· 5 min read
Good financial analysis is never one-dimensional. An investment thesis needs a bull case and a bear case. A market sizing needs top-down and bottom-up validation. A financial health assessment needs ratio analysis and narrative context.
These multiple perspectives are exactly what agent teams provide. Each agent brings a different analytical lens, and the tension between their outputs produces more rigorous conclusions than any single analysis could.
Here are three financial analysis configurations, each using a different coordination pattern.
Investment decisions are fundamentally about evaluating competing arguments. The Advisory Debate pattern is purpose-built for this.
Bull Case Analyst builds the strongest possible case for the investment. It identifies growth catalysts, competitive advantages, market tailwinds, and upside scenarios. This agent is deliberately optimistic — its job is to articulate why this investment could outperform.
Bear Case Analyst builds the strongest possible case against. It identifies risks, competitive threats, market headwinds, execution challenges, and downside scenarios. This agent is deliberately skeptical — its job is to find every reason the investment could fail.
Risk Assessor operates independently from both the bull and bear cases. It identifies risks that don't fit neatly into an optimistic or pessimistic narrative: regulatory changes, key-person dependencies, technology shifts, macroeconomic sensitivity. It assigns probability and impact estimates to each risk factor.
Moderator synthesizes all three perspectives into a balanced assessment. It evaluates the strength of the bull and bear arguments, weighs the identified risks, and produces a recommendation with explicit confidence levels and key assumptions.
The adversarial structure forces comprehensive analysis. The Bull Case Analyst can't ignore weaknesses — the Bear Case Analyst will highlight them. The Bear Case Analyst can't manufacture risks — the Bull Case Analyst will challenge unsupported concerns. The Moderator ensures neither side dominates the conclusion.
This mirrors how the best investment committees actually work: structured disagreement followed by synthesis.
Market sizing benefits from multiple independent methodologies. If top-down and bottom-up approaches converge on similar numbers, confidence is high. If they diverge significantly, you've identified an assumption worth investigating.
Top-Down Analyst starts from the total market and narrows down. It begins with industry-level revenue data, applies segmentation filters (geography, customer type, price point), and arrives at the addressable market by working from macro to micro. This approach is fast and grounded in published data, but can miss micro-level realities.
Bottom-Up Analyst starts from individual transactions and scales up. It estimates the number of potential customers, average revenue per customer, and purchase frequency, then multiplies up to total market size. This approach is grounded in unit economics but depends on accurate assumptions about customer counts and behavior.
Comparables Analyst sizes the market by analogy. It identifies similar markets (same product category in different geographies, adjacent product categories in the same geography) and extrapolates based on known scaling relationships. This approach serves as an independent sanity check on both top-down and bottom-up estimates.
The value isn't in any single number — it's in the convergence or divergence across methods.
Convergence (all three within 20% of each other): High confidence in the market size range. The assumptions behind each method are likely sound.
Partial divergence (two agree, one differs): Investigate the outlier's assumptions. It either caught something the other two missed, or it has a flawed input assumption.
Full divergence (all three materially different): The market is poorly understood. Don't pick a number — investigate why the methods disagree. That disagreement is the most valuable output of the analysis.
Assessing a company's financial health requires extracting data, computing metrics, and then translating those metrics into a narrative. Each step depends on the previous one.
Data Extractor takes financial statements (income statement, balance sheet, cash flow statement) and organizes the raw data into a structured format. It normalizes for accounting differences, identifies one-time items that should be excluded from recurring analysis, and flags data quality issues (restatements, changes in accounting methodology, missing periods).
Ratio Analyst takes the structured data and computes a comprehensive set of financial ratios: profitability (gross margin, operating margin, net margin, ROE, ROIC), liquidity (current ratio, quick ratio, cash conversion cycle), leverage (debt-to-equity, interest coverage, debt-to-EBITDA), and efficiency (asset turnover, inventory days, receivables days). Each ratio is compared against industry benchmarks and the company's own historical trends.
Narrative Writer takes the ratio analysis and produces the story. Numbers without narrative are meaningless in financial analysis. This agent identifies the key themes (improving profitability but deteriorating liquidity, strong growth funded by increasing leverage), connects them to business context, and produces an assessment that a non-financial stakeholder can understand and act on.
Each stage genuinely requires the previous stage's output. You can't compute ratios without clean data. You can't write a narrative without computed ratios. The pipeline structure enforces this natural dependency while keeping each agent focused on its specialty.
Financial analysis has a higher accuracy bar than most agent team use cases. A wrong number in a market overview is embarrassing. A wrong number in a financial model can mean millions in misallocated capital.
Source everything. Every data point in the output should trace to a specific source. "Revenue grew 23%" is insufficient. "Revenue grew 23% per the FY2025 10-K filing, page 47" is verifiable.
Flag uncertainty explicitly. Agent teams should distinguish between facts (directly from financial statements), calculations (derived from facts using standard methodologies), and estimates (assumptions or projections). Each category carries different reliability.
Cross-validate. Use the convergence principle from the market sizing example. When multiple analytical approaches agree, confidence is warranted. When they disagree, flag the discrepancy rather than picking a number.
Never present agent output as final financial analysis. Agent teams produce structured analytical starting points. A qualified human analyst should review the output before any investment decision. This isn't a limitation of the technology — it's basic fiduciary responsibility.
Match the pattern to the analytical need:
Most financial analysis projects combine multiple patterns. A full due diligence might use Parallel Workers for market sizing, Advisory Debate for the investment thesis, and Sequential Pipeline for the financial health assessment — all feeding into a final synthesis.