What Makes an AI Trustworthy When Asked to Change a Reservation?

A typhoon delays a flight, disrupting plans after arrival. A restaurant reservation needs to be moved, and a ride from the airport needs to be rearranged. Twilio CEO Khozema Shipchandler experienced just such a situation while returning to San Francisco from Tokyo. He said it would be helpful to have an AI agent handle those changes for him.

Contacting each provider to rearrange plans is certainly a hassle. Tell an AI what you need, and it could contact the restaurant and transportation provider to coordinate the changes. The convenience is easy to imagine. But there are also questions: Does the AI actually have permission to change the reservations? Has it communicated your wishes correctly?

Shipchandler emphasized more than how intelligent AI is. Whether a human or an AI handles customer service, he said, what matters is earning the other party’s trust.

The Other Party in a Conversation May Be an AI#

Shipchandler explained that Twilio’s platform has supported critical communications between businesses and their customers, among others, for 17 years. It provides cloud APIs that let companies integrate capabilities such as phone calls, SMS, video calls, and chat into their applications.

AI is taking that infrastructure into a new phase. In customer service, the party on the other end of a conversation is no longer necessarily human. If AI agents are going to answer questions and carry out tasks, they need infrastructure to support those customer interactions. There also needs to be a way to verify who an AI represents and what actions it is authorized to take.

Shipchandler cited an IDC forecast suggesting that the number of AI agents within enterprises could reach one billion by 2029. It remains a forecast, but the figure illustrates the potential scale of a future in which more AI agents handle customer service.

Acting on Someone’s Behalf Requires Checks#

Having restaurant and car reservations changed to accommodate a delayed flight is an easy-to-understand use case for AI agents. In practice, however, delegating the task may require more than simply saying, “Change my reservations.”

For a birthday dinner, for example, the AI would need to establish how late the reservation could be moved and whether the party size and restaurant preferences remain the same. For the ride, a change in arrival time does not necessarily mean it is acceptable to switch providers or pickup locations. The question is how much of the person’s intent the AI should clarify before contacting the providers.

For trustworthy customer service, users should be able to verify at least the following:

• What the AI is authorized to do
• How it has understood the request
• Whom it will contact about the changes and what the revised terms will be
• Whether the process is complete or still pending

These are practical checks suggested by the example, rather than requirements taken directly from Twilio’s remarks. For procedures such as reservation changes that may involve cancellations or charges, it makes sense to define the scope of the AI’s actions in advance. One option is to let it gather information, then ask the user for confirmation before finalizing anything.

Intelligence Also Requires an Understanding of Context#

Shipchandler reflected that, since the arrival of generative AI, technologists have focused on making AI itself smarter. He also stressed the importance of AI understanding the context of a situation and connecting closely with people.

In customer service, being able to provide an answer is not enough. Each time a delayed flight’s arrival time changes, a reservation that has already been adjusted may need to be moved again. Simply offering available time slots, without considering that the user is looking forward to a birthday dinner, can hardly be called a response that meets their wishes.

Understanding context involves more than choosing thoughtful words. It also means considering the purpose of a request and the consequences of a change, clarifying uncertainties, and clearly distinguishing what has been completed from what remains undone. Whether there is a point of contact to correct mistakes also affects how comfortable customers feel delegating a task.

Trust Can Be Assessed Throughout the Service Process#

“Whether human or AI, earning trust matters.” That is the central message of Shipchandler’s remarks discussed here. For companies considering AI adoption, trust is not an abstract slogan. It is something to assess throughout the process, from the initial inquiry to completion.

Before deployment, for example, it helps to select a few common requests and define how much the AI should handle in each case:

• Providing information and suggesting options
• Changing reservations after the user’s identity has been verified
• Handling procedures that change fees or cancellation terms
• Dealing with exceptions the AI cannot resolve on its own

The further down this list a task falls, the more carefully user confirmation and handoffs to human staff need to be designed. Once operations begin, companies should record misunderstandings of requests, stalled procedures, and the number of cases requiring a handoff to a staff member. Specific evaluation methods will vary by business, but tracking only the volume of requests handled can make it easy to overlook situations in which users struggled.

The remarks discussed here do not establish how Twilio implements such verification procedures. Companies considering deployment should separately check which actions AI can perform in their customer service operations, how users’ identities are verified, and how errors are addressed.

Reservation changes prompted by a delayed flight clearly illustrate where AI agents can help. At the same time, convenience can turn into anxiety if it remains unclear what was changed and under whose authority. Start by choosing one inquiry handled within your company, then write down what the AI can handle and when a human should check its actions.

Source: https://japan.zdnet.com/article/35253598/