How Amex exploits new AI tools

is rolling out AI aggressively across fraud, underwriting, customer service, marketing, and internal productivity, and management now describes “hundreds” of in‑flight use cases built on that stack.

Strategic framing

• Amex has used machine learning for more than a decade (fraud, personalization, operational streamlining) and is now layering generative AI on top of that foundation.

• The company groups Gen AI work around three goals: productivity (internal tools), protection (fraud, risk, security), and growth (spend stimulation, new products, more precise targeting).

• Executives say they have “hundreds” of AI use cases identified or underway across the organization, with a test‑and‑learn approach to scaling.

Fraud, risk, and protection

• Amex runs deep‑learning fraud models (RNN/LSTM plus gradient boosting) on GPU infrastructure, scanning tens of millions of daily transactions in milliseconds to meet a 2 ms decision‑latency target.

• Its “Gen X” fraud model executes more than 1,000 decision trees on billions of observations, automating over 8 billion decisions and ingesting data from over 1 trillion dollars in annual transactions.

• Earlier ML deployments identified an estimated 2 billion dollars in potential annual incremental fraud incidents before losses, and Amex continues to refine these models on a unified data platform.

Customer service and “AI‑assisted” agents

• Gen‑AI‑based “Travel Counselor Assist” is embedded directly in the workflow for Platinum/Centurion travel agents, surfacing real‑time itinerary options, recommendations, and context from LLMs plus live data.

• A “Knowledge Assist” chatbot lets frontline colleagues query approved internal content in natural language, returning synthesized answers rather than static article lists, which Amex says shortens handle time and improves accuracy.

• Call‑center ML tracks customer sentiment and agent performance across about 12,900 agents in 19 markets, generating feedback and coaching signals while supporting AI‑drafted responses and summaries.

Personalization, marketing, and product growth

• Amex uses its closed‑loop data (cardholder, merchant, and network information) to power real‑time offer engines and personalized experiences, and it is extending this into Gen‑AI‑driven search and discovery.

• A semantic “Search Optimization” capability in the mobile app lets customers ask natural‑language questions and receive more direct answers, reducing friction in self‑service servicing.

• AI underpins real‑time personalized offers and campaigns via its new cloud‑native “Lumi” data platform, which is designed to support fraud detection, offers, and governance from the same data estate.

Internal productivity and engineering

• Amex is scaling an AI coding assistant for engineers, using LLMs to suggest code and speed product development while keeping humans in the loop.

• Gen‑AI pilots also target document search, summarization, and knowledge management so employees can navigate policies, procedures, and technical documentation faster.

• Amex Ventures is actively investing in Gen‑AI startups and using pilots with them to seed new internal tools and capabilities across the enterprise.

Governance, “human in the loop,” and limits

• Public messaging emphasizes responsible deployment, with focus on trust, security, and a continued option to reach a human, in line with Trendex survey findings that customers want AI with empathy and human backup.

• The company stresses approved content sources and human oversight for Gen‑AI servicing tools, positioning AI as colleague assist rather than full automation in customer‑facing contexts for now.

• Under the hood, Amex is consolidating onto Lumi, its next‑generation cloud‑native data platform, to centralize data, support model governance, and avoid the fragmentation that can undermine explainability and compliance.