Verified
American Express
Piloting- Payments
- 76,800 employees
- $5B tech spend a year
- AI enablement layer, built in-house
- 11,000+ engineers on AI coding tools
American Express deploys AI through an enterprise AI enablement layer it built for generative and agentic tools. American Express says nearly all of its 76,800 colleagues have generative AI tools, and more than 11,000 engineers use AI coding assistants. Stephen Squeri said in Jul 2026 that AI cut coding cycle time 30% to 40%, and American Express puts that capacity into more projects rather than counting it as savings. For customers, American Express is building agentic commerce, including the ACE Developer Kit and Resy bookings inside ChatGPT and Claude.
americanexpress.comTheir approach to AI
- The platform
- American Express built an enterprise AI enablement layer, with shared starter code, model orchestration and an AI firewall, to launch generative and agentic tools such as Travel Counselor Assist and Dining Companion (Jan 2026).
- The models
- American Express uses open and closed-source models that its orchestration layer can swap by use case (Apr 2025), OpenAI models in its dining and hotel conversational experiences (Oct 2026), and GitHub Copilot for engineers.
- The reach
- American Express says it has deployed generative AI tools to nearly all colleagues worldwide (Jan 2026) but gives no usage count. More than 11,000 American Express engineers use AI-assisted development tools (Mar 2026), and 5,000 travel counselors use Travel Counselor Assist.
- The catch
- American Express's one AI return, coding cycle time down 30% to 40%, is capacity rather than savings, as Stephen Squeri said in Jul 2026. American Express framed the cut of about 5% of positions announced in Oct 2026 as preparing for AI, not as work AI already does.
Snapshot
- Employees
- ~76,800
- Dec 31, 2025, about 25,900 in the U.S. (2025 Form 10-K)
- Technology budget
- $5B a year
- Stephen Squeri, Jan 2026; a record level of technology development planned for 2026
- AI/ML headcount
- Not disclosedest.
- Several thousand of about 10,000 technologists worked on generative AI in May 2024
- Who owns AI
- Ravi Radhakrishnan, CIO
- Enterprise Data & AI led by Hilary Packer, per third-party profiles; not confirmed by American Express
- Platform
- Enterprise AI enablement layer
- Built in-house; described on the Jan 30, 2026 earnings call
- Models
- OpenAI, open and closed models
- Swapped by use case (Apr 2025); GitHub Copilot for engineers
- Reach
- Nearly all colleagues
- Generative AI tools deployed worldwide (Jan 2026); no usage count published
- Use cases
- Hundreds explored
- Mar 2026 letter; 70 in implementation out of 500 identified in Apr 2025
Key findings
American Express credits AI with one business return: Stephen Squeri said on Jul 24, 2026 that AI cut coding cycle time 30% to 40%. American Express spends that capacity on more technology projects rather than counting it as savings, and has published no firm-level AI figure.
American Express said in Jan 2026 that it had deployed generative AI tools to nearly all of its 76,800 colleagues. The Mar 2026 shareholder letter counted more than 11,000 engineering professionals on AI-assisted development tools and travel counselors in 19 countries on AI tools.
American Express took a $336 million restructuring charge on Oct 23, 2026 to eliminate about 5% of positions. CFO Christophe Le Caillec said the cut would make American Express leaner and prepare it for AI, without saying AI already does the work of the roles cut.
American Express published its ACE Developer Kit for agentic commerce and Agent Purchase Protection on Apr 14, 2026, and its Intent Intelligence framework on Oct 13, 2026. Hannah Bozian said in Oct 2026 that the American Express network sees thousands of agentic transactions a week.
American Express spends about $5 billion a year on technology (Jan 2026) and is moving its enterprise data platform and hundreds of production applications to Google Cloud BigQuery, a rollout Stephen Squeri said runs into 2027.
Where Amex sits
- Stage
- Piloting
American Express has reported one business return it credits to AI in the last 18 months, which places it in Piloting. Stephen Squeri said on Jul 24, 2026 that AI cut coding cycle time 30% to 40% and that American Express uses the capacity to work through more of its technology backlog. The Mar 2026 shareholder letter had put the cut at over 30% for more than 11,000 engineering professionals. Squeri said the coding gain is not a savings, and that the number of servicing representatives has stayed flat as the business grew, without giving a figure. CFO Christophe Le Caillec framed the Oct 2026 cut of about 5% of positions as preparing for AI rather than as work AI already does. On rollout, American Express says generative AI tools reach nearly all colleagues worldwide (Jan 2026) but has published no count of employees who use them. American Express falls short of Scaling because its one return sits in technology alone and it reports no firm-level AI figure.
How the stages are defined - Ownership
- CIO Ravi Radhakrishnan; Enterprise Data & AI under Hilary Packer (title unconfirmed)
- Platform
- Hybrid: AI enablement layer built in-house; models and coding tools bought
- Models
- OpenAI and other open and closed models; GitHub Copilot for engineers
- Hosting
- Own data centers and mainframes plus cloud; data platform moving to Google Cloud BigQuery; model hosting not disclosed
- Reach
- Generative AI tools for nearly all colleagues (Jan 2026); no usage count
How they deploy
Fraud and credit models first
American Express has used AI and machine learning since 2010, starting with credit risk and fraud detection. Stephen Squeri's Mar 2026 shareholder letter credits those models with helping keep American Express's credit and fraud metrics strong. Squeri said on Jul 24, 2026 that American Express is now adding unstructured data and agentic tools to those credit, risk and fraud decisions.
One enablement layer, swappable models
American Express builds generative AI on a shared enablement layer instead of training its own large language model. Hilary Packer, then CTO, said in Apr 2025 that the layer gives engineers common starter code, lets an orchestration layer swap open and closed-source models by use case, and wraps everything in an AI firewall. Stephen Squeri said on Jan 30, 2026 that the enterprise AI enablement layer supports the generative and agentic capabilities already in market.
Agentic commerce as the growth bet
Stephen Squeri's Mar 2026 letter calls AI agents that shop and pay for customers one of the largest changes in commerce since e-commerce. The letter sets 3 priorities: embed American Express payments in AI platforms, make Membership assets such as Resy, Amex Offers and Amex Travel bookable there, and build AI experiences in American Express's own apps. On Oct 23, 2026 Squeri named 5 areas where American Express is applying AI for customers: value propositions, acquisition, engagement, payments and servicing.
Restructure before automating
American Express announced on Oct 23, 2026 that it would eliminate about 5% of positions globally by removing management layers and widening spans of control. CFO Christophe Le Caillec told CNBC that American Express does not want to automate processes that are not well structured. Le Caillec said American Express is reorganizing first so that it can automate and use AI in the coming quarters and years.
Buy AI talent and products
American Express bought the expense management platform Center in Mar 2025. In Apr 2026 American Express agreed to buy Hypercard, a company whose AI agents categorize corporate expenses and check reports against policy, to bring agentic AI into the expense platform it is launching for business customers in 2026.
- Deployment model
- American Express runs AI as a central technology capability with use cases in each business. The Technology organization under CIO Ravi Radhakrishnan builds and runs the enterprise AI enablement layer, the coding tools and the data platforms. Third-party profiles place an Enterprise Data & AI group, combining data governance and AI engineering, under Hilary Packer. The businesses own their use cases. Servicing uses Travel Counselor Assist and a servicing portal for Card Member representatives, marketing uses AI agents for campaigns, and Global Commercial Services uses generative AI in sales. Global Merchant and Network Services, under Anna Marrs, also oversees the credit and fraud risk organization whose models score transactions (Oct 2026). A Generative AI Council of senior leaders from model building, technology, servicing and legal has prioritized use cases since 2023.
- Model providers
- American Express uses third-party models and has said it does not plan to build its own large language model (May 2024). Hilary Packer said in Apr 2025 that American Express uses open and closed-source models, swapped by use case. Packer said the internal IT chatbot had moved from the open-source BERT model to closed-source generative AI. Stephen Squeri said on Oct 23, 2026 that American Express built its proprietary dining and hotel conversational experiences on OpenAI models and is deepening its OpenAI partnership. About 9,000 American Express engineers used GitHub Copilot in Apr 2025. The Mar 2026 letter says American Express is scaling agentic coding tools, without naming them.
- Data and infrastructure
- American Express runs its own data centers and mainframes alongside cloud services, according to a Jul 2026 Goldman Sachs notice describing the infrastructure a former American Express technology executive ran. Stephen Squeri said on Jan 30, 2026 that a third-generation data and analytics platform, on the cloud, would roll out organization by organization through 2027. Squeri said the platform would support personalization, servicing, fraud and generative and agentic AI. Alphabet said in Apr 2026 that American Express is moving that enterprise data platform and hundreds of production applications to Google Cloud BigQuery. American Express has not said where its generative AI models run. American Express kept customer data out of its early generative AI pilots (May 2024) and tests tools in sandboxes with redacted customer data before deployment.
Platforms and internal tools
Enterprise AI enablement layerBy Apr 2025
The enterprise AI enablement layer is American Express's shared base for generative and agentic applications. The layer provides starter code for engineers, an orchestration layer that connects users to models and swaps them by use case, model risk validation, and an AI firewall around all of it. Stephen Squeri said on Jan 30, 2026 that the layer supports Travel Counselor Assist, Dining Companion and the generative AI tools deployed to nearly all colleagues.
- Models
- Open and closed-source models
Travel Counselor AssistPilots late 2023; all counselors by Oct 2024
Travel Counselor Assist is a generative AI tool that helps American Express travel counselors prepare for calls, build itineraries, search for deals and recommend hotels, restaurants and activities for Platinum and Centurion Card Members. The tool combines web data with American Express and customer data, and summarizes counselor calls. More than 85% of counselors said in Apr 2025 that the tool saves time and improves recommendations.
- Users
- About 5,000 travel counselors in 19 markets (Apr 2025)
AI-assisted development tools
American Express gives its engineers AI coding assistants, mainly GitHub Copilot for testing and code completion plus a feature for asking questions about code (Apr 2025). American Express is scaling agentic coding tools for the full software development cycle (Mar 2026).
- Users
- 11,000+ engineering professionals (Mar 2026); about 9,000 on GitHub Copilot (Apr 2025)
- Models
- GitHub Copilot
Internal IT support chatbotOct 2023
American Express's internal IT chatbot is its most heavily used technology support channel. The chatbot moved from BERT-based language processing to generative AI in Oct 2023 and now walks colleagues through problems step by step before handing them to a live engineer. American Express said in Apr 2025 that the chatbot had brought a 40% increase in its ability to resolve IT queries without a transfer to a live engineer, and that a similar colleague help center answered with 96% accuracy.
Servicing portal for Card Member representativesJul 2026
American Express launched a servicing portal with AI built in for its Card Member service representatives, which Stephen Squeri said on Jul 24, 2026 will reduce their handle time. Squeri said on Oct 23, 2026 that a growing suite of AI tools helps servicing colleagues find information and resolve issues, and that American Express had launched its first conversational AI voice agents.
ACE Developer Kit and Intent IntelligenceApr 14, 2026; Intent Intelligence Oct 13, 2026
The Amex Agentic Commerce Experiences (ACE) Developer Kit gives AI platforms specifications for agent registration, card enablement, purchase intent, tokenized payment credentials and cart context. Amex Agent Purchase Protection backs the kit, covering errors by registered agents. The Intent Intelligence framework added in Oct 2026 creates an intent ID and a proof-of-intent token before an agent buys, so American Express can stop a purchase that does not match what the Card Member asked for.
Third-generation data and analytics platformRolling out 2026 to 2027
American Express's third-generation data and analytics platform moves its enterprise data and hundreds of production applications to Google Cloud BigQuery and is meant to feed personalization, servicing, fraud and generative and agentic AI. Stephen Squeri called the platform American Express's third big data conversion in a decade. Squeri said in Jan 2026 that Card Member marketing was already seeing some cost benefit, without giving a figure.
Vendors and partners
OpenAIBy Mar 2026
Models behind American Express's dining and hotel conversational experiences; Resy in ChatGPT · Resy reservations went live in ChatGPT on Aug 10, 2026. American Express is working with OpenAI on an Amex Offers plugin (Oct 2026), and U.S. Business Gold and Platinum cards carry a $300 ChatGPT Business credit (Mar 2026).
AnthropicApr 2026
Resy connector in Claude · American Express connected Resy to Claude in Apr 2026 so diners can see live availability and book within a conversation.
Google CloudBy Apr 2026
BigQuery for the enterprise data platform · Alphabet said in Apr 2026 that American Express is moving its enterprise data platform and hundreds of production applications to BigQuery to support agentic commerce.
Microsoft and GitHubFeb 2023
GitHub Copilot for engineers; Microsoft Cloud and AI for expense tools · American Express and Microsoft agreed in Feb 2023 to build expense management tools on Microsoft Cloud and AI technologies, with the AI built by American Express itself.
PerplexityOct 2026
AI Skills for U.S. Business Card Members · Perplexity launched AI Skills, workflows such as cash flow forecasting and tax preparation, for U.S. American Express Business Card Members on Oct 1, 2026.
HypercardApr 2026
Acquired AI expense management company · American Express agreed in Apr 2026 to buy Hypercard, which builds AI agents that categorize and check corporate expenses; the two had launched an AI-powered card together in 2024.
Leadership
AI, data and platform leaders
Executives
AI org map
Central technology, data and AI capability under the CIO; use cases owned by the business lines
American Express places its AI platforms in the Technology organization under CIO Ravi Radhakrishnan, an executive officer since Jan 2022, and third-party profiles place an Enterprise Data & AI group under Hilary Packer inside it. American Express has no chief AI officer among its executive officers (Feb 2026). The Board's Risk Committee oversees the use of AI under an Artificial Intelligence Risk Management Policy, and a Generative AI Council of senior leaders has prioritized use cases since 2023.
American Express does not publish an AI or machine learning headcount. American Express said in May 2024 that several thousand of its roughly 10,000 technology employees worked on generative AI in some capacity and that it had hired a few hundred more. The Mar 2026 letter counts more than 11,000 engineering professionals using AI-assisted development tools. American Express had about 76,800 employees at the end of 2025.
Reporting lines
Chairman and CEO
Stephen Squeri
Technology
Ravi Radhakrishnan, CIO
Executive officer; a direct line to the CEO is inferred, not published
Enterprise Data & AI
Hilary Packer
Title and reporting line from third-party profiles, not confirmed by American Express
Technology Risk and Information Security
Chief Information Security Officer
The chief information security officer reports to the CIO (2025 Form 10-K)
Global Merchant and Network Services
Includes oversight of the credit and fraud risk organization (American Banker, Oct 2026)
Board Risk Committee
Primary board oversight of AI and generative AI risks and opportunities (2026 proxy statement)
Generative AI Council
Senior leaders from model building, technology, servicing and legal; prioritizes use cases and manages risk (since 2023)
Global Innovation
Luke Gebb
Leads the ACE Developer Kit; reporting line not published
American Express's Feb 2026 list of executive officers names a CIO but no chief AI, data or technology officer, and the company has not published who leads Enterprise Data & AI or where Berta Rodriguez-Hervas sits. Hilary Packer was CTO in Apr 2025; American Express has not said who holds that title now.
Where the AI workforce sits
Technology organization (engineering, Enterprise Data & AI)
Estimate: American Express publishes no AI headcount; in May 2024 several thousand of about 10,000 technologists worked on generative AI in some capacity.
~70-80%Business-line risk, marketing and servicing analytics
Estimate: the credit and fraud risk organization and business teams build and run models, but American Express gives no count.
~20-30%
Named people
Use cases
American Express says it has explored hundreds of AI use cases, and its Mar 2026 shareholder letter names work in customer service, marketing, technology, credit and fraud, and commercial sales. American Express has published a return it credits to AI for one use case. Stephen Squeri said in Jul 2026 that AI coding tools cut coding cycle time 30% to 40%, and American Express puts that capacity into more projects rather than counting it as savings. American Express gives no figure for its servicing, travel, marketing, fraud or sales tools and has reported no firm-level AI return. American Express framed the Oct 2026 cut of about 5% of positions as preparing for AI.
| Use case | Area | Status | ROI |
|---|---|---|---|
| AI coding and testing assistants More than 11,000 American Express engineering professionals use AI-assisted development tools, mainly GitHub Copilot, and American Express is scaling agentic coding tools across the software development cycle. Stephen Squeri said the gain goes into a large backlog of technology projects. | Technology | In production | Coding cycle time down 30% to 40%, capacity used for more projects (Jul 2026)Source |
| Travel Counselor Assist Travel Counselor Assist gives about 5,000 American Express travel counselors in 19 markets generative AI help with itineraries, deals and recommendations. A pilot cut about 60 seconds from calls (May 2024), and more than 85% of counselors said in Apr 2025 that the tool saves time. | Customer service | In production | |
| Card servicing tools and chatbot American Express card servicing teams use an AI-powered chatbot and, since Jul 2026, a servicing portal with AI built in. Stephen Squeri said on Jul 24, 2026 that the number of servicing representatives has not grown as the business grew and should fall over time through attrition, without giving a figure. | Customer service | In production | |
| AI search in the Amex app American Express uses AI to improve search in its U.S. mobile app, answering about 1 million Card Member inquiries a month (Mar 2026). | Customer service | In production | |
| Conversational AI voice agents American Express planned in Mar 2026 to pilot conversational AI agents to replace its legacy interactive voice response systems, and Stephen Squeri said on Oct 23, 2026 that it had launched its first few voice agents. | Customer service | Pilot | |
| Complaints and disputes American Express is expanding AI in complaints and disputes to find problems earlier, understand root causes and resolve them faster (Oct 2026). | Customer service | In production | |
| Campaign creation and compliance review American Express uses AI agents and other AI tools to create campaigns, develop content and personalize offers. American Express also uses AI-enabled workflows to shorten compliance review before campaigns launch, and Stephen Squeri said in Jul 2026 that campaigns now reach market faster. | Marketing | In production | |
| Fraud detection and credit underwriting models American Express has used AI and machine learning in fraud detection and credit underwriting since 2010. The credit and fraud risk organization's models evaluate transactions in real time, and American Express now adds unstructured data to its credit insights. | Credit & Fraud | In production | |
| Fraud claim classification American Express fraud teams are piloting AI that reads customers' fraud reports to classify incoming claims (Mar 2026). | Credit & Fraud | Pilot | |
| Generative AI for sales American Express uses generative AI in Global Commercial Services to trigger real-time prospect leads, research before calls, reprioritize prospect lists from call analysis and automate follow-ups. American Express is moving its sales teams to a new generative AI sales platform (Mar 2026). | Global Commercial Services | In production | |
| AI expense app and Insights Agent American Express plans an AI-powered expense app that automates receipt capture, policy checks and approvals, and an Insights Agent that analyzes business spending, as part of its 2026 commercial product roadmap with Hypercard's team. | Global Commercial Services | Announced | |
| Internal IT support chatbot American Express's generative AI IT chatbot guides colleagues through technology problems, and American Express said in Apr 2025 that it had brought a 40% increase in its ability to resolve IT queries without a transfer to a live engineer since its Oct 2023 launch. | Technology | In production | |
| Dining Companion and travel discovery Dining Companion uses Card Member spending insights to recommend restaurants, and Wishlist Builder in the Amex Travel app helps customers find hotels and trips. American Express's dining and hotel conversational experiences run on OpenAI models (Oct 2026). | Consumer products | In production | |
| ACE Developer Kit and Agent Purchase Protection American Express lets registered AI agents pay with tokenized card credentials under its ACE Developer Kit, protects Card Members from registered agent errors, and checks purchases against authenticated intent. Agentic volumes remain small, at thousands of transactions a week (Oct 2026). | Agentic commerce | In production | |
| Resy and Amex Offers on AI platforms Resy restaurant search and booking runs inside Claude (Apr 2026) and ChatGPT (Aug 2026), and American Express is working with OpenAI on a plugin to discover and enroll in Amex Offers (Oct 2026). | Agentic commerce | In production | |
| Cyber defense and platform resilience American Express uses AI to strengthen cyber defense and the resilience and availability of its critical platforms (Mar 2026). | Technology | In production |
Adoption
American Express spread AI in layers, starting with machine learning in credit and fraud in 2010. A Generative AI Council and the first generative AI pilots followed in 2023, then tools for travel counselors and engineers in 2024 and 2025, and generative AI tools for nearly all colleagues by Jan 2026. American Express's published counts cover single functions: about 5,000 travel counselors in 19 markets and about 9,000 engineers on GitHub Copilot in Apr 2025, and more than 11,000 engineering professionals on AI-assisted development tools in Mar 2026. Stephen Squeri's Mar 2026 letter says colleagues show strong engagement and widespread daily use, but American Express has not said how many of its 76,800 employees use AI tools each week.
By the numbers
Rollout, first pilot to today
2010
American Express began using AI and machine learning in credit risk analysis and fraud detection.
2023
American Express convened a Generative AI Council of senior leaders from model building, technology, servicing and legal, which identified 500 potential use cases.
Oct 2023
American Express put generative AI into its internal IT support chatbot, its most heavily used technology support channel.
May 2024
American Express said all of its U.S. travel counselors used a generative AI copilot, planned to extend it to 20 more countries by the end of 2024, and had several thousand technologists working on generative AI.
Apr 2025
Hilary Packer said about 9,000 American Express engineers used GitHub Copilot, Travel Counselor Assist served 5,000 counselors in 19 markets, and 70 generative AI use cases were in implementation.
Jan 2026
Stephen Squeri said American Express had built an enterprise AI enablement layer and deployed generative AI tools to nearly all colleagues worldwide.
Mar 2026
American Express's shareholder letter reported AI-assisted development tools for more than 11,000 engineering professionals and AI tools for travel counselors in 19 countries.
Apr 2026
American Express published the ACE Developer Kit for agentic commerce and connected Resy to Claude.
Jul 2026
American Express launched a servicing portal with AI built in for its Card Member service representatives.
Oct 2026
American Express launched its first conversational AI voice agents and published its Intent Intelligence framework for agent purchases.
Economics & controls
Spend and ROI
Technology budget
Stephen Squeri said on Jan 30, 2026 that American Express spends $5 billion a year on technology, split between running the business (infrastructure, software licenses, cybersecurity) and development such as new apps and core system modernization. Christophe Le Caillec said technology spend rose 11% in 2025. American Express planned a record level of technology development for 2026. Squeri said in Apr 2026 that strong first-quarter results let American Express lower internal return thresholds to fund more technology projects.
What AI itself costs
American Express does not disclose what it spends on AI. In May 2024 American Express declined to give its generative AI investment, saying it covered technology infrastructure, commercial software and licenses, and talent. Stephen Squeri said in Jul 2026 that agentic commerce work added investment that was not planned at the start of 2026.
Reported ROI
American Express has reported no firm-level return from AI. Stephen Squeri said on Jul 24, 2026 that few companies had committed to AI savings in their 2026 or 2027 profit and loss. On the same call, Squeri said the 30% to 40% cut in coding cycle time is not a savings because it lets American Express do more. Squeri also said AI tools have kept the number of travel and customer service representatives flat as the business grew. In Jan 2026, Squeri had said the new data platform was already contributing to lower Card Member marketing costs, without a figure or a share for AI.
Expense ratio
Christophe Le Caillec said on Jan 30, 2026 that operating expenses fell 4 points as a share of revenue from 2022 to 2025 while technology spend grew, without crediting AI. Le Caillec said on Jul 24, 2026 that AI will play a critical role in keeping that ratio falling. Stephen Squeri said on the same Jul 2026 call that American Express is further along on internal AI than on agentic commerce, and that both are early.
Headcount
American Express employed about 76,800 people at Dec 31, 2025, about 25,900 of them in the U.S. On Oct 23, 2026 American Express took a $336 million restructuring charge to eliminate about 5% of positions globally by cutting management layers and widening spans of control. The sale of its Global Business Travel Group stake partly funded the charge. Christophe Le Caillec told CNBC the cut would make American Express leaner and more agile and prepare it for AI. PYMNTS reported that Stephen Squeri described AI's productivity benefits separately, without attributing the full reduction to automation.
Governance and risk
Board oversight and AI risk policy
American Express's Board Risk Committee is primarily responsible for overseeing the risks and opportunities of the company's AI and generative AI use, including guardrails and training, according to the 2026 proxy statement. American Express has adopted an Artificial Intelligence Risk Management Policy that covers all AI systems, colleagues, products and lines of business. The policy follows a 3-lines-of-defense model, sets escalation to management or the Board, and governs third parties' use of AI.
Generative AI Council
American Express convened a Generative AI Council in 2023 to prioritize use cases, manage risk and guide responsible use, drawing senior leaders from model building, technology, servicing and legal. Ravi Radhakrishnan described the council and risk frameworks as the core of AI governance in Apr 2025.
Validation, data and human review
American Express validates generative AI through its model risk management process, including retrieval-augmented generation, and its enablement layer runs inside an AI firewall (Apr 2025). American Express kept customer data out of its early generative AI pilots (May 2024) and tests tools in sandboxes with redacted customer data. American Express keeps human oversight on automated actions and recommendations and pauses ideas with high hallucination rates.
Outside AI agents
Stephen Squeri said on Oct 23, 2026 that American Express will let outside AI agents into its systems only through approved API or Model Context Protocol (MCP) channels and will not let assistants run unchecked through its systems. Squeri said American Express will also watch for agents that game rewards or repeatedly query Resy. American Express registers agents under the ACE Developer Kit, checks purchases against authenticated intent, and covers Card Members for errors by registered agents.
Disclosed risks and regulation
American Express's 2025 Form 10-K lists risks from AI and machine learning, including biased or hallucinated outputs, data exposure, the difficulty of explaining model results and AI-enabled fraud and deepfake attacks. The 10-K names the EU AI Act and U.S. state AI laws among the rules American Express must follow. The Technology, Data, Resiliency Risk Committee, co-chaired by the CIO, oversees information security risk, and colleagues receive AI fluency and responsible use training (Mar 2026).
Recent signals
Oct 23, 2026
American Express took a $336 million restructuring charge to eliminate about 5% of positions. Stephen Squeri set out 5 areas where American Express applies AI for customers and said it had launched its first conversational AI voice agents. Squeri said outside agents may reach its systems only through approved API or Model Context Protocol channels.
Oct 23, 2026
CFO Christophe Le Caillec told CNBC that the restructuring would make American Express leaner and more agile and prepare it for AI, so that it can automate processes once they are restructured.
Oct 13, 2026
American Express published its Intent Intelligence framework, which records a Card Member's authenticated purchase intent before an AI agent buys. Hannah Bozian said the American Express network sees thousands of agentic transactions a week.
Oct 1, 2026
Perplexity launched AI Skills such as cash flow forecasting and tax preparation for U.S. American Express Business Card Members.
Oct 1, 2026
American Banker reported that Anna Marrs helps lead American Express's adoption of AI and generative AI and that the company uses AI agents in parts of credit underwriting, fraud detection and customer service.
Aug 10, 2026
Resy, owned by American Express, launched restaurant availability and booking inside ChatGPT for U.S. restaurants.
Jul 24, 2026
Stephen Squeri said AI cut coding cycle time 30% to 40% at American Express, that servicing headcount has stayed flat as the business grew, and that a new AI servicing portal will cut representatives' handle time.
Apr 27, 2026
American Express connected Resy to Anthropic's Claude so diners can check live availability and book a table in a conversation.
Apr 23, 2026
Stephen Squeri said American Express sees about a 30% benefit from AI for its programmers in coding and testing, and that on agentic commerce the company is not even in the first inning.
Apr 16, 2026
American Express agreed to buy Hypercard, whose AI agents categorize corporate expenses and check them against policy, for its 2026 expense management platform.
Apr 14, 2026
American Express published the ACE Developer Kit for agentic commerce and Amex Agent Purchase Protection, which covers Card Members for errors by registered AI agents.
Mar 25, 2026
Stephen Squeri's shareholder letter set out American Express's agentic commerce priorities and reported AI tools for nearly all colleagues, after a Citrini Research report on AI job losses had hit the stock in Feb 2026.
Mar 25, 2026
American Express launched the Graphite Business Cash card and announced AI tools for business customers, including an Insights Agent, an AI expense app and a $300 ChatGPT Business credit.
Jan 30, 2026
Stephen Squeri said American Express spends $5 billion a year on technology, is rolling out a third-generation data platform on the cloud, and has built an enterprise AI enablement layer.
Quick answers
- How does American Express deploy AI?
- American Express deploys AI through an enterprise AI enablement layer it built for generative and agentic tools. American Express says nearly all of its 76,800 colleagues have generative AI tools, and more than 11,000 engineers use AI coding assistants. Stephen Squeri said in Jul 2026 that AI cut coding cycle time 30% to 40%, and American Express puts that capacity into more projects rather than counting it as savings. For customers, American Express is building agentic commerce, including the ACE Developer Kit and Resy bookings inside ChatGPT and Claude.
- What does American Express use AI for?
- American Express says it has explored hundreds of AI use cases, and its Mar 2026 shareholder letter names work in customer service, marketing, technology, credit and fraud, and commercial sales. American Express has published a return it credits to AI for one use case. Stephen Squeri said in Jul 2026 that AI coding tools cut coding cycle time 30% to 40%, and American Express puts that capacity into more projects rather than counting it as savings. American Express gives no figure for its servicing, travel, marketing, fraud or sales tools and has reported no firm-level AI return. American Express framed the Oct 2026 cut of about 5% of positions as preparing for AI.
- Who leads AI at American Express?
- AI at American Express is led by Ravi Radhakrishnan (Executive Vice President and Chief Information Officer); Hilary Packer (Executive Vice President, Enterprise Data & AI); Berta Rodriguez-Hervas (Senior Vice President of AI). Ownership of AI on the org chart: CIO Ravi Radhakrishnan; Enterprise Data & AI under Hilary Packer (title unconfirmed).
- What AI platform does American Express use?
- American Express built an enterprise AI enablement layer, with shared starter code, model orchestration and an AI firewall, to launch generative and agentic tools such as Travel Counselor Assist and Dining Companion (Jan 2026).
- Which AI models does American Express use?
- American Express uses open and closed-source models that its orchestration layer can swap by use case (Apr 2025), OpenAI models in its dining and hotel conversational experiences (Oct 2026), and GitHub Copilot for engineers.
- How many Amex employees use AI?
- American Express says it has deployed generative AI tools to nearly all colleagues worldwide (Jan 2026) but gives no usage count. More than 11,000 American Express engineers use AI-assisted development tools (Mar 2026), and 5,000 travel counselors use Travel Counselor Assist.
- What return has American Express reported from AI?
- American Express has reported one business return it credits to AI in the last 18 months, which places it in Piloting. Stephen Squeri said on Jul 24, 2026 that AI cut coding cycle time 30% to 40% and that American Express uses the capacity to work through more of its technology backlog. The Mar 2026 shareholder letter had put the cut at over 30% for more than 11,000 engineering professionals. Squeri said the coding gain is not a savings, and that the number of servicing representatives has stayed flat as the business grew, without giving a figure. CFO Christophe Le Caillec framed the Oct 2026 cut of about 5% of positions as preparing for AI rather than as work AI already does. On rollout, American Express says generative AI tools reach nearly all colleagues worldwide (Jan 2026) but has published no count of employees who use them. American Express falls short of Scaling because its one return sits in technology alone and it reports no firm-level AI figure.
Open questions
- How many of American Express's 76,800 employees use its generative AI tools each week, given that the company reports access for nearly all colleagues but no usage count?
- Who owns AI at American Express below the CIO, given that the company has not announced Hilary Packer's move from CTO to Enterprise Data & AI or Berta Rodriguez-Hervas's appointment, and who is CTO now?
- What does American Express spend on AI, including model and token costs, inside its $5 billion technology budget?
- Which models and clouds run American Express's generative AI, beyond OpenAI models in its dining and hotel experiences and GitHub Copilot, and does any model run in its own data centers?
- Which roles does the Oct 2026 cut of about 5% of positions remove, what saving does American Express expect from it, and how much of the later automation will be AI?
- What do Travel Counselor Assist, the card servicing portal, the voice agents and the marketing agents return in cost or headcount, given that Stephen Squeri said servicing headcount has stayed flat without giving a figure?
- Has AI changed American Express's fraud losses or credit performance in a way the company can measure, given that it credits its models without a figure?
- How many agentic transactions run on the American Express network, and what has Amex Agent Purchase Protection cost so far?
- How large is the cost saving in Card Member marketing that Stephen Squeri tied to the new data platform in Jan 2026?
Sources
American Express
Filings, letters and earnings calls
- 2025 Annual Report, with Stephen Squeri's shareholder letter and Form 10-K (Mar 2026)
- 2026 proxy statement (Mar 2026)
- Q3 2026 earnings release, Form 8-K exhibit (Oct 23, 2026)
- Q3 2026 earnings call transcript, TradingKey (Oct 23, 2026)
- Q2 2026 earnings call transcript, AllMind (Jul 24, 2026)
- Q4 2025 earnings call transcript, AllMind (Jan 30, 2026)
Partners
News
- TechCrunch: Amex introduces a framework to prevent AI agents from making unauthorized purchases (Oct 13, 2026)
- Digital Transactions: Amex's next step for agentic commerce is a framework (Oct 2026)
- HR Reporter: Amex to cut 5% of workforce to prepare for AI (Oct 2026)
- The Next Web: American Express cuts 5% of staff and won't let AI agents run rogue (Oct 26, 2026)
- PYMNTS: Amex plans job cuts, sets AI rules (Oct 23, 2026)
- Unite.AI: Perplexity launches business Skills for American Express Card Members (Oct 1, 2026)
- American Banker: The Most Powerful Women in Finance, Anna Marrs (Oct 1, 2026)
- Verdict Foodservice: Amex links Resy with Claude (Apr 27, 2026)
- The National CIO Review: American Express links growth to technology (Apr 23, 2026)
- Digital Transactions: Why Amex is going full steam ahead in AI (Apr 23, 2026)
- American Banker: Amex agrees to buy more tools to feed its AI plans (Apr 16, 2026)
- American Banker: Amex introduces agentic commerce developers kit (Apr 14, 2026)
- Fortune: American Express unveils tools for AI developers (Apr 14, 2026)
- American Banker: Amex doubles down on AI for businesses (Mar 25, 2026)
- PYMNTS: Squeri says service and security will win agentic commerce (Mar 25, 2026)
- Bloomberg Línea: American Express defends its AI efforts (Mar 2026)
- CO/AI: American Express AI tool helps 5,000 travel agents (Sep 2025)
- VentureBeat: How Amex uses AI to increase efficiency (Apr 2, 2025)
- Technovation podcast, Metis Strategy: Ravi Radhakrishnan on scaling AI at American Express (Apr 28, 2025)
- American Banker: American Express unveils its approach to generative AI (May 2, 2024)
- TechCrunch: Amex and Microsoft turn to AI for expense reports (Feb 9, 2023)
- The Org: Hilary Packer, EVP and Head of Enterprise Data & AI
- AI Magazine: Top 100 Women in AI 2026, Berta Rodriguez-Hervas
Data confidence
- Confirmed
- Stated in a primary or reputable public source.
- Estimated
- Derived or approximated from public reporting.
- Unconfirmed
- Reported, but not yet corroborated.
American Express profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Nov 9, 2026