It's been nearly a year since the travel industry was hit hard by ChatGPT and generative artificial intelligence.
By late February last year, three months after OpenAI announced its natural language processing tools, Trip.com had created a chatbot within an app built on OpenAI's API, and a month later Expedia Group and Kayak became the first travel companies to create a plugin that integrates with ChatGPT. . Very quickly, it seemed like everyone was marketing new tools developed using this technology.
To learn more about how travel companies' use of technology has changed over the past year, PhocusWire reached out to some of the early adopters in hopes that their lessons may be helpful to others. I'm trying to take it. We start with a few startups that started using generative AI early on.
For some, it was important to hone what they created to get more out of the tool, both in terms of productivity and depth of insight. For others, it was important to learn what works best and what doesn't, and focus on the former.
And in at least one case, the purpose was to find out more about a traveler than you could ever hope to find out when they casually asked a being who couldn't judge them anything. For them, it might be worth remembering. Content shared with a chatbot may not always remain with the chatbot.
Check back for more articles in the PhocusWire series about early adopters of generative AI in the travel industry and the lessons they share in the coming weeks.
Magpie: Dig deeper into your analytics with GenAI
Magpie, a content management system for tour and activity providers, is built on ChatGPT's application programming interfaces (APIs) designed to help tour and activity suppliers create marketing content optimized for online search. It was last February that we released a new tool.
Christian Watts, Magpie's founder and CEO, said the app was well-received by his customers, who are not professional writers and sometimes struggle with effective product descriptions.
Subscribe to our newsletter below
“We haven’t stopped yet. [the past year] “We just tweaked the initial product,” Watts said, but those “tweaks” included adding translations into more than 80 languages and adding human responsibility for vetting the product. Given the level of data analysis involved, which would have been unrealistic, this was quite an understatement. material.
“We're really focused on 1707286059 About reviews. I think a review is appropriate for this matter. “There’s so much data in the review,” he said. “Hopefully someone would sit down and read those reviews… We are human and sometimes the messages from so many reviews get lost. , or from the last three months, to create a summary and find some of the issues that are actually occurring.”
He gave the example of a tour company bus driver named Joe who consistently receives bad reviews. If one customer complains directly to the travel company, another complains to TripAdvisor, and a third complains to Google, the problem may go unrecognized for a long time.
“We don't triangulate, so we don't know until three months later that Joe the driver shouldn't be driving the bus for some reason.” [the different review sources]'' Watts said. “But now we can just throw them all in.” [Magpie’s tool] and bring out such tendencies. ”
Looking to the future, Watts looks forward to digging deeper into analytics and building insights. The platform already includes client product descriptions, frequently asked questions, a list of different websites, and reviews from all sources. Connecting that information creates opportunities for great analysis.
“It wasn’t about that, it was about the insight,” Watts said. “So you can find some things that might not be mentioned in the product description. Maybe we'll give everyone a free ice cream at the end, and that's what people talk about most in the reviews.” So let's mention free ice cream at the beginning of the explanation. This may be a silly example, but it's the insight to spot something like that that can make a big difference. .”
Turneo: Using GenAI where it's most powerful
For Matija Marijan, CEO and co-founder of Turneo, a year with generative AI has proven to be a “huge learning curve.”
One of PhocusWire's Hot 25 Travel Startups of 2024, Turneo is an e-commerce platform for hotels and other travel brands looking to create experiences for their guests. One of our first experiments with generative AI was using ChatGPT to create a travel chatbot that acts as a virtual concierge, providing hotel guests with recommendations for local experiences they can book. .
“Early on, both we and our clients were surprised by the initial results of our generative AI products,” Marijan says. “But once the novelty factor wore off, it all boiled down to a simple question: Can this AI feature solve the problem I have, and do it in a better way overall?”
The answer wasn't necessarily yes. While adding products designed to simplify the lives of experience organizers and resellers, the company is eliminating chatbots and saying that “AI will seamlessly replace the tedious tasks you previously had to deal with.” Now processed.” The impact on productivity was significant, Marijan says.
“What we learned as we added products to the product and built new products on top of it is that certain tasks, like creating itineraries with precise schedules, are very difficult for GPT,” he says. said. “Since then, we've evolved these products to use GPT where GPT is good. [such as] It can write text, but relies on other forms of AI or human experts for things that generative AI isn't good at. ”
For example, Turneo now encourages hotel concierges to put together itineraries for guests. AI speeds up writing and provides recommendations, but human authors provide hyper-personalized service.
“We still strongly believe in GenAI,” Marijan said. “We've been using it for a while and now we can choose our deployment locations more wisely to get the best results.”
D3x (formerly Akin): Never stop learning
Over a year ago, D3x co-founder and CEO Jason Noronha and his team discovered that they could use ChatGPT to turn guest messages into API calls, and they started experimenting with it.
“It was shocking. That wasn't possible before,” he recalled.
Noronha speaks from experience. Having previously co-founded India's first Backpacker Hostel and then co-founded his chain of smaller hostels based in his home country, he moved to New York to study at Columbia University. When I needed to develop a remote management system. Communication with guests proved to be one of his biggest challenges.
His new company (formerly known as Akin after rebranding to D3x) is using ChatGPT-4 to create a personalized service that can respond to customer emails and reviews for its hospitality clients. We created a multilingual AI concierge.
When we saw what ChatGPT could do, we thought, “This changes everything.”
Jason Noronha – D3x
“When I saw what ChatGPT could do, I thought, 'This changes everything.' This changes all the previous systems that we had always built using workflows.”
The particular path taken to process data through a workflow request can get caught up in a human conversation. Previously, if someone sent a message requesting an airport transfer and asked about the weather, toothpaste, or staying an extra night, the system would sometimes stall during the trip to the airport.
ChatGPT's natural language processing understands what guests want and accesses the right information. Even better, approximately 60% of queries are of the simple type that can be answered by a chatbot, so this system can help free up facility employees and provide a human touch when needed. Ta.
So what has changed after using this system for a year?
“We know the numbers,” Noronha said. “This was an idea and a vision that we were pitching, and now he's actually implemented it in probably 200 properties and has some very large clients as well. I can tell you the gender.”
The software also continues to improve. By studying how human operators edit generated AI responses, you can train your models to provide better responses. “Right now he's hitting the 90s in terms of duplication rate. That's good validation that the response is useful.”
Twelve months in, not all of the lessons learned from using generative AI are about software quality. Some deal more with the human condition.
One strange thing they noticed was that people felt more free about what they said when they realized they were talking to a chatbot asking questions that humans might be embarrassed to ask. In one instance, the person had arrived early, but records revealed that he did not want to pay to book an extra night. Once the asker realized the conversation was with a chatbot, he asked for more details. What was the reception seat like? Does anyone care if he sleeps in the lobby?
Eventually, the AI told the questioner that he might consider visiting the on-site movie theater and taking a nap there.
“It was really weird just watching that interaction,” Noronha said.
It gets better.
Finally, the traveler asked if the chatbot could share photos of movie theater seats. There were no photos in the system — at least not at the time.
And that was a new lesson. Soon they will be adding all the photos they can to their database. This is not to help guests find a comfortable place without paying extra, but in anticipation of new questions that may arise.
