Hopper is simply a recreation institution increasing rapidly. Last year, it reportedly earned $700MN successful revenue. That represents a 40 percent uptick from 2022.
With that accelerating maturation came caller customers pinch different preferences, for illustration those who default to nan telephone erstwhile articulating their needs.
Hopper had antecedently prioritized customer work done chat and self-service options, arsenic a digital-first brand. Yet, it recognized nan opportunity for sound to support its description .
Just adding sound wasn’t enough, however. It wanted to orchestrate AI-led sound experiences that propulsion successful unrecorded agents erstwhile champion for nan customer.
In doing so, Hopper worked pinch PolyAI to deploy a front-end GenAI supplier for nan channel.
Now, nan marque resolves 10-15 percent of its queries via nan adjunct while cautiously monitoring nan effect connected captious customer and supplier outcomes.
As much interaction centers hole to deploy GenAI Agents, CX Today reached retired to Jo Lai, Head of Customer Experience astatine Hopper, for master champion practices. Here’s what she had to share.
1. Maintain an Accurate & Usable Knowledge Base
GenAI Agents usage retrieval-augmented procreation (RAG) to observe a customer’s intent and make responses.
Unlike older bots, which struggle pinch varied phrasing, nan RAG exemplary allows these GenAI Agents to understand questions phrased successful galore different ways, overmuch for illustration ChatGPT.
From there, it matches nan query to a predefined knowledge guidelines article and presents an accurate, applicable consequence that fits nan customer’s tone.
The problem is that those knowledge bases often person gaps aliases see outdated articles. As such, maintaining an meticulous and usable knowledge guidelines is key.
“There’s nary shortcut here,” noted Lai. “If nan knowledge guidelines isn’t kept up-to-date, you can’t expect a ample connection exemplary (LLM) to capable successful nan gaps connected its own. This is wherever group often mention “hallucinations,” wherever nan exemplary whitethorn make incorrect information.”
Thankfully, Hopper already had a robust knowledge base. However, it leveraged PolyAI to spot opportunities to refine its contented while testing its GenAI Agent.
Contact centers tin besides impact high-performing, knowledgeable agents and squad members pinch beardown journalistic skills successful nan knowledge-creation process. That whitethorn thief to further bolster knowledge guidelines content.
2. Focus connected Your Biggest Five Demand Drivers First
Hopper didn’t instantly link nan GenAI Agent to its knowledge guidelines and fto it loose.
Instead, it started by testing its apical 5 queries, ensuring nan knowledge guidelines was complete for those, and building assurance successful nan tech.
In prioritizing these contacts, Hopper could reap quick, but important ratio gains.
“The testing took astir six to 8 weeks, helping america quickly stitchery insights without striving for perfection upfront,” added Lai.
After 2 months, we felt fresh for a broader rollout.
Yet, moreover past Hopper could instrumentality guardrails truthful that nan GenAI Agent didn’t ever connection a default response. The exertion could instantly observe queries that required much of a quality touch and walk those straight to a human.
3. Understand nan Best Channel connected a Case-by-Case Basis
While a customer whitethorn scope retired via voice, that is not needfully nan champion transmission to resoluteness their query. As such, Hopper is ensuring its GenAI Agent is multimodal.
For instance, if a customer calls successful pinch a mobility that could beryllium answered by an online support article, its GenAI Agent sends an SMS pinch a nexus to that article. Lai added:
Voice assistants tin place which transmission is astir suitable for a solution and guideline nan customer there, which is what we, arsenic work leaders and customers, really want – conscionable an businesslike measurement to resoluteness our issues.
Now, Lai and her squad are moving connected processing that imagination and exploring caller ways to amended AI-led conversations.
She continued: “Imagine if a customer asks for their formation number; nan adjunct could not only show them nan number but besides nonstop a matter pinch a formation locator nexus and recreation details.
“True omnichannel, wherever an orchestrator tin seamlessly guideline customers crossed channels, would beryllium a existent game-changer.”
4. Switch nan Focus from Containment & Deflection
For Lai, erstwhile evaluating nan occurrence of GenAI aliases akin tools, it’s basal to attraction connected rumor solution alternatively than metrics for illustration containment aliases deflection.
“It tin sometimes consciousness for illustration nan extremity is to forestall customers from reaching a quality agent, but nan existent nonsubjective should beryllium to resoluteness nan customer’s rumor effectively,” she noted.
If a sound adjunct can’t resoluteness nan issue, it should quickly and easy way nan customer to an agent. The attraction should ever beryllium connected providing resolutions arsenic quickly arsenic possible.
Indeed, nan existent metric for GenAI Agent occurrence should beryllium outcomes: did nan customer really execute what they wanted, aliases did they extremity up connected a different transmission a fewer minutes later?
Steve Blood, a erstwhile VP & Analyst of nan Gartner Sales & Customer Service Practice, precocious dived deeper connected this point successful speech pinch CX Today.
5. Create a Contextual Escalation Process
As Lai noted, Hopper focuses connected providing a speedy, seamless escalation process erstwhile nan GenAI Agent cannot resoluteness nan query.
“The transcript from nan sound adjunct is disposable to nan agent, truthful they person discourse and don’t commencement from scratch,” she continued. “Customers don’t request to repetition themselves, which is thing they greatly appreciate.”
Contact centers tin besides expect beforehand which queries nan GenAI Agent won’t beryllium capable to resoluteness and orchestrate an acquisition that blends AI and unrecorded agents.
So, alternatively of passing nan customer straight to a unrecorded agent, they tin usage nan AI to stitchery pertinent accusation for nan beforehand. They whitethorn besides mechanize immoderate of nan steps to a resolution. In itself, that’s a large win.
Thanks to PolyAI for facilitating this speech pinch Jo Lai and Hopper. If you’re looking for much insights from CX experts, sign up to nan CX Today newsletter.