KPMG developed a generative AI tool that quickly and accurately analyzes unstructured data from private companies, including financial information. The tool mimics the analytical reasoning of a Chartered Professional Accountant (CPA), achieving 98% accuracy in interpreting financial statements. As a result, it extracts more data and delivers richer analyses in significantly less time.
KPMG automation consultants worked closely with CPAs to outline the logical processes accountants use to identify key financial metrics. The AI model uses thousands of specialized prompts to incorporate contextual information, such as industry-specific guidelines and financial details, enabling accurate and precise analysis.
For example, rather than searching for specific terms like “total debt,” which may not appear in every statement, the team created an AI-based persona that mirrors CPA-level thinking. This allows the program to intelligently identify relevant metrics, such as long-term and short-term debt, while excluding liabilities that do not incur interest. The model can also answer specific questions about the company, including key financial ratios, or even exposure to cryptocurrencies.
The tool features advanced multi-modal capabilities, enabling it to read files in any format while maintaining the document’s structure and logic. To mitigate risk, the team used anonymized financial documents in a secure environment.
KPMG integrated the AI tool into the client’s workflow. On a monthly basis, it ingests the financial statements of portfolio companies and outputs the necessary fields. Data analysts receive prompts to confirm accuracy before forwarding the information to investment analysts, all within a user-friendly interface for seamless validation.
