Hopper is a travel company growing rapidly. Last year, it reportedly earned $700MN in revenue. That represents a 40 percent uptick from 2022.
With that accelerating growth came new customers with different preferences, like those who default to the phone when articulating their needs.
Hopper had previously prioritized customer service through chat and self-service options, as a digital-first brand. Yet, it recognized the opportunity for voice to support its expansion.
Just adding voice wasn’t enough, however. It wanted to orchestrate AI-led voice experiences that pull in live agents when best for the customer.
In doing so, Hopper worked with PolyAI to deploy a front-end GenAI agent for the channel.
Now, the brand resolves 10-15 percent of its queries via the assistant while cautiously monitoring the impact on critical customer and agent outcomes.
As more contact centers prepare to deploy GenAI Agents, CX Today reached out to Jo Lai, Head of Customer Experience at Hopper, for expert best practices. Here’s what she had to share.
1. Maintain an Accurate & Usable Knowledge Base
GenAI Agents use retrieval-augmented generation (RAG) to detect a customer’s intent and generate responses.
Unlike older bots, which struggle with varied phrasing, the RAG model allows these GenAI Agents to understand questions phrased in many different ways, much like ChatGPT.
From there, it matches the query to a predefined knowledge base article and presents an accurate, relevant response that fits the customer’s tone.
The trouble is that those knowledge bases often have gaps or include outdated articles. As such, maintaining an accurate and usable knowledge base is key.
“There’s no shortcut here,” noted Lai. “If the knowledge base isn’t kept up-to-date, you can’t expect a large language model (LLM) to fill in the gaps on its own. This is where people often mention “hallucinations,” where the model may generate incorrect information.”
Thankfully, Hopper already had a robust knowledge base. However, it leveraged PolyAI to spot opportunities to refine its content while testing its GenAI Agent.
Contact centers can also involve high-performing, experienced agents and team members with strong journalistic skills in the knowledge-creation process. That may help to further bolster knowledge base content.
2. Focus on Your Biggest Five Demand Drivers First
Hopper didn’t immediately connect the GenAI Agent to its knowledge base and let it loose.
Instead, it started by testing its top five queries, ensuring the knowledge base was complete for those, and building confidence in the tech.
In prioritizing these contacts, Hopper could reap quick, but significant efficiency gains.
“The testing took about six to eight weeks, helping us quickly gather insights without striving for perfection upfront,” added Lai.
After two months, we felt ready for a broader rollout.
Yet, even then Hopper could implement guardrails so that the GenAI Agent didn’t always offer a default response. The application could immediately detect queries that required more of a human touch and pass those directly to a human.
3. Understand the Best Channel on a Case-by-Case Basis
While a customer may reach out via voice, that is not necessarily the best channel to resolve their query. As such, Hopper is ensuring its GenAI Agent is multimodal.
For instance, if a customer calls in with a question that could be answered by an online support article, its GenAI Agent sends an SMS with a link to that article. Lai added:
Voice assistants can identify which channel is most suitable for a resolution and guide the customer there, which is what we, as service leaders and customers, really want – just an efficient way to resolve our issues.
Now, Lai and her team are working on developing that vision and exploring new ways to improve AI-led conversations.
She continued: “Imagine if a customer asks for their flight number; the assistant could not only tell them the number but also send a text with a flight tracker link and travel details.
“True omnichannel, where an orchestrator can seamlessly guide customers across channels, would be a real game-changer.”
4. Switch the Focus from Containment & Deflection
For Lai, when evaluating the success of GenAI or similar tools, it’s essential to focus on issue resolution rather than metrics like containment or deflection.
“It can sometimes feel like the goal is to prevent customers from reaching a human agent, but the real objective should be to resolve the customer’s issue effectively,” she noted.
If a voice assistant can’t resolve the issue, it should quickly and easily route the customer to an agent. The focus should always be on providing resolutions as quickly as possible.
Indeed, the real metric for GenAI Agent success should be outcomes: did the customer actually accomplish what they wanted, or did they end up on a different channel a few minutes later?
Steve Blood, a former VP & Analyst of the Gartner Sales & Customer Service Practice, recently dived deeper on this point in conversation with CX Today.
5. Create a Contextual Escalation Process
As Lai noted, Hopper focuses on providing a speedy, seamless escalation process when the GenAI Agent cannot resolve the query.
“The transcript from the voice assistant is available to the agent, so they have context and don’t start from scratch,” she continued. “Customers don’t need to repeat themselves, which is something they greatly appreciate.”
Contact centers can also anticipate beforehand which queries the GenAI Agent won’t be able to resolve and orchestrate an experience that blends AI and live agents.
So, instead of passing the customer directly to a live agent, they can use the AI to gather pertinent information for the beforehand. They may also mechanize some of the steps to a resolution. In itself, that’s a big win.
Thanks to PolyAI for facilitating this conversation with Jo Lai and Hopper. If you’re looking for more insights from CX experts, sign up to the CX Today newsletter.
