Artificial Intelligence at American Express

American Express is an American financial services company headquartered in New York City, New York. The company employs more than 76,800 people globally and reported fourth-quarter 2025 revenue of $18.98 billion, with fiscal 2026 revenue growth guidance of 9-10 percent.

American Express began applying machine learning to fraud detection in 2010, making it one of the earliest adopters of AI among financial services companies. Today, the company has provided access to leading AI tools to nearly all of its 76,800 colleagues globally and has scaled AI-assisted development tools to more than 11,000 engineering professionals, reducing coding cycle times by over 30 percent.

Its machine-learning-powered fraud detection model monitors more than $1.2 trillion in transaction value every year, generating a fraud decision in milliseconds for every card transaction worldwide. The company is currently exploring more than 70 generative AI use cases across the organization and, through Amex Ventures, is investing in generative AI startups focused on trust and safety, enterprise efficiency, and data-driven experiences.

For financial institutions operating at a global transaction scale, AI systems are increasingly influencing how fraud risk, transaction approvals, and customer retention workflows are managed operationally. In the case of American Express, many of these systems are embedded directly into high-volume decision-making processes rather than positioned as standalone AI initiatives.

Publicly available reporting and company documentation point to two particularly visible applications of AI within the organization:

Real-time fraud detection during transaction authorization

Predictive customer retention and merchant targeting

In both cases, machine learning systems examine large volumes of behavioral and transactional data to support faster operational decisions, reduce manual review workloads and improve customer experience outcomes.

Real-Time Fraud Detection During Transaction Authorization

Fraud prevention is one of the most visible AI applications inside American Express. The company has discussed using machine learning systems to evaluate transactions in real time during payment authorization workflows.

According to a Harvard Business School Digital Initiative analysis, American Express applies machine learning models to analyze purchasing patterns, spending behavior, merchant activity, and transaction anomalies during authorization requests.

The business problem is significant for payment providers. Fraudulent transactions cause direct financial losses, while incorrectly declining legitimate purchases can erode customer trust and transaction volume.

The AI systems appear to process multiple forms of transaction and behavioral data, including:

Historical transaction records

Spending frequency and velocity

Merchant category activity

Geographic purchasing patterns

Device and account information

Real-time authorization signals

Instead of relying solely on static fraud rules, machine learning models assess whether a transaction deviates from expected customer behavior.

According to a NVIDIA case study on American Express’s infrastructure, the company uses GPU-accelerated AI systems that can process fraud decisions within milliseconds.

For customers, the workflow impact is largely invisible unless suspicious behavior is identified. Legitimate purchases are processed through authorization, while higher-risk transactions may trigger additional verification or review.

For fraud operations teams, machine learning systems help prioritize which transactions require human investigation. This reduces the volume of lower-risk transactions that require manual review, allowing analysts to focus on more complex fraud cases.

The use of AI in fraud detection has been consistently discussed in company materials, academic analyses, and infrastructure partner reports, suggesting that machine learning plays an operational role in American Express’ transaction security processes.

Reducing Manual Review Workloads – Machine learning systems help fraud teams focus investigative resources on higher-risk transactions instead of reviewing large volumes of routine activity.

Improving Transaction Approval Accuracy – Behavioral models can help distinguish legitimate purchase anomalies from potentially fraudulent activity during authorization workflows.

Agentic Commerce Infrastructure for AI-Powered Transactions

American Express is also investing in what it describes as “agentic commerce” — a model in which AI agents can perform commerce-related tasks on behalf of customers, including product purchases, travel bookings, reservations, and payment transactions. Rather than focusing on customer analytics, this initiative centers on creating the infrastructure required for AI agents to participate securely in commercial transactions.

The business problem stems from a challenge facing the emerging agentic AI ecosystem. Traditional payment systems were designed around direct human actions, where a customer explicitly selects products, enters payment details, and authorizes purchases. As AI agents become capable of executing tasks on behalf of users, payment providers must establish mechanisms to verify agent identity, authenticate customer intent, and maintain transaction security.

To address this challenge, American Express introduced its Agentic Commerce Experiences (ACE) developer kit. According to company statements, the platform includes five interconnected capabilities designed to support agent-driven transactions:

Agent registration and verification

Customer account enablement

Intent validation and authentication

Tokenized payment credential issuance

Transaction context and authorization controls

​

The data processed within these workflows differs from that of traditional fraud detection systems. In addition to transaction and payment information, the platform evaluates customer purchase intent, authorization permissions, agent credentials, and transaction context before allowing an AI agent to complete a purchase.

For customers, the intended workflow shifts from executing every transaction manually to defining purchasing intent and approval parameters that verified AI agents can act upon. American Express has stated that customers will be able to manage spending controls, purchase approvals, and active AI-agent permissions through its digital channels.