A capable AI agent can answer routine questions quickly and still leave a customer worse off.
Nick Mehta, entrepreneur in residence at Bessemer Venture Partners and former CEO of Gainsight, encountered that gap when he called an airline about a complex flight change. An automated agent spent about 10 minutes collecting his account details before transferring him to a person who requested the same information and a fresh explanation of the problem. A human representative ultimately told him the change could be made in the airline’s app.
“They just Frankenstein all this together,” Mehta said.
Mehta described the experience during Newsweek’s July 30 webinar, “Humanizing AI: Personalization, Engagement and the Customer Experience,” presented by Cognizant and moderated by Gabriel Snyder, Newsweek’s executive editor, enterprise.
He was joined by Arun Chandra, chief operating officer at NiCE; Jon Hyman, co-founder and chief technology officer at Braze; and Adrian McDermott, chief technology officer at Zendesk.
Each part of the airline’s service had something to offer: The app worked, the human representative was helpful and the automated agent could likely resolve common requests. The failure came in the handoffs: The conversation did not follow Mehta from one channel to the next.
The agent could not compensate for the gaps between channels. Context has to travel across them, employees need tools they will use and success must be measured through business and customer outcomes.
McDermott said companies often optimize an AI agent for the most common requests without examining what happens when a problem falls outside that range.
“You got to look at the whole journey,” he said.
McDermott compared the economics of improved service to Jevons paradox, a principle describing how efficiency can reshape demand.
“The faster, more efficient and more enjoyable you make service, the more people will use it,” he said.
Greater efficiency, in other words, does not necessarily reduce demand for service. It can raise expectations and draw out requests that customers previously abandoned because getting an answer required too much effort.
Hyman described personalization as a means of creating relevance and meaningful interactions. Even basic context—such as a recent purchase, an order number or a previous conversation—can help an automated system or employee understand what someone needs without making that person start over.
AI also gives companies the ability to respond to individual comments and actions with a level of detail that would have been difficult to provide manually. Hyman described a Braze customer using a large language model to interpret a product review and generate a tailored follow-up.
The same technology can produce messages that feel robotic or insincere. Hyman recalled receiving an email so clearly written by an AI system that he was unsure whether the other person was participating in the conversation.
“You really got to make sure that you’re buying goodwill with a customer,” he said.
Chandra added another requirement: People should have a say in when and how a company engages with them. Personalization can become intrusive when a business tracks behavior and delivers offers without understanding whether the recipient wants them.
Sales and service channels also need access to the same information. Without integrated data, Chandra said, neither an automated agent nor a person will be able to provide a favorable experience.
Drawing on his four years leading customer experience at Disney, Chandra said the company evaluated service through a broader business lens.
“For Disney, every interaction is a value-creating opportunity,” he said.
Treating each interaction as a, code output, cases resolved by AI and average handle time can show activity without revealing whether the technology improved the relationship or helped the business
Data, knowledge and context, Chandra said, ultimately depend on whether employees adopt the tools and workflows surrounding them.
“The binding factor across all of this is change management,” he said.
Employees need to be brought into the redesign as companies change how work moves through the organization. Applying AI to a process built in 2012 or 2015, Chandra warned, will reproduce outdated outcomes.
He recommended taking high-value use cases one at a time, redesigning each workflow in a contained environment and determining where AI can add value. Leaders can then identify what the process requires instead of attempting to overhaul the entire organization at once.
Hyman called for experiments that compare AI-supported work with a company’s current performance. A temporary increase in email opens, conversions or productivity means little unless it lasts and contributes to the measurements the business ultimately cares about.
Higher output can be especially deceptive. An engineering team may produce more code without generating more revenue. A marketing group may send more messages without improving retention, engagement or order value.
McDermott made a similar case for moving beyond transactional measures.
“All of this code we’re generating, is it actually generating any more revenue?” he asked. “Are customers more engaged, are they more satisfied?”
Redesigning service will also change the work assigned to people. McDermott said routine ticket triage and other basic support work are likely to become heavily automated, but he distinguished the tasks employees perform from the broader reason their roles exist.
“When that task goes away, that doesn’t mean the purpose goes away,” he said.
Mehta said broader, more generalist customer service roles could give employees responsibility for a wider range of problems and reduce the number of times customers are transferred. Chandra pointed to relationship-building and empathy as areas where people will continue to play an important role, while predicting substantial demand for re-skilling.
A human handoff should carry the interaction forward, not reset it. The employee should receive the history, context and tools needed to solve the problem. Otherwise, the customer is left to manage the company’s internal boundaries.
AI gives businesses new ways to make routine help nearly instantaneous, anticipate some needs and extend high-touch attention to people who previously could not receive it. Those capabilities create value only when the experience holds together and the organization looks beyond the quickest savings.
“Before, I think the idea was you can either deliver a great experience, like high-end service, or you can save money,” Mehta said. “And the truth is with AI, you can do both. But if you’re short-sighted, you’re going to get neither.”
Newsweek’s next AI webinar in this series will examine how established companies can move AI beyond promising pilots and into the systems, workflows and operating models that run the business.
“Overcoming Barriers to AI Transformation in Legacy Industries,” scheduled for Wednesday, Oct. 21, and presented by Cognizant, will be moderated by Gabriel Snyder, Newsweek’s executive editor, enterprise. The live discussion will explore how leaders can clarify ownership, meet governance and risk requirements and align technology investments with people, processes and measurable outcomes without disrupting critical operations.
Request Reprint & Licensing
View Editorial & AI Guidelines
