Contact center AI providers are making many promises right now.
In a way, that’s excellent, as customer service leaders need a vision.
That vision can help them to allay pressure from execs, understand possible labor savings, and cultivate real interest in helping customers.
Yet, implementing AI always brings expected and unexpected challenges.
Not all of these are the tech provider’s fault. Sometimes, it’s the organization’s internal data systems, decision-making processes, or outdated tools.
Such issues can create a gap between the original promise of AI and the reality.
Tripadvisor is no stranger to this dichotomy. Yet, it’s now achieving meaningful results, including 70 percent containment and improved employee engagement.
How has it achieved that? Tim Barker, Director of Business Operations, AI & Automation at Tripadvisor, explained: “We put in real effort: good knowledge bases, ongoing tuning, and actual work. No magical AI that tunes itself – yet!”
Speaking at Customer Contact Week 2025, Barker shared many more golden nuggets of advice for implementing contact center AI. Here are ten excellent examples.
1. Think of AI in Two Buckets
When implementing contact center AI, Tripadvisor has three core focuses: boosting customer satisfaction, reducing agent pain, and validating ROI.
It looks through each lens as it considers new AI use cases, which it splits into two buckets: customer-facing AI and agent-facing AI.
Customer-facing AI can be scary. Yet, Tripadvisor has found success not only in reducing pressure on service staff but – in some instances – producing better responses.
For Barker, agent-facing AI is trickier, as many use cases focus on supporting agents in crafting responses, not the true pain points in the employee experience. He said:
Our agents didn’t need help writing responses; they had templates. The issue was the prep, the research. Writing a great email takes time. We had to find ways to support that part, not just throw in a co-pilot.
Ultimately, for Tripadvisor, the key was to position AI to help agents do what they’re already great at faster and more efficiently.
2. Understand What Makes for a Successful Pilot
Barker believes that the success of a contact center AI pilot comes down to the targeted use case and the rigor the team brings to the process. He said:
When I talk to peers or sister brands, the main difference between projects that succeed and those that stall is how deep people go in understanding what they’re solving for, and whether they treat AI as “set it and forget it” or as a tool that needs real attention.
3. Push Past the Flashy Demos
Contact centers need to get past flashy demos and ask: does it work in my world? Can it handle my use cases? Can I compare solutions clearly and see measurable results?
After making this point, Barker admitted that it’s easy to become lured in by the impressive capabilities of large language models (LLMs). After all, it’s simple to see they’re great at question-answering and have real potential, especially in service scenarios and multilingual support.
However, contact centers must consider the human impact, how it fits into their tech stack, and the change management process (more on all this below).
4. Vet Your Vendors Carefully
According to Barker, the best AI vendors are those that make an effort to understand each business, its jargon, and its teams.
He also suggests that contact centers should seek providers that customize their solutions to fit unique needs, not those that present a one-size-fits-all solution.
Finally, he suggests picking a provider that’s transparent about its LLM choices.
5. Validate Your ROI Properly
Proof of Concepts (PoCs) are essential to validate ROI. “Even if it’s not perfect, you’ll learn a ton,” stated Barker.
If you have done a PoC, get really clear about the success criteria, internally and with the vendor, so everyone agrees on what “success” looks like.
Consider a conversation automation use case. To validate ROI here, start by measuring volume by question type, knowing the average handle time (AHT) per type, and identifying where AI can take over efficiently. This allows contact center leaders to prioritize high-impact use cases.
From there, Barker suggests asking to prove value early with free or low-cost POCs.
“See if their solutions work in high-impact areas,” he said. “Don’t expect them to solve everything; focus on the top use cases first.”
6. Don’t Overlook Robotic Process Automation (RPA)
Summarizations and notation tools are often touted as the easy wins for contact center AI. Yet, Barker also advocates for an older technology: RPA.
RPA helps automate repetitive work. “These pay off quickly, especially if you understand your cost structure,” said Barker.
We have about 17–18 web-based tools our agents use. They jump between them constantly, five to nine clicks per interaction—just to get the information they need. With RPA, many of those steps can happen behind the scenes, saving time and effort.
Indeed, instead of manually checking the status in a separate tool, automation can do that and guide agents appropriately.
7. Expand Next Best Action to Edge Cases
When the contact center automates simpler contacts, agents receive tougher queries, increasing their cognitive load. Barker noted:
Think of it like your computer when you try to open a massive Excel file, everything slows down or crashes. That’s what contact centers risk doing to their agents every day.
To counteract this, Tripadvisor has invested in next best action technology, not only for the simple stuff but the edge cases where the tech can pay real dividends.
“While agents know the high-volume flows well, it’s those rare edge cases where memory lapses and errors creep in,” continued Barker. “AI can guide agents in real time, helping ensure compliance and reducing mistakes.”
8. Be Wary of Overlapping Providers
Contact centers often utilize multiple AI tools. As such, Barker asks them to consider: are you creating synergy or wasting money?
“Trying to compare vendors is tough; it’s not apples-to-apples; it’s a fruit salad,” he said. “That’s okay, but be clear on who owns what part of the journey.”
Here, the Tripadvisor man also stresses that businesses must ensure vendor role clarity, stressing that some will hand over the technology and walk away.
As such, be crystal clear on who’s responsible for what, and sell that internally to get the right resourcing.
9. Respect the Change Management Process
Change management is huge, and Barker admitted to making mistakes with this previously. “I’ve done it poorly,” he said. “The first time, I kept everything quiet and launched AI under wraps. It freaked people out.
Now, I involve the team, ask for feedback, and walk them through it. It makes a big difference.
10. Sit on Your AI Results for a While
Another key lesson for Barket is not to share initial performance metrics too early. Let results stabilize over a few weeks before reporting out.
If positive, he suggests going back to the vendor and asking: “What would it take to get even better?”
“Don’t settle,” he summarized. “Tech evolves, and so should your expectations.”
Want regular updates from respected CX industry practitioners? Subscribe to the CX Today Newsletter.
