How Amex uses AI to increase efficiency: 40% fewer IT escalations, 85% travel assistance boost

America express It is a huge multinational company with about 80,000 employees, so it is always something to imagine – it always comes with a worker who fights a laptop in WiFi input or Fritz.

However, everyone knows before, especially with a conversation, it can be a nervous experience. Automated tools can offer uncertain answers or walls of the connections that the employees actually clicked “to a person” and the “to take me to a person.”

Amex infused to prepare this obsolete scenario generative ai The interior is supported by Chatbot. ChatBot now has more intuitive interactions, adapts for feedback and walks users through step-by-step problems.

As a result, AMEX significantly reduced the number of employees who need to grow in a live engineer. AI is increasingly able to solve problems.

“Unlike the list of links to people, Hilary Packer told Amex EU and CTO insturebeat. “As we return to work quickly, the productivity is better.”

Evaluation and accuracy ‘holy grail’

IT ChatBot is just one of the many AI successes of AMEX. The company has no opportunity: in fact, a special parliament first identified 500 potential use of 500 potential in the case, potential use of up to 70 in various stages of up to 70.

“From the beginning, we wanted to facilitate the establishment of our teams Gen AI solutions and explained the “package” to fit.

This is delivered through a basic guarantee layer that provides “general recipes” or startup code, which engineers can follow the engineers to ensure sequence throughout the apps. Orchestrate layers connect users with models and allow models to change the models based on use. “AI Firewall” covers all this.

While not entering the features, Packer used the open and closed source models and a large model used the risk management and verification process, including the accuracy of the expanded generation (dwarf) and other emergency engineering techniques. The accuracy is criticized in an adjustable industry, and the basic information should be so far, so the team consumes the company’s knowledge bases, protecting the company’s knowledge bases, according to the source of the best data.

“Verification and accuracy is currently a generative AI’s holy grail,” Packer said.

AI reduces an increase in 40%

Built-in IT chatBot – AMEX’s most severely used technological support function – was a natural early use.

Initially equipped with traditional natural language handling (NLP) models – specifically combines a closed source Gen AI to provide more interactive and personal assistance – specifically.

The package explained that instead of just proposing a list of knowledge base, ChatBot involves working with questions that follow users, clarify their issues and provides step-by-step solutions. It can create generalized and appropriate response in a clear and short format. If the employee still does not receive the necessary answers, AI can increase unresolved problems to live engineers.

For example, when an employee’s connection problems, ChatBot can offer a few problems to return them to WiFi. As Packer explains, “can be interactive with his colleague and” did it solve your problem? “If they say no, you can continue and give them other solutions.”

Since the beginning of October 2023, AMEX saw an increase of 40% in the ability to resolve their surveys without transferring the live engineer. “We are always our colleagues in their way, very quickly,” said Packer.

85% of travel consultants report efficiency with AI

Amex has 5,000 travel consultants that help firmly repair the routes for the company’s most elite map (black) card and platinum card members. This is the most highest level of customers firm and wait for a certain customer service and support. Thus, the consultants must be as knowledgeable as possible about a certain place.

“Travel consultants are stretched in a large number of different areas,” Packer said. For example, a customer can ask about the visit sites in Barcelona, ​​the next asks about the five-star restaurants in Buenos Aires. “It’s not trying to keep all this in the head of someone, isn’t it?”

AIU Agent, which helps to treat individual travel recommendations, “Assistant Travel Consultant Assistant” to optimize the process. Thus, tools, tools, owners are combined with AMEX data and customer information and combined with customer information and combined with customer information, can draw information about the highest levels (such as cardholders) from the Internet (such as the highest levels). Packer said it helps create a single, accurate, timely look.

The AI’s wife now supports 5,000 travel consultants in 19 of AMEX’s market, and more than 85% of them have stored the instrument and improve the quality of the instrument and improves the quality of the tool. “So it’s really a really productive tool,” Packer said.

Although it can take a total of the EU, the package stressed the importance of keeping people in the loop: The information obtained by the EU is combined with travel consultants and institutional knowledge to make special recommendations reflecting the interests of customers.

Because during this technology managed, customers want a person’s recommendations that can provide customers, context and compatibility – not only a common route drawn based on the main search. “Want to know what you talked to someone who will think about the best vacation for you,” Packer said.

EU advanced colleague assistant, coding companion

Among the other dozens of use, Amex applied the EU for “Colleague Help Center” – this was 96% of the rate similar to Chatbot; Developed search optimization, which returns the result, based on the intention of the words more sought than literal words, causes 26% of answers; And AI coding assistants increased the productivity of developers by 10%.

The AMEX’s 9,000 engine now uses GitHub Copilot, mainly for testing and code complement. The packager explained that developers have a talk-to-code feature that allows you to ask the code about the code. Finally, the company wants to expand the latest software development life cycle (SDLC) and API documents until the end.

It should be noted that the Packer said that more than 85% of the codes, the company said the company’s approach to the AI ​​AI.

“Not only does not work, but when a colleague interacts with him, do they like?” Said Packer. “We have a few pilots we say we can achieve the conclusion, but we do not have great colleague satisfaction. Do we want to continue this? Is this really the right result for us?”