Claude Code Finance App: How a Fractional CFO Built a Full Three-Way FP&A Tool in 6 Weeks With Zero Coding Experience

A fractional CFO with no coding background used Claude Code to build a fully functional, client-ready FP&A application in six weeks. The app handles three-way financial modelling, multi-entity group consolidation, intercompany reconciliations, prepayments, accruals, and real-time integrations with Xero, QuickBooks, and Sage. Total monthly cost: $160. Annual saving versus previous software: $8,000.

In CFO Connect's third live-build session, Kevin Steele (fractional CFO and founder of Inflective Intelligence) demonstrated the app on screen in front of a live audience, then walked through exactly how he built it. Below is the distilled playbook.

Key Takeaways

• A non-technical finance professional can build a production-ready FP&A application using Claude Code with no prior coding experience

• The step-by-step iteration method (rather than a "grand vision" prompt) is the most effective and token-efficient approach to building with Claude Code

• Screenshots pasted directly into Claude Code are more effective than long descriptive prompts for UI and layout changes

• A backlog file inside Claude Code replaces a product roadmap and keeps development prioritised and continuous

• Deep finance knowledge is non-negotiable; Claude Code builds the mechanics, but human financial expertise is what makes outputs trustworthy for live clients

• Total infrastructure cost for a production-grade finance app runs at approximately $160/month

What is Claude Code, and why are finance professionals building with it?

What is Claude Code?

Claude Code is a development environment within Claude that allows non-technical users to build custom software applications by describing what they need in plain language. For finance teams, it means building internal tools, dashboards, and full applications without requiring a development team or prior coding knowledge.

What is a three-way financial model?

A three-way financial model links the profit and loss account, balance sheet, and cash flow statement so that changes in one automatically flow through to the others. It is the backbone of any credible FP&A process and the accounting foundation Kevin built Inflective Intelligence on before adding any front-end features.

What is Inflective Intelligence?

Inflective Intelligence is an FP&A application built entirely in Claude Code by Kevin Steele, a fractional CFO with no prior coding experience. It is now in live use with paying clients and handles forecasting, consolidation, reporting, and real-time integrations across multiple accounting systems.

Why are finance professionals building their own tools with Claude Code?

Because the mid-market FP&A software category has consistent gaps. As Kevin put it during the session: "I tried a lot of the well-known FP&A and forecasting apps for the sub-ERP market. They just weren't as accurate on cash flows as I'd have liked. There was always something annoying that I couldn't get right. I learned I cannot just have a tool that works the way I want." Claude Code removes the dependency on software vendors to fix what bothers you.

1. What can you actually build with Claude Code as a finance professional with no coding experience?

The answer is: a production-grade, client-ready FP&A application — if you approach it correctly.

Kevin started with a single message to Claude Code five to six weeks before this session. No architectural planning document. No technical briefing. Just: "I'm looking to build an app with a profit and loss account with drivers where I can update to populate. Eventually I want to move on to balance sheet and then cash flow — a full three-way forecast. But I want to take it step by step and start with the P&L."

Claude recommended a tech stack. Kevin did not fully understand what most of it meant. He approved it anyway, answered Claude's clarifying questions (forecast horizon, driver types, actuals vs. forecast, business type), and let Claude run.

The first working P&L with drivers took approximately 10 to 15 minutes.

Six weeks later, Inflective Intelligence includes:

• Three-way linked financial model (P&L, balance sheet, cash flow)

• Driver-based forecasting with custom formula assignment

• Multiple scenarios (baseline, best case, worst case, master budget)

• Goal seek functionality on both historical and forecast data

• Prepayments, accruals, deferred income, and accrued income modules

• Aged debtors and creditors with exact timing profiles rather than averages

• Fixed assets (straight-line and reducing balance depreciation)

• Loans module with full capital/interest or interest-only amortisation schedules

• Group consolidation across multiple entities and currencies

• FX rate auto-population

• Intercompany elimination workflows

• Headcount module with org chart generation

• Valuation module with multiple methodologies and sector multiples

• Reporting suite with AI-written commentary

• Live integrations with Xero, QuickBooks, and Sage

• API connectors for Google Analytics, HubSpot, Asana, and Stripe

• Shared dashboards with crypto-token authentication for client access

• Intelligence module that ingests emails (Outlook) and meeting notes for AI context

Infrastructure: GitHub for code, Supabase for data, Vercel for deployment. All three are SOC 2 compliant. All three carry multi-factor authentication.

What to do this quarter

Before you write a single prompt, identify the one finance problem that costs your team the most hours every month. Write one paragraph describing it: what the input is, what the output should look like, and who needs to review it. That paragraph is your first Claude Code prompt.

2. What is the right method for building a finance app in Claude Code?

Iterative, bite-sized steps beat comprehensive upfront planning every time.

The most common mistake Kevin sees online is the "grand vision" approach: writing a massive design document, feeding it all into Claude at once, and expecting a finished product. This does not work well. It burns through tokens, produces inconsistent outputs, and makes debugging nearly impossible.

Kevin's method:

1. Start with the smallest possible working version. Build the P&L first. Nothing else. Get it right.

2. Act as quality assurance at every step. Before moving to the next feature, verify the current one works as expected using your own finance knowledge.

3. Use screenshots instead of long descriptions. Take a screenshot, circle the problem area with the snipping tool, paste it directly into Claude Code, and describe what you want changed. "It's a lot easier to show Claude an image than to type out a three-paragraph explanation."

4. Maintain a backlog file. Tell Claude to create a backlog file in the project. Every time you have a new idea, say "add this to the backlog." Claude builds up a prioritised development list and can cross-reference it mid-session: "We're working on this — we could actually knock off that backlog item at the same time."

5. Let Claude handle everything mechanical in the background.Your job is to define the outcome and verify the output. Claude's job is to write every line of code.

Kevin went from zero coding knowledge to a production app because he treated every session as a QA review, not a development sprint. His development speed increased week by week because iterating step by step meant he was inadvertently learning how the system worked.

What to do this quarter

Set up Claude Code following the official installation guide (approximately 30 to 45 minutes). Create a new project. Write one prompt describing your first feature. Approve the tech stack Claude recommends. Type "continue" until you have a working version one, however basic. Then open a backlog file and note what you want to add next.

👉 For a broader introduction to Claude, Claude Cowork, and how they layer together for finance teams, read our Claude for Finance Teams playbook.

3. How do you handle financial accuracy and QA when building with Claude Code?

You cannot outsource financial judgment to Claude. The tool builds mechanics; your expertise is what makes the output trustworthy.

Kevin spent seven hours getting the prepayments module right. An expense prepaid over 12 months but with cash hitting only across 10 of those months. It sounds straightforward. The tax posting kept landing in the wrong period. Retained earnings kept drifting. The cash flow kept breaking.

The fix was not a better Claude prompt. The fix was Kevin's own understanding of how a three-way model behaves: "You know the impact on corporation tax. You know the impact on sales tax. You know the impact on retained earnings. You know the impact on cash. So you can just manually verify with the demo company. Here's what it was before. I make the change. Is this what I expect to happen?"

His warning for the audience was direct: "Unless you've got pretty deep finance knowledge, you cannot just spin up an app like this and expect it to work for real clients. People are making business decisions with your information. It cannot be wrong."

The implication for finance leaders considering Claude Code: the technical barrier has collapsed. The professional bar has not.

What to do this quarter

For any module you build that touches cash or the balance sheet, create a parallel test using a simple Excel model you trust. Run the same inputs through both. If the outputs match, the logic is correct. If they diverge, use your finance knowledge to identify where Claude's implementation has broken down and explain the correct behaviour in your next prompt.

4. How do you build integrations with accounting software in Claude Code?

Build the first integration from first principles; use it as a template for every integration after.

Kevin built Xero first. He described it as the most technically involved integration because he had no prior experience with API connections. It took approximately one day.

Once Xero was done, QuickBooks and Sage together took another day. The pattern was identical: understand how the API exposes data, write a Claude Code prompt describing how to map the API response to the app's internal data structure, verify the mapping against known actuals, and iterate.

His prediction: a Stripe integration from scratch could be completed in two hours. Any standard REST API with documentation follows the same pattern once you have done it once.

The integrations enable a core workflow: import actuals directly from the accounting system, map nominal codes to the app's P&L and balance sheet structure, assign timing profiles and VAT treatment per nominal, and then run the three-way model against real figures rather than demo data.

What to do this quarter

If your firm uses Xero, QuickBooks, or Sage, read the API documentation for your system before writing any prompt. Understanding the data structure your accounting platform exposes will make your Claude Code integration prompt significantly more precise and the output significantly more reliable.

5. How do you manage security and data privacy for a Claude Code finance app?

Security is not an afterthought; it is the foundation that makes client adoption possible.

Kevin's security architecture:

• Authentication: Multi-factor authentication across GitHub, Supabase, and Vercel — email, authenticator app, and passkey

• Data storage options: Offline mode (data stays local, nothing touches the cloud) and cloud mode (Supabase with SOC 2 compliance)

• Encryption: Optional zero-knowledge encryption, meaning even Kevin cannot read client data if the client enables it — only ciphertext is visible

• Access control: Row-level security enforced at the database layer; email, Google, and Microsoft SSO support

• AI usage: Entirely opt-in; clients who do not want AI functionality simply do not enter an API key

• Integration security: OAuth 2.0 with PKCE; connector tokens are never stored server-side

• Ongoing hygiene: Regular scans for security leaks, scheduled internal audits, strict content security policy

Before commercial release, Kevin has committed to an independent third-party security test. His framing: "Try your hardest to break the security on this app."

For finance leaders evaluating Claude-based tooling more broadly, the same principles apply: use Claude Enterprise (which does not train on your data), involve your legal team before sharing confidential data, and build a one-page internal policy on approved use cases before any team-wide rollout.

See Anthropic's Trust and Safety documentation for the full enterprise policy framework.

What to do this quarter

Before sharing any live financial data with a Claude-based tool, confirm the subscription tier you are using, confirm your legal team has reviewed the use case, and document which data categories are approved for use. This one-page policy will save significant friction during any internal audit.

6. How much does it cost to build and run a finance app with Claude Code?

At scale, a custom-built FP&A application runs at approximately $160 per month in infrastructure.

Kevin's monthly stack:

• Claude Max Plan (development environment + AI) — ~$100

• Supabase (data hosting, SOC 2 compliant) — ~$30–35

• Vercel (app deployment) — ~$30–35

• Total — ~$160/month

Against this, Kevin eliminated approximately $8,000 per year in software fees from his previous FP&A provider. The payback period on development time was measured in weeks.

On token consumption: Kevin has never run out of tokens on the Max Plan, despite building continuously for six weeks. His key insight on token management: "If you feed some massive plan into Claude Code, it's just gonna chew up your tokens so, so, so quickly." Iterating step by step is both faster and more economical than front-loading a large context.

What to do this quarter

Before committing to a Claude Code build project, calculate your current annual software spend on tools that do not fully meet your needs. If that figure exceeds $2,000, a custom build almost certainly delivers positive ROI within the first year, even accounting for development time.

7. Should finance teams buy existing FP&A software or build with Claude Code?

The decision depends on your role, your organisation's governance structure, and your specific frustrations with existing tools.

Kevin's framing:

If you're employed in a company, that's going to limit your ability to do what you want — enterprise issues, data issues. If you're a fractional CFO, you can go crazy and code your own app.

For employed finance leaders, the build path carries procurement, IT, legal, and data governance overhead that makes it genuinely complex. For fractional CFOs, independent controllers, and finance consultants, those constraints are significantly reduced.

The trigger for Kevin was not ambition. It was accumulated frustration: "I cannot just have a tool that works the way I want. I'll try a different one. Okay, that was fixed there, but now this thing's wrong." Claude Code does not solve the problem of bad software. It removes the dependency on software vendors entirely.

The question to ask before deciding: is the gap between what your current tool does and what you need it to do wide enough to justify six to eight weeks of iterative development? For Kevin, the answer was unambiguous. For a 200-person finance team on an enterprise ERP, it probably is not.

For teams that sit between those extremes — looking to automate specific recurring workflows without building an entire application — Claude Cowork and Zapier offer faster time-to-value. See our Claude Code, Claude Cowork, and Zapier finance automation playbook for a comparison of those approaches.

What to do this quarter

Write down the three things your current FP&A tool does that consistently frustrate you. If any of them have been on your list for more than 12 months with no resolution from the vendor, that is a signal the build path is worth exploring. If all three could be addressed with a better prompt or a Zapier workflow, start there first.

A simple 6-week Claude Code build plan for finance professionals

Week 1: Install Claude Code (30–45 minutes). Build a working P&L with at least three drivers using demo data. Get comfortable with the approve/continue/iterate cycle.

Week 2: Refine the P&L until you are satisfied with the driver logic, layout, and output format. Use screenshots to communicate UI preferences. Create your backlog file.

Week 3: Build the balance sheet. Map the P&L to balance sheet. Verify the linkage using a parallel test in Excel or Google Sheets before proceeding.

Week 4: Build the cash flow statement. Test the three-way link rigorously. Include at least one prepayment and one accrual scenario to stress-test the cash timing logic.

Week 5: Add the first integration with your primary accounting system. Import live actuals into your model and verify against known figures.

Week 6: Build reporting output (at minimum: one formatted summary view, one variance analysis view). Share with one trusted colleague or client for feedback. Start working down your backlog.

FAQ: Building Finance Apps with Claude Code

Can a finance professional with no coding experience really build a production-ready app in Claude Code?

Yes. Kevin Steele did it in six weeks, with zero prior coding experience, using only the iterative method described in this article. The app is now in live use with paying clients.

How long does it take to build a three-way financial model in Claude Code?

A basic working P&L with drivers can be operational in 10 to 15 minutes. A full three-way model with accurate prepayments, accruals, and balance sheet linkage took Kevin approximately four weeks of iterative development.

What accounting software integrations can you build with Claude Code?

Any system with a REST API and documentation is buildable. Kevin built Xero, QuickBooks, and Sage integrations. He estimates a Stripe integration would take approximately two hours.

How much does it cost to build and run a Claude Code finance app?

Approximately $160 per month covering Claude Max Plan ($100), Supabase ($30–35), and Vercel ($30–35). Kevin saves approximately $8,000 per year in software fees previously paid to a third-party FP&A provider.

How do you ensure financial accuracy when Claude Code builds the logic?

You act as QA at every iterative step. You verify each module's output against your own financial knowledge and a parallel test model. Claude handles the code; you handle the financial judgment. The two cannot be separated.

Is a Claude Code finance app secure enough for real client data?

With proper architecture (SOC 2 compliant hosting, multi-factor authentication, row-level database security, and optional zero-knowledge encryption), yes. A third-party security audit before any commercial release is strongly recommended.

Should I buy existing FP&A software or build with Claude Code?

Buy if you are in an employed role with enterprise governance constraints or if your needs are largely met by existing tools. Build if you are independent, if your recurring frustrations with existing tools are specific and persistent, and if you have the finance depth to act as QA throughout the build process.

Closing thought: Claude Code did not lower the technical bar. It removed it entirely.

The story from this session is not really about software.

It is about what happens when the execution constraint disappears. Kevin did not become a developer. He became the product manager, quality assurance function, and client delivery team for a piece of software that no vendor had built the way he needed it built.

The technical barrier to building custom finance tools has effectively collapsed. What remains is what has always separated good finance professionals from great ones: the depth to know when a number is wrong, the rigour to verify before signing off, and the judgment to decide what is worth building in the first place.

Claude Code is not the opportunity. The opportunity is what you do when the thing that was stopping you no longer exists.

To go deeper on the broader Claude toolkit for finance — including how Claude Chat, Cowork, and Code work together as a three-layer automation stack — read the CFO Connect Claude for Finance Teams playbook.