Hopper CEO Frederic Lalonde poses outside the company’s Montreal office in 2021. Mr. Lalonde says his company’s new app resolves customer issues four times faster than human agents.Christinne Muschi/The Globe and Mail
Hopper Inc. became one of Canada’s most valuable private technology companies by using artificial intelligence to build a challenger to online travel giants Expedia EXPE-Q and Booking.com BKNG-Q.
Now the Montreal company is leaning into generative AI, launching a virtual customer service portal called HTS Assist (HTS stands for Hopper Technology Solutions) that airlines and travel providers can offer their customers.
Users can book or cancel reservations by typing or speaking into Assist, which responds in a conversational, AI-generated voice, handling arrangements and responding to questions in a fraction of the time that human agents would take. During a demonstration Monday, Jo Lai, senior vice president of AI solutions and customer experience with HTS, used Assist to cancel a flight, book a car and get advice on the best way to transfer between two London airports – in five minutes.
Hopper’s rule for innovation: kill what you started and make it better
CEO Fred Lalonde said Hopper intended “to unlock something that is not a cheap chatbot or a layer on top of ChatGPT that will hallucinate, but a fully connected, modern system that’s capable of selling, rebooking, and looks like” a generative AI platform.
The app resolves customer issues four times faster than human agents and has achieved an 88-per-cent score on customer satisfaction surveys in tests, the same as call-centre employees, Mr. Lalonde said. He added the app cuts Hopper’s cost per call by 92 per cent, to US$1.12.
Hopper has built the app to handle call surges due to disruptions such as storms. It can pick up on caller cues and upsell services. “I’ve been shocked at how quickly this technology has been evolving,” Mr. Lalonde said, referring to large language model-powered AI that has upended the technology landscape.
With its all-in approach to AI, Hopper is on the leading edge of a trend to adopt “agentic AI” tools that can make decisions, reach into other systems to conduct commerce, store and recall memories and instantly adapt to changing conditions. Agentic AI “could upend the travel industry,” said a September study from McKinsey & Co. and travel news site Skift. Booking.com is also developing a personalized agentic AI.
It’s a bet on the mounting capability of AI despite some recent studies that have raised doubts on the usefulness, limitations and investment returns of the nascent technology.
Many customer service-focused companies have cautiously adopted chatbots, using them to handle basic questions and tasks, while relying on humans to handle complicated “edge cases.” A report this year from MIT found 95 per cent of AI pilot projects had failed, while online consumer finance company Klarna recently backtracked on plans to replace its customer service agents with AI. LLMs that power generative AI apps such as ChatGPT also have a propensity to invent wrong answers.
Tech founders leaving Canada at accelerating rate, survey finds
Hopper started experimenting with generative AI after OpenAI launched ChatGPT in late 2022. Ms. Lai said Hopper heard from customers that they wanted the option “to, quote, talk to someone” during critical moments to explain what they needed “in a back-and-forth nuanced conversation,” she said. “At the end of the day it was more about resolving their issue.” But that didn’t necessarily mean they needed to speak with a human.
Her team designed a chatbot, trained mostly on OpenAI and Google LLMs, that would “completely unbake the anatomy of a conversation” and respond to a customer’s needs conversationally in a way that was “shorter and better and helped customers resolve their issues.”
Hopper first tested its chatbot on users of its eponymous consumer-facing mobile app, then with customers of some of its 30 corporate clients including Sumitomo Mitsui Banking Corp. Assist was trained on 14 million customer interactions in several languages and is “really good at navigating edge cases,” Ms. Lai said.
Hopper said it uses its own specialized language model architecture, informed by travel-related customer conversations and integrations with verified travel data sources. Rather than improvising answers, the app’s responses are limited to information gleaned from facts and policies.
Mr. Lai predicted Assist will eventually be able to respond to almost any customer enquiry, but the extent to which corporate customers use it will come down to how many calls “they’d prefer a person to handle.”
Hopper first made its mark by predicting the best times for people to book flights, even telling customers to wait, based on past price quotes gleaned from travel booking systems.
It later built a suite of lucrative insurance-like products for travellers, including the ability to freeze a price or to buy the right to cancel a flight for full refund for any reason. Hopper then focused on signing up corporate clients to power their travel booking sites with its suite of products. That business now accounts for 90 per cent of its US$500-million-plus in revenues. Hopper expects to turn its first operating profit in 2025, Mr. Lalonde said.
With a report from Joe Castaldo