How to Innovate in the Age of Agentic AI: American Express's Approach

Autonomous AI agents are becoming a reality in the corporate world. American Express shows how global companies can stay ahead in the age of agentic AI: it takes a strategy, not spontaneous experiments.

Why Amex is looking at agentic AI

Traditional AI models require constant human intervention. Agentic AI is different: the agent makes decisions itself and carries out actions in a chain (reads data → thinks → acts), all without constant prompts. For a payments company, this means automating complex processes: transaction verification, risk management and customer service.

• The agent works in real time
• It chooses its own tools and data sources
• It learns from historical data
• It requires fewer human interventions

Four approaches from American Express

According to an Amex executive, successful innovators at the company do the following:

First: start with the pain, not the technology. Don't hunt for a sophisticated use case just to tick a box. Pick a real problem that costs the company a lot. For Amex, that was automating fraud protection and payment verification.

Second: experiment in small iterations. Don't build for years. Run sprints of 2-4 weeks, collect fast feedback, and launch quickly on real data (in a controlled environment). Amex tests agents on a subset of transactions before scaling.

Third: build into existing processes. Agentic AI should not replace everything at once. Good agentic AI complements existing workflows: it helps people make decisions faster rather than removing them entirely.

Fourth: watch the risks from day one. An autonomous agent can make a mistake that a human will not notice. You need monitoring, logging of all decisions, fast rollback and an escalation chain.

Results at Amex

The company has not disclosed specific figures, but says it has seen a noticeable reduction in transaction processing time and in fraud-protection errors. The key change is cultural: the company has started to see agentic AI not as a panacea but as a tool for specific tasks.

What this means

For large fintech companies and banks, agentic AI is not futurism but a tool for competitive advantage. Amex has shown that the winner is not whoever launches the most complex agent first, but whoever quickly ties AI to real revenue and risk. Everyone else needs to move in the same direction.