How to Analyze Financial Reports Locally with AI

Why financial data shouldn't leave your machine

Financial statements are the most sensitive documents an organization produces. A balance sheet reveals total assets, liabilities, and equity. An income statement shows revenue, margins, and profitability. A cash flow statement exposes liquidity position and cash burn. Together, they tell the complete financial story of a business.

When a CFO uploads these documents to a cloud AI service for analysis, that financial story passes through third-party infrastructure. The AI provider's terms of service may allow data retention for model training. Even if they don't, the data transits external networks and resides temporarily on external servers. For public companies, this creates material non-public information (MNPI) exposure. For private companies, it exposes the financial position to the AI vendor's data handling practices.

This is not a theoretical concern. Financial data that reaches external systems becomes subject to the data handling, breach notification, and regulatory compliance practices of those systems. If a cloud AI vendor suffers a breach, your financial data could be exposed alongside thousands of other organizations' data. If an auditor asks how financial analysis is performed, "we upload our statements to a third-party AI" raises questions that "we analyze them locally on our secured workstations" does not.

The solution is local analysis. docrew processes financial documents on your machine. The PDFs never leave your computer. The extracted data, computed ratios, trend analyses, and comparison reports all stay local. You get AI-powered financial analysis with zero data exposure.

Extracting data from financial PDFs

Financial statements arrive as PDFs -- from your accounting system, from auditors, from portfolio companies, from public filings. Before analysis can begin, the numbers need to be extracted into a structured format.

This is harder than it sounds. Financial PDFs are not simple tables. They contain hierarchical line items (Assets > Current Assets > Cash and Cash Equivalents), subtotals at multiple levels, comparative columns (current year vs. prior year), footnote references, and formatting variations that make automated extraction tricky.

docrew handles financial PDF extraction by reading the document structure contextually. The agent identifies the statement type (balance sheet, income statement, cash flow statement), recognizes the hierarchical structure of line items, and extracts values with their full context.

Here is the workflow for extracting data from a set of quarterly financial statements.

Step 1: Collect the PDFs. Place all financial statement PDFs in a folder. These might be quarterly reports, annual reports, or individual statements exported from your accounting system. Label them clearly (Q1-2025-financials.pdf, Q2-2025-financials.pdf, etc.) to help organize the output.

Step 2: Request extraction. Tell the agent: "Extract all financial statements from these PDFs. For each statement, extract every line item with its label, hierarchy level, and values for each period shown. Preserve the original statement structure and identify the statement type, reporting entity, and period."

Step 3: Review structured output. The agent produces spreadsheets with extracted data: one sheet per statement type, with line items preserving their hierarchy. A balance sheet extraction looks like:

Level | Line Item                    | Q4 2025    | Q4 2024
1     | ASSETS                       |            |
2     | Current Assets               |            |
3     | Cash and Cash Equivalents    | 2,450,000  | 1,890,000
3     | Accounts Receivable          | 3,200,000  | 2,870,000
3     | Inventory                    | 1,800,000  | 1,650,000
2     | Total Current Assets         | 7,450,000  | 6,410,000

The hierarchical structure is preserved, enabling both human review and programmatic analysis.

Ratio analysis

With extracted financial data, the agent can compute the financial ratios that controllers, analysts, and CFOs use to assess business health. Rather than pulling out a calculator or building formulas in a spreadsheet, you describe what you need and the agent computes it.

Liquidity ratios -- current ratio, quick ratio, cash ratio -- measuring the ability to meet short-term obligations.

Profitability ratios -- gross margin, operating margin, net margin, return on assets, return on equity -- measuring earnings relative to revenue, assets, or equity.

Efficiency ratios -- AR turnover, inventory turnover, days sales outstanding, days inventory outstanding -- measuring how well the company uses its assets.

Leverage ratios -- debt-to-equity, interest coverage, debt-to-assets -- measuring the debt structure.

The agent computes all requested ratios, shows the underlying values used in each calculation, and flags any ratios that fall outside typical industry ranges if you provide benchmarks.

A typical instruction: "Compute all liquidity, profitability, efficiency, and leverage ratios for the last four quarters. Show the trend for each ratio. Flag any ratio that deteriorated by more than 10% quarter-over-quarter."

Trend identification

Single-period ratios are useful but trends tell the real story. A current ratio of 1.5 is healthy. A current ratio that dropped from 2.1 to 1.5 over four quarters is a warning sign.

docrew identifies trends across multiple periods by analyzing extracted data from sequential financial statements. The agent examines each line item and computed ratio across time, identifying:

Directional trends. Revenue growing at 8% quarterly. Gross margin declining from 42% to 38% over six quarters. AR days increasing from 35 to 48 over a year.

Acceleration and deceleration. Revenue growth accelerating (5%, 7%, 9%, 12%) versus decelerating (12%, 9%, 7%, 5%). The direction matters less than the rate of change.

Seasonal patterns. Q4 revenue consistently 30% higher than Q1. Inventory buildup in Q3 ahead of holiday sales. Payroll costs spiking in Q1 from annual bonuses.

Anomalies. A single quarter where SGA expenses jumped 40% while revenue was flat. An inventory write-down in Q2 that distorted the margins. These stand out against the trend and warrant investigation.

The trend analysis output is a narrative report: "Revenue grew 7% year-over-year to $12.4M in Q4 2025. Gross margin contracted 2.3 percentage points to 38.1%, driven by raw material cost increases. Operating expenses remained controlled at 22% of revenue, consistent with the trailing four-quarter average. Net income declined 8% despite revenue growth, indicating margin pressure that merits attention."

This is the kind of analysis that a financial analyst produces after hours of work with spreadsheets. The agent produces it in minutes from the source PDFs.

Comparative analysis across periods

Finance teams regularly compare performance across periods: this quarter versus last quarter, this year versus last year, actual versus budget, subsidiary versus subsidiary.

docrew handles multi-document comparison by extracting data from all documents, aligning line items across periods, and computing variances.

Period-over-period comparison. The agent reads Q4 2025 and Q3 2025 financial statements, aligns every line item, and produces a variance report showing absolute change and percentage change for each item. Revenue up $200K (+7%). Cost of goods sold up $150K (+9%). The margin squeeze is immediately visible.

Year-over-year comparison. Same approach but comparing Q4 2025 to Q4 2024. This removes seasonal effects and shows underlying growth or contraction.

Actual versus budget. Provide the budget document alongside actuals. The agent extracts both, aligns the line items, and computes variances. Revenue $12.4M actual versus $13.0M budget (-4.6%). Marketing expense $1.2M actual versus $1.0M budget (+20%). The over-spend categories and under-performing revenue lines are immediately identified.

Multi-entity comparison. For organizations with subsidiaries, divisions, or business units, the agent can compare financial statements across entities. Which division has the highest margin? Where is AR growing fastest? Which unit is consuming the most cash?

The comparative output includes both the raw numbers and a summary highlighting the most significant variances -- the items a CFO would want to see first.

The docrew workflow for quarterly financial review

Here is a concrete workflow for a controller performing a quarterly financial review using docrew.

Preparation. Collect the following PDFs: current quarter financial statements (balance sheet, income statement, cash flow statement), prior quarter statements, same quarter prior year statements, and the current quarter budget.

Step 1: Extract all statements. Place all PDFs in a folder. Ask the agent to extract all financial data, maintaining statement type and period labels.

Step 2: Compute ratios. Request a full ratio analysis for the current quarter, with trailing four-quarter history.

Step 3: Run comparisons. Ask for three comparisons: quarter-over-quarter, year-over-year, and actual versus budget. Request that the agent highlight variances exceeding 5% of the base period.

Step 4: Generate trend report. Request a trend analysis covering the last four quarters, identifying directional trends, acceleration or deceleration, and anomalies.

Step 5: Produce executive summary. Ask the agent to synthesize the ratio analysis, comparisons, and trends into a one-page executive summary suitable for the CFO or board. Key metrics, significant variances, and items requiring attention.

The entire process takes 30 to 60 minutes of interaction time, replacing what typically takes a full day of spreadsheet work. The controller reviews and validates the output rather than producing it from scratch.

Practical scenario: quarterly review for a $50M company

A $50M revenue manufacturing company performs quarterly financial reviews. The controller collects financial statements from the ERP, auditor working papers, and budget documents. The review covers the parent company and three subsidiaries.

Without docrew: The controller spends two full days extracting data into Excel, building comparison sheets, computing ratios, creating charts, and drafting the quarterly review memo. The work is accurate but labor-intensive, and the analysis depth is limited by available time.

With docrew: The controller collects 16 PDFs (four statements each for four entities) plus the consolidated budget. docrew extracts all data in 20 minutes. The controller then requests ratio analysis, trend reports, inter-entity comparisons, and a consolidated variance analysis. The agent produces all of this in another 30 minutes of processing. The controller spends 2 hours reviewing, validating, and adding qualitative commentary. Total time: 3 hours instead of 16.

The additional time saved is often reinvested in deeper analysis. Instead of stopping at top-level variances, the controller can ask the agent to drill into specific line items: "Why did subsidiary B's inventory increase 25% while revenue was flat? Show me the inventory composition breakdown for the last four quarters." This kind of drill-down analysis rarely happens in the manual process because there's no time for it.

Handling different statement formats

Financial statement PDFs come in many formats. Statements from your own accounting system have one layout. Audited financials from your external auditor have another. Public company filings follow SEC formatting. Portfolio company reports follow whatever format the portfolio company uses.

docrew handles this variety without configuration. The agent reads each document contextually, identifying:

• The statement type based on content (a document listing assets and liabilities is a balance sheet, regardless of what it's titled)
• The reporting periods based on column headers and dates
• The line item hierarchy based on indentation, bolding, and subtotal patterns
• The values based on their position in the table structure

This means a controller can analyze a mix of internally generated statements, auditor-prepared statements, and third-party reports in a single workflow. The agent normalizes the data into a consistent structure regardless of the source format.

Privacy advantage for sensitive financial data

Every financial analysis performed with docrew happens entirely on your machine. Consider what stays local:

• Raw financial statements with detailed line items
• Computed ratios revealing liquidity, profitability, and leverage
• Trend data showing multi-quarter financial trajectory
• Variance analyses exposing budget misses and operational issues
• Executive summaries containing the most sensitive insights

None of this passes through external servers. For organizations where financial data privacy is a regulatory requirement or a competitive necessity, this is a fundamental advantage over cloud-based analysis tools.

The analysis is also reproducible. The source PDFs and the agent's output both reside on your system. An auditor can review the source documents, the extraction, and the analysis without involving any third party. The chain of custody for the financial data is clean.

Getting started

If you perform periodic financial analysis and want to keep your data local:

1. Collect your most recent financial statements as PDFs -- one quarter's worth of balance sheet, income statement, and cash flow statement.
2. Install docrew and point the agent at the folder.
3. Request extraction and ratio analysis to see the output quality.
4. Add comparative periods -- prior quarter or prior year -- and request variance analysis.
5. Build your recurring quarterly workflow with the analysis steps that matter most to your organization.

The first run establishes the extraction pattern and analysis framework. Subsequent quarters reuse the same approach with new data, making quarterly financial review a streamlined, private, and thorough process.