The revolution of financial analysis with AI

A controller analyzing financial statements from 5 subsidiaries takes between 3 and 5 days to produce a consolidated report with ratios, trends, and alerts. With AI, that same analysis can be completed in 2 hours. Not because AI is smarter than the controller, but because 80% of the time is spent on mechanical tasks: extracting data, calculating ratios, comparing periods, and formatting tables.

AI does not replace financial judgment. It amplifies it. A CFO who understands how to use AI for financial analysis can process 10 times more information, detect patterns that would otherwise go unnoticed, and devote their time to what truly matters: interpreting the numbers and making decisions.

In this module we will cover the 5 areas where AI transforms financial analysis:

• Ratio analysis: calculating, interpreting, and contextualizing financial ratios with specialized prompts.
• Benchmarking: comparing metrics against competitors and sectors using public data.
• Balance sheet reading: extracting insights from complex financial statements in minutes.
• Financial modeling: building valuation and projection models with AI assistance.
• Prompt templates: ready-to-use prompts for your daily financial workflow.

Ratio analysis with AI assistance

Financial ratios are the universal language of corporate finance. AI radically changes how we calculate, interpret, and communicate them.

Automated ratio calculation

The most basic use case is mass calculation. You can upload a balance sheet and an income statement to Claude or ChatGPT and ask it to compute 30 ratios in seconds. But that is only the beginning. The interesting part comes next.

When you ask AI to calculate ratios, do not just ask for the number. Ask for context. A current ratio of 1.2 means nothing without knowing that the pharmaceutical sector median is 2.1 and that your company had 1.8 two years ago.

Contextualized interpretation

AI can cross-reference multiple dimensions of analysis that would take a human hours to combine:

• Temporal: how each ratio has evolved over the last 5 fiscal years.
• Sectoral: where the company stands relative to direct competitors and sector medians.
• Regulatory: whether any ratio is approaching critical thresholds (bank covenants, solvency requirements).
• Causal: which balance sheet or income statement factors are driving each ratio, and in what direction.

A well-constructed prompt for ratio interpretation should include: the financial data, the company's sector, its size (revenue, employees), geography, and target ratios or covenants if they exist.

Ratio alert detection

One of the most powerful applications is configuring AI as an early warning system. You can define thresholds for each ratio and ask it to analyze financial statements specifically searching for deviations. For example:

• Current ratio below 1.5 (yellow alert) or below 1.0 (red alert).
• Days sales outstanding growing more than 15% year-over-year.
• EBITDA margin dropping more than 200 basis points from the prior quarter.
• Debt/EBITDA ratio exceeding 3x (a typical bank covenant).

AI reviews all ratios against your thresholds and generates a prioritized alert report. What previously required a manual ratio-by-ratio review is now fully automated.

DuPont analysis with AI

The DuPont model decomposes return on equity (ROE) into its three components: net profit margin, asset turnover, and financial leverage. It is a classic tool in financial analysis, but most controllers compute it only once a year because disaggregating it by business unit or subsidiary is tedious.

With AI you can perform a DuPont breakdown by business line, by geography, by quarter, and visualize exactly where profitability is being created or destroyed. What was once an academic exercise becomes an operational management tool.

Automated financial benchmarking

Financial benchmarking involves comparing your company's metrics with those of your sector, direct competitors, or internal targets. It is one of the most time-consuming tasks in finance because it requires gathering data from multiple sources, normalizing it, and building the comparisons.

Data sources for benchmarking

AI can work with benchmarking data from several sources:

• Public sector reports: industry classification databases, national statistics offices, central bank data (such as central balance sheet offices), and Eurostat. You can upload PDFs of these reports to Claude and ask it to extract sector medians.
• Public company filings: your competitors' annual accounts are public. You can download them from commercial databases like SABI, Orbis, or directly from company registries and analyze them with AI.
• Financial databases: Bloomberg, Refinitiv, S&P Capital IQ. If you have access, you can export data and process it with AI.
• Audit and consulting firm reports: Deloitte, PwC, EY, and KPMG publish annual sector benchmarks that are a goldmine.

Process automation

The AI-assisted financial benchmarking workflow has 4 steps:

1. Collection: upload your company's financial statements along with reference data (sector reports, competitor filings).
2. Normalization: ask AI to harmonize accounting criteria (IFRS vs local GAAP, different depreciation policies, different consolidation perimeters).
3. Comparison: generate comparative tables with percentiles. Your company at the P25, P50, or P75 of the sector for each metric.
4. Narrative: produce an executive report with key conclusions and areas for improvement.

A controller can complete this entire exercise in an afternoon instead of the 2 to 3 weeks it traditionally requires.

Operational KPI benchmarking

Beyond classic financial ratios, AI enables benchmarking of operational metrics that correlate with financial performance:

• Cost per employee (labor productivity).
• Revenue per FTE (revenue generation efficiency).
• Capex as a percentage of revenue (investment intensity).
• Working capital days (cash cycle efficiency).
• SG&A as a percentage of revenue (administrative efficiency).

Intelligent balance sheet reading

Reading a balance sheet is easy. Understanding it is another matter. A 50-page balance sheet with notes, line item breakdowns, and intercompany transactions can take an entire day of careful reading. AI compresses that process.

Extracting data from financial statements

The first step is converting financial statements into data that AI can process. There are several scenarios:

• PDF with tables: Claude processes PDFs directly and extracts tables with high accuracy. You can upload annual accounts filed with the company registry and ask it to extract the balance sheet and income statement.
• Excel: ChatGPT with Data Analysis and Claude with Code Interpreter process spreadsheets. This is the ideal format because it preserves the numerical structure.
• XBRL: the standard digital financial reporting format. AI can parse XBRL files and extract the taxonomies.

Vertical and horizontal analysis

Two classic analyses that AI accelerates enormously:

Vertical analysis: each line item as a percentage of total assets (on the balance sheet) or net revenue (on the income statement). AI calculates the percentages and highlights line items that deviate from the typical sector structure.

Horizontal analysis: the change in each line item between periods. This is where AI shines because it can analyze 5 or 10 fiscal years simultaneously and detect trends that a 2-period analysis would not reveal. A line item growing at 5% annually seems stable, but if it has been growing at that rate for 7 years while revenue remains flat, there is a structural problem.

Detecting signals in the balance sheet

AI can act as a first-line financial auditor, searching for warning signals in the balance sheet:

• Receivables growing without revenue growth: possible deterioration of collections or aggressive revenue recognition.
• Inventory growing faster than sales: possible obsolescence or overproduction.
• Increase in intangibles without acquisitions: possible aggressive capitalization of expenses.
• Short-term debt replacing long-term debt: possible refinancing stress.
• Provisions systematically reversed: possible earnings management.

These signals are not proof of problems, but indicators that merit deeper investigation. AI detects them in seconds across a volume of data that would take an analyst hours to review.

AI-assisted financial modeling

Financial modeling is the core work of the analyst and the controller. Building a three-statement model (balance sheet, income statement, cash flow) in Excel can take between 2 and 5 days. AI does not build the model for you, but it accelerates every phase of the process.

Model structure

You can ask AI to generate the complete structure of a financial model: tabs, assumptions, formulas, balancing checks. It is especially useful for:

• Standard models: DCF, LBO, three-statement model. AI knows the canonical structure and can generate the skeleton with the correct formulas.
• Assumption definition: asking it to identify the 10 key assumptions you need to define for your sector and company size.
• Integrity checks: generating automatic verification formulas (assets = liabilities + equity, cash variation reconciles with the cash flow statement).

Sensitivity and scenarios

Where AI adds the most value in modeling is in sensitivity analysis. Traditionally, an analyst builds 3 scenarios (base, optimistic, pessimistic) manually. With AI you can:

• Generate 10 or 20 scenarios combining variations in the main assumptions.
• Run Monte Carlo simulations to understand the distribution of possible outcomes.
• Identify the variables with the greatest impact on the final value (tornado charts).
• Create stress scenarios based on real historical events (COVID, 2008 financial crisis, 2022 rate hikes).

Model documentation

One of the classic problems of financial modeling is the lack of documentation. Models are passed from analyst to analyst without anyone knowing why certain assumptions were chosen. AI can:

• Generate automatic model documentation: what each cell calculates, where each assumption comes from.
• Create a user manual for the model so anyone on the team can update it.
• Produce an assumptions memo that justifies each assumption with market data.

Prompt templates for finance

Generic prompts do not work in finance. A prompt that says "analyze these financial statements" will produce a generic result. Effective financial prompts include specific context, target metrics, and the desired output format.

Template 1: Monthly ratio analysis

"Act as a senior financial controller. I will provide the balance sheet and income statement of [COMPANY] as of [MONTH/YEAR] close. Calculate the following ratios: current ratio, acid test, days sales outstanding, days payable outstanding, inventory turnover, EBITDA margin, net margin, ROE, ROA, leverage ratio, interest coverage, and net debt/EBITDA. For each ratio: (1) show the current value, (2) compare it with the same month of the prior year, (3) indicate whether it is above or below the target threshold [include thresholds], (4) note the trend over the last 6 months if data is available. Present results in a table and at the end include a 100-word paragraph with the top 3 alerts."

Template 2: Sector benchmarking

"Compare the financial metrics of [COMPANY] with sector medians for [INDUSTRY CODE] in [COUNTRY]. Sector data available in the attached document [upload central balance sheet report]. Metrics to compare: EBITDA margin, net margin, ROE, liquidity, leverage, days sales outstanding, and days payable outstanding. For each metric, indicate the percentile where the company falls and provide a 1-to-5 star rating. At the end, identify the 3 areas where the company has the greatest room for improvement."

Template 3: Express due diligence

"Review the attached financial statements of [TARGET COMPANY] as if you were performing a preliminary financial due diligence. Look for: (1) revenue quality (recurrence, customer concentration, seasonality), (2) EBITDA normalization (extraordinary items, related-party transactions, owner expenses), (3) balance sheet health (hidden debt, contingencies, asset quality), (4) operating cash flow trends. Present your findings as a list of red flags, yellow flags, and green flags."

Template 4: Board report

"Generate a 2-page executive financial report for the Board of Directors of [COMPANY] using the attached data. Structure: (1) Executive summary (5 lines), (2) P&L highlights with variance vs budget and vs prior year, (3) Balance sheet highlights focused on working capital and debt, (4) Cash flow with year-end forecast, (5) Top 3 financial risks, (6) Top 3 opportunities. Professional tone, no unnecessary jargon. Include actionable recommendations."

AI tools for financial analysis

Not all AI tools are suited for finance. These are the most relevant for a finance department:

General assistants

• Claude (Anthropic): the best for long financial documents. Context window of 1 million tokens. You can upload complete annual accounts and analyze them in one go. Very accurate at following complex analysis instructions.
• ChatGPT with Data Analysis: executes Python code in real time. Ideal for complex calculations, charts, and statistical analysis. Reduces the risk of arithmetic errors.
• Gemini with Google Sheets: direct integration with spreadsheets. Useful if your team works in Google Workspace.

Finance-specialized tools

• Bloomberg Terminal with AI: if you already have Bloomberg, its new AI features enable document analysis and insight generation directly on market data.
• Datarails: FP&A platform with integrated AI. Connects with your ERPs and generates automated reports.
• Cube: cloud FP&A with AI capabilities for planning and analysis.
• Stampli: AP automation with AI for invoice classification and approval.

ERPs with integrated AI

Major ERPs are integrating native AI:

• SAP Joule: AI assistant within SAP S/4HANA. Natural language queries on financial data.
• Oracle Fusion AI: cash flow predictions, transaction anomaly detection.
• Microsoft Dynamics 365 Copilot: predictive analytics in the financial module.

Limitations and risks

Using AI in financial analysis carries real risks that you must manage:

Arithmetic errors

LLMs are not calculators. They can make mistakes in arithmetic operations, especially with large numbers, nested percentages, or currency conversions. Solution: always use tools with code execution (Code Interpreter) for calculations, or verify with Excel.

Hallucinations about market data

If you ask AI for sector data without providing it, it may fabricate it. Plausible but false data. Solution: always provide reference data as input; never trust data the AI "remembers."

Confidentiality

Financial statements are sensitive information. Uploading your company's accounts to a cloud LLM has confidentiality and potentially regulatory implications. Solution: use enterprise plans that guarantee your data is not used for training (Claude Team, ChatGPT Enterprise) or on-premise solutions.

Confirmation bias

AI tends to confirm the narrative you present. If you say "the company is in good financial shape, analyze this data," it will find arguments to confirm it. Solution: always request a neutral analysis, with red flags and green flags, without biasing the prompt.

Ejercicio practico

Puntos clave