Artificial Intelligence at American Express
American Express is an American monetary providers firm headquartered in New York City, New York. The firm employs greater than 76,800 individuals globally and reported fourth-quarter 2025 income of $18.98 billion, with fiscal 2026 income progress guidance of 9-10 p.c.
American Express began making use of machine studying to fraud detection in 2010, making it one of many earliest adopters of AI amongst monetary providers corporations. Today, the corporate has provided entry to main AI instruments to almost all of its 76,800 colleagues globally and has scaled AI-assisted improvement instruments to greater than 11,000 engineering professionals, decreasing coding cycle occasions by over 30 p.c.
Its machine-learning-powered fraud detection mannequin monitors greater than $1.2 trillion in transaction worth yearly, producing a fraud choice in milliseconds for each card transaction worldwide. The firm is at the moment exploring greater than 70 generative AI use circumstances throughout the group and, by way of Amex Ventures, is investing in generative AI startups centered on belief and security, enterprise effectivity, and data-driven experiences.
For monetary establishments working at a worldwide transaction scale, AI techniques are more and more influencing how fraud threat, transaction approvals, and buyer retention workflows are managed operationally. In the case of American Express, many of those techniques are embedded immediately into high-volume decision-making processes reasonably than positioned as standalone AI initiatives.
Publicly out there reporting and firm documentation level to 2 notably seen purposes of AI throughout the group:
- Real-time fraud detection throughout transaction authorization
- Predictive buyer retention and service provider concentrating on
In each circumstances, machine studying techniques look at massive volumes of behavioral and transactional information to help quicker operational choices, cut back handbook evaluation workloads and enhance buyer expertise outcomes.
Real-Time Fraud Detection During Transaction Authorization
Fraud prevention is likely one of the most seen AI purposes inside American Express. The firm has mentioned utilizing machine studying techniques to judge transactions in actual time throughout fee authorization workflows.
According to a Harvard Business School Digital Initiative analysis, American Express applies machine studying fashions to research buying patterns, spending conduct, service provider exercise, and transaction anomalies throughout authorization requests.
The enterprise drawback is important for fee suppliers. Fraudulent transactions trigger direct monetary losses, whereas incorrectly declining authentic purchases can erode buyer belief and transaction quantity.
The AI techniques seem to course of a number of types of transaction and behavioral information, together with:
- Historical transaction data
- Spending frequency and velocity
- Merchant class exercise
- Geographic buying patterns
- Device and account info
- Real-time authorization indicators
Instead of relying solely on static fraud guidelines, machine studying fashions assess whether or not a transaction deviates from anticipated buyer conduct.
According to a NVIDIA case study on American Express’s infrastructure, the corporate makes use of GPU-accelerated AI techniques that may course of fraud choices inside milliseconds.
For clients, the workflow impression is essentially invisible until suspicious conduct is recognized. Legitimate purchases are processed by way of authorization, whereas higher-risk transactions might set off extra verification or evaluation.
For fraud operations groups, machine studying techniques assist prioritize which transactions require human investigation. This reduces the quantity of lower-risk transactions that require handbook evaluation, permitting analysts to give attention to extra complicated fraud circumstances.
The use of AI in fraud detection has been persistently mentioned in firm supplies, educational analyses, and infrastructure companion studies, suggesting that machine studying performs an operational position in American Express’ transaction safety processes.
Reducing Manual Review Workloads – Machine studying techniques assist fraud groups focus investigative assets on higher-risk transactions as a substitute of reviewing massive volumes of routine exercise.
Improving Transaction Approval Accuracy – Behavioral fashions might help distinguish authentic buy anomalies from probably fraudulent exercise throughout authorization workflows.
Agentic Commerce Infrastructure for AI-Powered Transactions
American Express can also be investing in what it describes as “agentic commerce” — a mannequin through which AI brokers can perform commerce-related duties on behalf of shoppers, together with product purchases, journey bookings, reservations, and fee transactions. Rather than specializing in buyer analytics, this initiative facilities on creating the infrastructure required for AI brokers to take part securely in industrial transactions.
The enterprise drawback stems from a problem dealing with the rising agentic AI ecosystem. Traditional fee techniques have been designed round direct human actions, the place a buyer explicitly selects merchandise, enters fee particulars, and authorizes purchases. As AI brokers turn out to be able to executing duties on behalf of customers, fee suppliers should set up mechanisms to confirm agent id, authenticate buyer intent, and keep transaction safety.
To handle this problem, American Express launched its Agentic Commerce Experiences (ACE) developer kit. According to firm statements, the platform contains 5 interconnected capabilities designed to help agent-driven transactions:
- Agent registration and verification
- Customer account enablement
- Intent validation and authentication
- Tokenized fee credential issuance
- Transaction context and authorization controls
The information processed inside these workflows differs from that of conventional fraud detection techniques. In addition to transaction and fee info, the platform evaluates buyer buy intent, authorization permissions, agent credentials, and transaction context earlier than permitting an AI agent to finish a purchase order.
For clients, the meant workflow shifts from executing each transaction manually to defining buying intent and approval parameters that verified AI brokers can act upon. American Express has said that clients will have the ability to handle spending controls, buy approvals, and energetic AI-agent permissions by way of its digital channels.