Case Study: TRIPADVISOR

Tripadvisor Group is a global travel platform spanning hotels, restaurants, cruises, and tours and experiences, with approximately 2,500 contact center agents handling post-booking customer support across the group’s portfolio. With a January 1, 2026 contract effective date and a summer peak deadline that could not move, Tripadvisor set out to deploy a production-grade AI voice agent in a single travel season. An in-house team of eight, supported by NiCE Professional Services and NiCE partner Bell and building on the existing NiCE CXone platform with NiCE AI Agents (Cognigy), took the system from concept to first live calls in two and a half months. The agent, branded Vesper, captures intent and verifies identity in natural language before a human ever picks up, processes queries in parallel against multiple data sources, and hands off to a human agent with full context preserved when needed. Peer-tested against a comparable build at a leading global technology platform, Tripadvisor’s first iteration was assessed as exceeding what a 45-person team had reached in eighteen months. The deployment is already returning a customer sentiment score of 90 percent against the 71 percent recorded when human agents handle the same interaction types.

After deploying our first iteration, two and a half months from a standing start, we had an external peer review of our Cognigy Agent. The feedback was that what our team of 8 had achieved was beyond what they’d managed in three iterations, over 18 months, with a team of 45.

David Fox

Senior Director of Global Telecom Services
Tripadvisor Group

01 Before

A seasonal operation running on a manual sequence

Tripadvisor is a portfolio of travel brands. Hotels. Restaurants. Cruises. Tours and experiences. Viator, the group’s tours and experiences marketplace, now accounts for close to half of group revenue. Behind every booking sits a contact center of approximately 2,500 agents: tenured onshore colleagues and offshore business process outsourcers who flex up and down with the seasons.

The seasonal rhythm is everything. The vast majority of post-booking contact lands in the summer months. Miss the summer peak and the business case waits a year. The deadline is set by the calendar, not by leadership.

The call types arriving at volume were not simple. A traveler pulling into Barbados the next morning wants to know where the glass-bottom boat will collect them. The agent must consult the supplier’s website. Attempt a live call. Wait for a callback. Return to the customer. All in real time. All while the traveler is already outside their comfort zone.

Across the operation, every call opened with the same two-step manual sequence. Intent capture happened agent-by-agent. Identity verification happened agent-by-agent. Two minutes consumed before the actual conversation began. Multiplied across 2,500 agents and an entire summer peak, the cost in agent time, system fragmentation, and customer patience was substantial.

The platform to change this already existed. Tripadvisor had been running on NiCE CXone as the operating core of its global telecoms ecosystem. What was needed was the will, the design, and the deadline.

02 Opportunity to change

Start with the business problem, the technology comes second

The decision to build did not begin with AI. Tripadvisor had built a discipline around a simple test: every technology initiative either generates more revenue or reduces cost. If it cannot be connected to one of those two outcomes, it is a hobby. David Fox, Senior Director of Global Telecom Services at Tripadvisor Group, applied that discipline to the conversational AI evaluation from the first conversation.

If your only objective is to have AI and hope good things will happen, they probably won’t and you’ll wind up with a really bad investment. Everything starts with: this is what we’re trying to achieve.

David Fox

Senior Director of Global Telecom Services
Tripadvisor Group

When the team began evaluating, the question was not whether to explore conversational AI. The discipline was in filtering it. The team identified specific call types where agents were adding limited value: information retrieval, identity verification, logistics lookups. Calls where the work was sequential and the answer was already in a database somewhere, if you could only reach all of them at once.

Tripadvisor issued an RFP and tested the market. NiCE AI Agents (Cognigy) was selected. The decision came down to two factors in equal measure: the quality of the conversational AI, and the architecture of keeping everything in platform. Adding a third-party AI layer would have meant latency across already lengthy call legs, a new vendor to manage, and data residency questions across a global operation. NiCE sat over the existing network. The AI lived where the call already lived.

The contract was effective January 1, 2026. The summer peak was in June. The clock was already ticking.

03 NiCE solution

Eight people, two and a half months, first live calls inside a single season

The build ran two parallel tracks from day one. The first was infrastructure: standing up the agentic capabilities and integrating the platform into Tripadvisor’s global telecoms ecosystem. The second was conversational design: mapping exactly what a customer might say, every branch the conversation could take, every data source required to answer it, and every failure mode that needed handling before a single live call was attempted.

The second track was the more work intensive. Tripadvisor’s operating principle was that eighty percent of the work in a deployment of this kind is conversational design. The technology, once correctly integrated, will do what it is built to do. Getting the conversation right is where projects of this kind land or collapse.

The core build team numbered eight: four telecoms engineers, two back-office systems engineers, and two product managers.

Tripadvisor’s in-house team led the build, with NiCE Professional Services and NiCE partner Bell providing targeted delivery support across the engagement. “I’ve jokingly called it my baby,” Fox said. “But the more I look at it, the more I realize it’s actually a little bit like bringing up a child. You’ve got to guide it in its every step. You’ve got to correct it when it goes wrong.”

The agent that already knows what you need

Tripadvisor’s branded AI voice agent, Vesper, is built on NiCE AI Agents (Cognigy) and integrated into the group’s global telecoms platform. When a customer calls, Vesper greets them in natural language. No menu. No keypress. It captures their intent, verifies their identity securely before any account data is disclosed, and attempts to resolve the query in full. For calls it cannot contain, it hands off to a human agent with full context already transferred. The agent does not ask the customer to start again.

Even on calls Vesper does not contain, the impact is immediate. The two-minute opening sequence of every call, intent capture and identity verification, has been removed from agent workload across the system. The processes still happen for the customer; Cognigy now handles them. They no longer consume the two minutes of agent time they used to. Vesper also processes queries in parallel. A human agent can have one screen open at a time. Vesper can simultaneously query the supplier’s site, check the booking system, cross-reference logistics data, and surface the answer in the time it would take a person to load the first tab.

The iteration that made it real

The first live automated calls went out in two and a half months. What followed was a rigorous, data-driven iteration cycle the team had designed for from the outset.

One discovery came quickly. In the US market, a significant proportion of customers, conditioned by years of legacy IVR experiences, respond to any automated greeting with “agent” before the system has finished its first sentence. The build team, largely UK and Europe based, had not fully anticipated the scale of this behavior. Rather than treating it as a failure, the team refined the opening message to explain what Vesper could do. They added a bridging step: when a customer requests an agent, Vesper confirms a few details first, so the handoff goes directly to the right person rather than the next available one. The customer still gets a human. They get there faster.

The bridging step is also where Tripadvisor’s posture on agentic AI separates from the rest of the category. While Vesper is preparing the handoff, it is also capturing intent, gathering analytical insight on what brought the customer to call, and verifying identity. The customer gets the human conversation they asked for. Tripadvisor gets an enhanced view of demand and the ability to automate a handful of process steps inside the same interaction. The agentic layer is not about deflecting customers away from humans. It is about putting the business in a position to learn from every call it takes.

Every call is now transcribed and available for analysis. That data is surfacing insights the team can now act on systematically: error patterns on the website driving avoidable contacts, supplier behaviors generating complaint volumes, navigation friction sending customers to the phone. For the first time, every pattern, every friction point, every avoidable contact is visible inside one analytical layer.

After deploying our first iteration, two and a half months from a standing start, we had an external peer review of our Cognigy Agent. The feedback was that what our team of 8 had achieved was beyond what they’d managed in three iterations, over 18 months, with a team of 45.

David Fox

Senior Director of Global Telecom Services
Tripadvisor Group

04 Results

Faster than anyone had done it, better than the humans on day one

Tripadvisor peer-tested its first iteration with a team that had built a comparable AI contact center solution at a leading global technology platform. That team numbered 45 and took 18 months across three iterations. Tripadvisor’s first iteration, delivered by a team of eight in two and a half months, was assessed by those peers as exceeding the quality of their third. That pace was not accidental. It was the product of disciplined scoping, a platform integration that avoided the complexity of a multi-vendor architecture, and a team small enough to move without drag.

On customer sentiment, the AI is already outperforming human agents. Across early-deployment data, Vesper left customers feeling positive about the interaction 90 percent of the time on a measured sentiment score. Human agents handling the same interaction types scored 71 percent. Intent is captured on every call in both modes; the difference is in how the customer felt when the call ended. The AI is not approximating human performance on that measure. It is leading it.

Vesper launched at the start of 2026. The most compelling data, full call resolution rates and A/B results measured at scale, arrives through May and June as new use cases come online and the summer peak builds. That is not a caveat. It is the nature of a deployment designed to prove itself against the hardest test a travel business faces: the season when everything happens at once. Tripadvisor projects that 20 to 30 percent of post-booking call volume will flow through Vesper by mid-summer, with 75 percent containment anticipated on that subset of calls as the platform scales. The proof is being built in real time.

The A/B testing framework is live. Every new use case runs against the previous handling method before adoption. The team is measuring everything: error patterns, supplier behaviors, call drivers that could be eliminated upstream. The information that was impossible to act on at scale before Vesper is now arriving in full. Every call is data. The data is changing the business.

It’s a little bit like painting the Golden Gate Bridge. The day you stop doing it is the day you start gently sliding into obsolescence.

David Fox

Senior Director of Global Telecom Services
Tripadvisor Group

05 Future

The summer is the proof. What comes after is the scale.

The Q2 roadmap adds two capabilities: proactive supplier outreach, where Vesper initiates outbound contact with operators to resolve logistics queries before the customer calls in, and real-time translation, extending the platform across language barriers that currently route calls to specialist agent pools. Both extend the same principle: remove the human from the parts of the conversation where they add little value, so the human can be fully present for the parts where they matter most. Proactive outreach means a customer may never need to call at all.

The architecture Tripadvisor built is designed for horizontal scale. Vesper is not a single-brand, single-language, single-channel deployment. It was built from the first line of code to run across the Tripadvisor Group portfolio, across languages, and across voice and non-voice channels. The infrastructure decision, keeping everything in platform, was not just about latency on day one. It was about not having to rebuild the foundations when the scope expands.

New use cases are already being evaluated not only for volume impact but for the subtler efficiency gains. “Don’t forget to take the wins where you can find them,” Fox said. “If you do something in a couple of weeks and it will address nearly all of 5% of your conversations, rather than something that can do half of 20% in six months, get those small things going in parallel to the preparation for going after the big fish.”

It all starts with the conversation, the customer. You have to start with a business problem. Then 80% of your work is what we’ve called conversational design. Get that right and you can make the rest happen. But have the technology in an orderly manner so that you can layer that conversational design on something that doesn’t hit roadblocks through your technology stack.

David Fox

Senior Director of Global Telecom Services
Tripadvisor Group