United StatesFinance and insurance2019 – ongoing

JPMorgan Chase

Machine-learning payment fraud decisions

JPMorgan said machine learning let it approve 1 million additional legitimate customers

WorkingCompany's own claimOfficial ↗

Problem

Payment fraud controls could reject legitimate customers while allowing fraudulent activity. Decisions also needed to occur quickly at the point of sale.

Approach

JPMorgan used machine learning in payment-fraud applications that made transaction decisions in milliseconds. In its shareholder letter covering 2018, it reported improved approval and rejection decisions alongside lower fraud losses.

Results

1 millionadditional legitimate customers JPMorgan said it could approve rather than decline for suspected fraudCompany's own claim[1]

approximately 1 millionadditional fraudsters JPMorgan said it could decline rather than approveCompany's own claim[1]

Takeaway

JPMorgan reported improvements on both legitimate approvals and fraud rejection. It did not specify the measurement window for the additional customer counts.

Sources

2018 Chairman and CEO Letter to Shareholders
JPMorgan Chase2019Company publication

Source summaries

[1]JPMorgan Chase2019

JPMorgan described live machine-learning fraud decisions and additional legitimate approvals. It also reported approximately 1 million additional fraudsters declined; its separate annual financial-benefit estimate was a forecast.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026