“In travel, it’s not just people asking random questions to have fun with this thing, they are actively trying to gather information.”
The announcement of the multimodal GPT-4 by OpenAI and Microsoft this week has sent every tech industry scrambling to unlock all its capabilities, and the travel space is no exception. In the four months since its release, ChatGPT had been adopted at breakneck speed by travel and hospitality players – using its groundbreaking language processing abilities to develop new tools to engage with clients, consumers, and travellers.
One such tool is MyRealTrip’s AI Trip Planner, an interactive platform that allows users to converse with the ChatGPT-powered AI to plan itineraries and receive recommendations for tours and activities. Like many adopters of ChatGPT, the AI Trip Planner wasn’t in MyRealTrip’s pipeline, but with ChatGPT catching on like wildfire, the South Korean company couldn’t resist the urge to implement the AI engine in a new service.
And, they built this new service in 2 days.
That’s a ridiculously short time to build an app, but in the case of ChatGPT, it’s not be all that surprising.
In recent weeks, we’ve quoted a number of industry experts and key players who have vouched for the speed at which ChatGPT can be implemented. The tech’s highly accessible nature is what makes it a viable tool not just for large companies, but small startups as well. This was echoed at PhocusWire’s LinkedIn audio event regarding the adoption of ChatGPT when Paul English (Co-founder, Kayak) said, “I absolutely think the GPT’s API will allow small startups with just a handful of engineers to build compelling apps on top of it.”
Still, developing any sort of offering to consumers requires the channeling of vital resources in the form of people, time, and talent. However, there is something that can make the rush seem worthwhile – a sense of fun.
WiT caught up with Jonathan Chung (Jonny), Chief Experience Officer, MyRealTrip, to get his thoughts on ChatGPT, the anticipation of GPT-4, and how the company built a fully functional AI planner in under 48 hours.
Q: What gap in the market were you addressing by adopting ChatGPT and how did you know it was the right move?
Now that travel is back, people are travelling a lot, and they’re thinking of what to do in new cities and oftentimes they have to gather that information from search engines like Google, but a lot of Koreans use Naver.
A lot of the information on Naver is now outdated, so we were thinking of having this AI planner to help plan these trips.What we added on top of that were maps and locations and some pictures. So, it’s not just chatting but it’s giving [users] a little more information that’s tailored for travel.
Q: I’m guessing the AI Trip Planner wasn’t in the pipeline before ChatGPT “broke the internet”. How did you know this was worth pursuing and rejigging the pipeline for?
We talk a lot about how, when the whole industry moved from PC to mobile 10 years ago, the companies that caught on that wave of the changes in technology are really doing well. We really think AI will be the next ‘thing’ in travel. We do believe it solves a lot of different use cases, not just the planning… but translations and [other elements] regarding communications.
We were thinking a lot about AI before, and when ChatGPT came out we knew we had to do this right away. And when this came out we were like, let’s do this in 2 days.
Q: Every company has a finite amount of resources, so how do you decide to channel talent, time, and effort to this new project while still maintaining productivity all around?
When Covid hit, the company had to really build our muscles in adjusting to the market. That’s when we pivoted to domestic travels. When the market came back, the company was already moving really fast to adapt to these changes.
A little more background here – it’s actually not the C-suite that came up with the idea of the AI Trip Planner. It was one of the data scientists, Chulmin Lee (Lead Data Analyst), who was browsing and thought, this is so cool, we should do it. He mopped up some code by himself and demoed it and everybody thought it was cool, too.
I helped to host a hackathon. We said let’s have fun. We all have work, but this is so cool, we should do it, so let’s timebox it to 2 days. Whatever we come up with at the end of the second day, let’s just post and see how that goes. We started Thursday morning, on Friday we uploaded, and we’re like… wow, that was fun!
Q: What did you learn from that initial release, once people started interacting with the AI and you could see it in a real-world scenario?
Initially, the speed at which the responses were coming was slow. I think it was similar for anyone who was using ChatGPT at the time. So after rolling it out, we were trying to improve the performance and quality. We had a few preset questions that we knew were going to return better results and answers, and we cached some of them so it wouldn’t take so long for people to get the first few initial answers to their questions.
And that was critical because a lot of the [users] have heard of ChatGPT but they haven’t used it as much so it was really important for them to just try something to see what comes back. And from that data, we noticed they were typing in some free text or natural language to search for something relevant to the trip that they were planning.
Secondly, we framed it as an AI Trip Planner, but people were asking all sorts of questions. Initially we were wondering if we should block them, or should we put them on a guardrail – but then we said no, that ruins the magic of AI. We should let them [ask away], and if it’s travel-related we should provide richer feedback. Those were some things we learned right after we launched the product.
Q: Were there any major challenges when trying to implement ChatGPT into a new product like the AI Trip Planner, or any challenges with development at all?
For ChatGPT… it was new, but it was built in a way that was easy to intergrate. One [obstacle] was the English – Korean translation that had to happen, and obviously the quality of the answers were better in English, but because we had to serve in Korean, that was a bit hard to figure out in the beginning.
Other than that, we were just trying to find the most scrappy ways to get this out and see how people react. We were thinking about the user experience, actually. The tech was [fairly easy] to integrate.
Q: Studies show that ChatGPT is being adopted at a faster rate than even the biggest social media platforms like TikTok and Instagram. Do you think this will flood the market with too many AI products, and maybe create too much competition?
Correct, even after we launched, we saw 6 – 10 other services that launched [their own AI products] right away. People are actively looking at others, actively looking at use cases. I think this will really be a big mark in this industry. We’ve seen Blockchain and so many new things come out, but when it comes to end-user adoption and how we see people using it, I think [ChatGPT] is the most realistic way of thinking about what’s next.
And because it’s so clever – it has its own intelligence – we really do think that it’ll work in travel. Let’s say I use Google Translate in Japan or I can’t communicate in English, I have to speak in a certain way to know that it’ll translate the right way. If I have nuanced sentences, I know it’s not going to translate [well].
But because of ChatGPT’s intelligence, and we know the voice tech is coming out, that synergy that it would create between all of the new technologies – I think it’ll be big.
It’ll also get rid of the market opportunities as well. We know that there was a startup [working on] a trip planner, and it was the core of their product… and they sold it right away because they knew they were not going to be able to compete here. So, I think it’s an opportunity but also for the things that we thought would work in the future… [ChatGPT] could replace those entirely so yes, we do think that this is a big market.
Q: What would you say are some limitations of the AI tool?
Obviously, it’s not as stable as it could be. We’re waiting for GPT-4 or the next version to come out soon. Sometimes it doesn’t have the right data – the data point stops at 2021 so it’s not current – but all of those things are what I would expect from something like this. So, limitations-wise I didn’t see too much.
Some people are fearful of this, I could see why that is because of the speed at which it improves by itself… but other than the two or three things I mentioned, I don’t see that much of a limitation.
Editor’s note: GPT-4 had not been released at the time of this interview. However, the new model is now available to the general public via ChatGPT Plus, OpenAI’s $20 monthly ChatGPT subscription.
Q: With ChatGPT going strong and GPT-4 on the horizon, what is next for MyRealTrip and the AI Travel Planner?
Right now, we’re doing usability studies and user interviews to see how people are reacting to it. One thing that’s different from what we expected before launching is… we thought people would find this fun and entertaining, but people are actually looking for high-quality answers.
We were thinking people would ask lightweight questions, but people are asking very detailed questions like ‘I’m going to send my parents to Seoul by themselves, they’ll stay for 3 days, I need to know what restaurants they can go to’… and that’s why in travel, it’s not just people asking random questions to have fun with this thing, people are actively trying to gather information.
We’re trying to integrate this into the search engine within our product as well. And then there’s customer service. Sometimes, demand [at MyRealTrip] is so high that it’s just unrealistic for us to take all the calls and figure it out in under a minute. So, I definitely think that integrating [ChatGPT] into our customer service will be another step that we’ll be actively discussing.
