Key Takeaways
- • J.P. Morgan's Sarah Gang and Gabriele Butti describe machine learning as a tool for extracting, standardizing, and validating data from non-standard private credit documents, with the biggest time savings in data extraction, which the article says falls from about 45 minutes per company to 30 seconds.
- • Traders ranked machine learning as the second most influential technology in the industry over the next three years, behind generative AI, with 18% of traders seeing it as having the greatest influence, according to the J.P. Morgan Markets e-Trading Survey.
- • Lumonic (part of PitchBook) reports third-party underwriting technology adoption among roughly 150 private credit lenders doubled year-on-year from about 10% to 20%, with nearly all growth driven by AI-native tools.
- • Predictive models can assign a probability of default across a portfolio, allowing analysts to prioritize names with higher default risk and intervene earlier.
- • Industry participants stress that human judgment remains essential, given model accuracy requirements, the consequences of errors flowing into EBITDA and covenant calculations, and data security constraints on private credit information.
Summary
Machine learning is increasingly being deployed by private credit analysts to improve efficiency as the asset class scales. The sector was worth around $3.5 trillion in 2024, according to the article, and is expected to grow further. Analysts have traditionally spent days manually reviewing financial statements, extracting key figures, and forming credit risk views. Machine learning can now identify relevant data in seconds, according to J.P. Morgan, freeing analysts to focus on risk assessment and investment selection.
J.P. Morgan's Sarah Gang, Global Head of Underwriting, Credit Financing, said the bespoke nature of private credit documentation, which contrasts with the more uniform delivery of public market disclosures, makes extraction difficult. Oxane Partners Managing Director Kanav Kalia described machine learning as most useful in the ingestion and document intelligence layer, where data is extracted, classified, standardized, and validated before entering the analytical workflow. Kevin Hsu, Chief Executive of Lumonic, part of PitchBook, said underwriting and due diligence is among the most active areas of adoption.
Alternative data is also gaining use. A LexisNexis Risk Solutions 2025 Alternative Credit Data Impact Report surveyed 875 lending professionals across 10 countries, finding that 67% had increased confidence in decisions supported by alternative data, and three-quarters said it had improved portfolio performance. Data types include payment flows, supply chain activity, and receivables performance. Gang and Butti also described predictive models as a way to estimate default probabilities and flag covenant breaches earlier.
The article repeatedly emphasizes that human oversight remains necessary. Kalia noted that credit analysis is still fundamentally a judgment exercise, and Gang warned that people must be able to identify when outputs look wrong. Hsu cautioned that errors in EBITDA can flow into leverage ratios and covenant decisions, so firms must be able to halt AI systems. Security and confidentiality also constrain adoption, which Gang said is driving interest in proprietary systems.
J.P. Morgan's Sarah Gang, Global Head of Underwriting, Credit Financing, said the bespoke nature of private credit documentation, which contrasts with the more uniform delivery of public market disclosures, makes extraction difficult. Oxane Partners Managing Director Kanav Kalia described machine learning as most useful in the ingestion and document intelligence layer, where data is extracted, classified, standardized, and validated before entering the analytical workflow. Kevin Hsu, Chief Executive of Lumonic, part of PitchBook, said underwriting and due diligence is among the most active areas of adoption.
Alternative data is also gaining use. A LexisNexis Risk Solutions 2025 Alternative Credit Data Impact Report surveyed 875 lending professionals across 10 countries, finding that 67% had increased confidence in decisions supported by alternative data, and three-quarters said it had improved portfolio performance. Data types include payment flows, supply chain activity, and receivables performance. Gang and Butti also described predictive models as a way to estimate default probabilities and flag covenant breaches earlier.
The article repeatedly emphasizes that human oversight remains necessary. Kalia noted that credit analysis is still fundamentally a judgment exercise, and Gang warned that people must be able to identify when outputs look wrong. Hsu cautioned that errors in EBITDA can flow into leverage ratios and covenant decisions, so firms must be able to halt AI systems. Security and confidentiality also constrain adoption, which Gang said is driving interest in proprietary systems.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.