-level detail with explanations
  • Updated full-year forecast
  • Action items for material variances

Output: Complete budget package ready for management review.

Option 4: Cash Flow Forecasting​

What it does: Rolling 13-week cash forecast with scenario analysis and liquidity alerts.

Workflow sequence:

  1. MCP pulls AP aging, AR aging, recurring expenses from systems
  2. Financial Modeler builds 13-week cash projection
  3. Scenario Analyst creates conservative/base/aggressive scenarios
  4. Risk Assessor identifies potential liquidity stress points
  5. Validator checks formula integrity and assumption consistency
  6. Narrative-generator creates treasury summary
  7. Human reviews and makes funding decisions

Key metrics tracked:

  • Weekly cash position
  • Minimum cash threshold breaches
  • Payment timing optimization opportunities
  • Scenario-based runway calculations

Output: 13-week cash forecast with scenario comparison and alerts.

Capstone Deliverables Checklist​

Regardless of which project you choose, your Finance Digital FTE should include these deliverables:

□ CLAUDE.md with domain context (from Lesson 2)
  ├── Chart of accounts structure
  ├── Materiality thresholds
  ├── Fiscal year and reporting calendar
  └── Company-specific policies and rules

□ 2-3 Finance Skills with full specifications (from Lesson 3)
  ├── SKILL.md files with persona, logic, context, safety
  ├── Example inputs and outputs
  └── Confidence level definitions

□ MCP integration design document (from Lessons 4, 7)
  ├── Which systems are connected
  ├── Read vs write permissions
  ├── Data refresh frequency
  └── Error handling approach

□ Subagent orchestration plan (from Lesson 8)
  ├── Which subagents are used
  ├── Invocation sequence (sequential/parallel/conditional)
  ├── Context handoff between agents
  └── Output aggregation approach

□ Governance checklist (from Lesson 9)
  ├── Human approval checkpoints
  ├── Audit logging requirements
  ├── Access control matrix
  └── Escalation procedures

□ Complete workflow specification document
  ├── Trigger conditions (when does workflow run?)
  ├── Input requirements (what data is needed?)
  ├── Processing steps (what happens in order?)
  ├── Output formats (what gets produced?)
  └── Success criteria (how do we know it worked?)

□ Sample outputs for stakeholder review
  ├── Example of final deliverable
  ├── Intermediate artifacts (for debugging)
  └── Governance documentation (audit trail)

This checklist ensures you've integrated all chapter concepts into a complete, production-ready system.

Workflow Specification Template​

Use this template to document your capstone workflow:

# Finance Digital FTE: [Workflow Name]

## Overview

**Purpose**: [One sentence describing what this workflow accomplishes]

**Trigger**: [When/how often this workflow runs: monthly, on-demand, etc.]

**Owner**: [Who is responsible for reviewing outputs]

## Component Inventory

### Skills Used

| Skill | Purpose in Workflow |
|-------|---------------------|
| [skill-name] | [What it does] |
| [skill-name] | [What it does] |

### MCP Connections

| System | Data Accessed | Read/Write |
|--------|---------------|------------|
| [system] | [data types] | [R/W/Both] |

### Subagents

| Subagent | Role | Triggers |
|----------|------|----------|
| [agent-name] | [specialty] | [when invoked] |

### Governance Checkpoints

| Checkpoint | Criteria | Approver |
|------------|----------|----------|
| [checkpoint-name] | [what triggers review] | [role] |

## Workflow Sequence

### Step 1: [Name]
**Agent/Skill**: [which component]
**Input**: [what it receives]
**Processing**: [what it does]
**Output**: [what it produces]
**Governance**: [any checkpoints]

### Step 2: [Name]
[Continue for each step...]

## Output Specification

### Primary Deliverable
[Describe the main output: format, content, audience]

### Supporting Artifacts
[List any intermediate outputs, logs, or documentation]

## Success Criteria

- [ ] [Measurable criterion 1]
- [ ] [Measurable criterion 2]
- [ ] [Measurable criterion 3]

## Error Handling

| Error Type | Detection | Response |
|------------|-----------|----------|
| [error category] | [how detected] | [what happens] |

## Maintenance Notes

**Skill Updates**: [When/how skills need revision]
**System Changes**: [What to update if connected systems change]
**Governance Review**: [Periodic review schedule]

From Digital FTE to Sellable Asset​

Your Finance Digital FTE isn't just a personal productivity tool. It's intellectual property with market value.

What Makes It Valuable​

  1. Encoded Expertise: Your variance analysis approach, classification rules, and governance frameworks represent years of professional experience

  2. Consistent Quality: Unlike manual work that varies with attention and energy, skills produce consistent outputs every time

  3. Scalability: The same Digital FTE that handles your company's monthly close can handle ten companies' closes with minor customization

  4. Time Compression: Workflows that take human teams hours compress to minutes of orchestrated AI execution plus focused human review

Customization Points​

When packaging your Digital FTE for other finance teams, they'll need to customize:

ComponentCustomization Required
CLAUDE.mdChart of accounts, thresholds, policies
SkillsConfidence levels, output formats, escalation triggers
MCPConnection credentials, system-specific mappings
SubagentsPersona adjustments, workflow sequencing
GovernanceApproval hierarchies, audit requirements

The core logic—your expertise in variance analysis, classification, risk assessment—remains constant. The configuration adapts to each client's context.

Packaging Options​

PackageWhat's IncludedTarget Buyer
Skill LicenseIndividual skills (variance-analyzer, classifier)Finance teams building their own systems
Workflow TemplateComplete workflow spec + skillsTeams wanting ready-to-configure solutions
Managed ServiceFull Digital FTE + configuration + supportTeams wanting hands-off automation
Custom BuildBespoke Digital FTE for specific requirementsEnterprise clients with unique needs

Your capstone project is the prototype. With refinement and documentation, it becomes the product.

Production Deployment Considerations​

Before putting your Finance Digital FTE into production, validate these areas:

Data Security​

  • Sensitive financial data stays within approved systems
  • MCP connections use secure authentication
  • Audit logs capture data access
  • Skills don't leak confidential information in reasoning chains

Reliability​

  • Workflow handles missing or incomplete data gracefully
  • Error states trigger appropriate escalation
  • Retry logic exists for transient failures
  • Human override is always available

Compliance​

  • Governance checkpoints align with company policy
  • Audit trail satisfies regulatory requirements
  • Access controls enforce separation of duties
  • Documentation supports external audit requests

Maintenance​

  • Skills include version numbers for tracking changes
  • CLAUDE.md updates have a clear process
  • MCP connection changes are tested before deployment
  • Regular review schedule exists for governance rules

Try With AI​

Complete your capstone by working through these prompts with Claude Code.

Prompt 1: Design Your Complete Finance Digital FTE​

Help me design a Finance Digital FTE for [monthly close / accounts receivable /
budget variance / cash forecasting - pick one].

I need:
1. Complete component list: What Skills do I need? What MCP connections?
   What Subagents? Show me how they connect.
2. Workflow sequence: What runs in what order? What's sequential vs parallel?
3. Governance checkpoints: Where does a human need to review?
4. Output specifications: What does the final deliverable look like?

Give me a complete architecture document I can use to implement this.

What you're learning: This prompt forces you to think through the entire system architecture. You're not just listing components—you're designing how they interact. The architecture document becomes your implementation roadmap. Pay attention to the data flows between components; that's where integration complexity lives.

Prompt 2: Validate Your Specification​

Review my capstone specification: [paste your workflow specification]

Evaluate against these criteria:
1. Are all four component types present? (Skills, MCP, Subagents, Governance)
2. Is the workflow sequence clear and complete?
3. Are governance checkpoints sufficient for finance compliance?
4. What's missing for production deployment?

Score each criterion and tell me what to fix.

What you're learning: Self-review is essential for production systems. This prompt trains you to evaluate your own work against explicit criteria. The scoring forces specificity—you can't hide behind "looks good." When Claude identifies gaps, pay attention to whether they're missing components or unclear specifications. Both matter, but they require different fixes.

Prompt 3: Package for Sale​

I've built a Finance Digital FTE for [your workflow choice]. I want to package
it to sell to other finance teams.

Help me think through:
1. What makes this valuable to a buyer? What problem does it solve?
2. What would a buyer need to customize for their company?
3. How would I document it for someone who didn't build it?
4. What pricing model makes sense: license, subscription, or service?

Give me a product positioning statement and customization guide.

What you're learning: Building is one skill; productizing is another. This prompt shifts your perspective from creator to seller. You're forced to articulate value (not just features), identify customization points (where does your logic end and their context begin?), and think about documentation for users who lack your context. The pricing model question makes the business real—what's the recurring value you're delivering?

Safety Note: Your Finance Digital FTE handles sensitive financial data and produces outputs that influence business decisions. Before deploying to production: test with historical data where you know the correct answers, validate all governance checkpoints actually trigger, ensure audit logs capture sufficient detail for compliance review, and maintain human oversight on all material items. The Digital FTE augments your judgment; it never replaces your responsibility for financial accuracy.