Kepler's Verifiable AI System Outperforms Frontier Models on Financial Document Analysis
Kepler, a platform built for buy-side financial research, extracts the correct line item from SEC filings 94% of the time. Frontier language models alone achieve 38-46% accuracy on the same task. The company detailed its architecture in a recent Anthropic customer profile.
The gap reflects a fundamental design choice: Kepler treats language models as one stage in a larger system, not the entire solution. The platform combines AI for interpretation and reasoning with deterministic code for data retrieval, calculation, and citation.
How the Architecture Works
A financial ontology sits between the two components. This proprietary dictionary maps analyst terminology-EBITDA, free cash flow, segment revenue-to exact line items in underlying filings. Every formula is explicit. Every calculation is reproducible.
Language models interpret the analyst's question and decompose it into a plan. Code retrieves data, runs calculations, and generates citations. The model never invents a number.
"One wrong assumption early in a financial analysis breaks everything downstream," said Vinoo Ganesh, CEO of Kepler. "On our workloads, Claude was the model that consistently held the plan together. Other models would start strong and then quietly drop a constraint by step five."
Implications for Daily Workflows
Every figure in a Kepler answer traces back to its filing, page, and line item. An analyst can defend a number in an investment committee memo at the level of the underlying 10-K.
Calculations remain reproducible across runs. The same query returns the same number each time. Outputs are auditable end-to-end, eliminating rework during compliance reviews or examiner requests.
Current Deployment and Scale
Buy-side analysts at private equity firms, hedge funds, and investment banks currently use Kepler. The platform indexes 26 million SEC filings, 50 million additional public documents, and 1 million private documents across 14,000 companies and 27 global markets.
The company built the platform in under three months, informed by interviews with 147 financial firms. Expansion into private credit underwriting is underway.
Learn more about AI for Finance and how Claude is being applied in regulated industries.
Your membership also unlocks:
