AI for Financial Projections in Startups

1. Preparing the Foundation: Inputs and Assumptions

The quality of any financial model, AI-generated or not, is dictated by the quality of its inputs. The principle of "garbage in, garbage out" is especially true here. Before we can prompt an AI to build a model, we must provide it with structured historical data and clear forecasting assumptions.

A highly effective way to structure this is by creating two simple tabs in an Excel or Google Sheet:

• Performance: Contains your historical financial statements (Income Statement, Balance Sheet, Cash Flow Statement). For a new venture with no history, this would be your initial balance sheet (e.g., initial investment) and perhaps some key operational metrics from market research.
• Assumptions: A clear table outlining the drivers for your forecast. This includes items like revenue growth rates, margins, and working capital days. You can also define different scenarios (e.g., base, best, worst) here.

The following video explains this preparation step clearly.

A Note on Data Privacy
As we've discussed, uploading sensitive data is a concern. For your startup, this is less of an issue, but the principle is important. If you were working with confidential data, you could adapt the technique you learned previously: instead of uploading the file, you would describe the structure and headers to the AI.

For instance, you could tell Gemini: "My 'Assumptions' tab has columns: 'Line Item', 'Forecasting Method', 'Base Case Value', 'Best Case Value'. The 'Performance' tab contains a standard income statement for the last 3 years." This provides the necessary context without exposing the raw data, and the AI can guide you with formulas and logic. The video "How I Perform a Financial Analysis With AI in 5 minutes" demonstrates this principle well in its second half (LINK, from 03:02 onwards), which you may recall from an earlier module.

2. The Master Prompt: Building a 3-Statement Model

With your inputs prepared, you can now instruct the AI to build the model. A single, comprehensive prompt is far more effective than a series of small requests. This "master prompt" should act as a detailed project brief for your AI analyst.

A great prompt for financial modeling has several key components:

1. Role & Objective: Tell the AI what it is (e.g., "a financial modeling assistant") and what to do ("Build a fully linked three-statement model").
2. Describe Inputs: Explain where to find the data (e.g., "The uploaded Excel file has two tabs: 'performance' with historicals and 'assumptions' with drivers").
3. Forecasting Logic: Specify how to use the assumptions to project the future.
4. Linkages & Schedules: Instruct the AI to ensure the three statements are linked and to create necessary supporting schedules (e.g., for debt, working capital).
5. Integrity Checks: Demand a balanced model by including checks (e.g., Assets = Liabilities + Equity).
6. Formatting & Output: Define the desired output format (e.g., "Keep historicals, append 5 forecast years, use transparent formulas").

This next video provides a perfect, step-by-step breakdown of how to construct such a prompt.

Here is a template based on that structure, which you can adapt and use in Gemini.

Actionable Prompt Template:

"Act as an expert financial modeling assistant. Your objective is to build a fully linked, three-statement financial model (Income Statement, Balance Sheet, and Cash Flow using the indirect method) with a 5-year projection horizon.

INPUTS:
I have uploaded an Excel file with two tabs:

1. 'Performance': This contains the historical financial statements and defines the desired reporting format.
2. 'Assumptions': This contains all forecasting drivers and rules, organized by line item, forecasting method, and driver value for the 'Base Case'.

FORECASTING LOGIC:
For each line item in the financial statements, apply the corresponding forecasting method and driver value from the 'Assumptions' sheet to project the next 5 years. Use standard accounting equations and relationships.

LINKAGES & SCHEDULES:

• Ensure all three statements are fully linked (e.g., Net Income links to Retained Earnings and Cash Flow Statement; ending cash on CFS links to the Balance Sheet).
• Create supporting schedules for: Working Capital, CAPEX and Depreciation, Debt Roll-forward, and Equity Roll-forward. Ensure these schedules feed into the main financial statements correctly.

INTEGRITY CHECKS:
Include a 'Checks' section at the bottom of the model. This section must include a row that validates 'Total Assets = Total Liabilities + Equity' for all forecast years, showing 'OK' or 'Error'. The balance sheet must balance.

OUTPUT:
Generate a new Excel file. The 'Performance' sheet should be updated to include the 5 forecast years appended to the historical data. All formulas in the forecast period must be transparent and traceable. Do not use hard-coded numbers in formulas; link everything back to the assumptions sheet or other calculated cells."

Once executed, Gemini (with its "Advanced Data Analysis" capability) will process the file and generate the model, often providing a summary of its work and the downloadable file.

3. Critiquing the Model: The Evaluator-Optimizer Pattern

The first draft from the AI is just that—a draft. Your expertise is crucial for the next step: critique and refinement. A powerful mental model for this is the Evaluator-Optimizer pattern. The AI acts as the "Optimizer" by generating the model, and you (or another AI instance) act as the "Evaluator" to critique it, prompting iterative improvements.

To apply this pattern, you can open a new conversation with Gemini, re-upload the model it just created, and give it a new role: a skeptical investor or a senior finance director.

The same CFA Institute article provides an excellent, detailed example of a prompt used to validate financial metrics. While the context is slightly different, the structure is a perfect template for critiquing your model.

Here’s how you can adapt that concept to critique your startup's financial model.

Actionable Critique Prompt:

"Act as a skeptical venture capital analyst reviewing a financial model for a new venture. I have uploaded the model.

Your task is to critique the model's assumptions and identify key risks.

1. Review Assumptions: Go through the 'Assumptions' tab. For each key driver (e.g., revenue growth, COGS margin, customer acquisition cost), comment on whether it seems realistic, optimistic, or pessimistic for a startup in the [your industry] sector. Suggest alternative, more conservative assumptions.
2. Identify Key Risks: Based on the model, what are the top 3-5 financial risks to this business? (e.g., 'The model is highly sensitive to customer churn rate', 'The break-even point is not reached until year 4, indicating high cash burn').
3. Check for Nuance: Are there any nuances the model might be missing? For example, does the revenue model account for seasonality? Does the cost structure reflect economies of scale over time?
4. Provide a Summary: Write a brief executive summary of your findings, concluding with your overall confidence in this financial plan and suggesting specific areas that need to be revised for greater credibility."

This process transforms the AI from a simple calculator into a thinking partner, helping you pressure-test your own assumptions.

4. Stress-Testing with Sensitivity Analysis

A key part of any robust financial critique is understanding how the model's output changes when key assumptions are varied. This is where sensitivity analysis comes in. You can ask the AI to build this for you.

After your main model is built, you can ask Gemini to add this analysis.

Actionable Prompt for Sensitivity Analysis:

"Using the financial model you just created, perform a sensitivity analysis.

1. Identify the top two most sensitive assumptions in the model (e.g., 'Revenue Growth Rate' and 'EBITDA Margin').
2. Create a 2-dimensional sensitivity table.
   • The rows should represent the 'Revenue Growth Rate', with a range of [Your Assumption - 5%] to [Your Assumption + 5%] in 2.5% increments.
   • The columns should represent the 'EBITDA Margin', with a range of [Your Assumption - 2%] to [Your Assumption + 2%] in 1% increments.
3. The values in the table should be the resulting 'Net Income' for Year 5.
4. Present this as a new tab in the Excel file."

This final step adds a layer of professional rigor to your model, showing that you've considered a range of outcomes beyond the base case.

Conclusion

In this lesson, you've learned a complete, end-to-end workflow for leveraging AI in financial modeling for a new venture. By combining your domain expertise with AI's processing power, you can dramatically accelerate the creation of robust, well-reasoned financial projections.

Key Takeaways:

• Structured Inputs are Essential: A well-organized Performance (historicals) and Assumptions sheet is the prerequisite for a good AI-generated model.
• Use a Master Prompt: A single, detailed prompt that specifies the role, inputs, logic, linkages, checks, and output format yields far better results than a series of small requests.
• Embrace the Evaluator-Optimizer Pattern: Use AI to generate the first draft (the "Optimizer") and then use a new AI instance (or your own critical eye) as the "Evaluator" to critique and refine the model.
• Prompt for Critique and Stress-Testing: Explicitly ask the AI to act as a skeptical analyst and to build sensitivity analyses to pressure-test your model's credibility.

Next Up:
With your business plan, pitch deck, and financial model now in place, you have a solid internal foundation for your startup. In our next lesson, we'll turn our attention outward and begin generating the content needed to attract your first customers. We will learn how to use AI to generate marketing copy for key channels like landing pages and email announcements.