Listen on your favorite app
 |
 |
 |
 |
 |
 |
Welcome to Travel Buddy
In this episode of the Travel Buddy Podcast, Brandon Giella, Rachel Satow, and Ian Andersen welcome back Ravneet Ghuman, Head of Data Science and Machine Learning at Switchfly. Ravneet explains how Switchfly integrates AI—especially generative AI—into travel and loyalty platforms to improve productivity, accelerate model deployment, and enhance customer experiences. Key features include natural language hotel search, neighborhood insights for city stays, and transparent recommendations to build user trust while maintaining privacy compliance (GDPR, CCPA).
They discuss measurable business impacts, such as increased searches, engagement, and conversions, citing features like similar hotels and neighborhood-based recommendations. The conversation covers AI vs. machine learning distinctions, challenges with data availability, energy use, and the future balance between AGI ambitions and practical efficiency. Ravneet emphasizes gradual, user-friendly feature integration to drive adoption while streamlining the travel planning process. Looking ahead, Switchfly will focus on operationalizing proven experiments, boosting productivity, and delivering business value through AI-driven enhancements.
Key Highlights
- Ravneet explains Switchfly’s approach to generative AI and machine learning, emphasizing how these technologies drive internal productivity and efficiency, accelerate labeling and automation, and support engineers with practical applications in daily workflows.
- The episode spotlights Switchfly’s rigorous commitment to privacy and transparency, including GDPR and CCPA compliance, user behavior analyzed only in aggregate, and features like AI-powered explainable recommendations for destinations and hotels.
- Following the launch of Neighborhood Insights and supporting models, Switchfly observed a 4% increase in user search activity and a 1.6% increase in total conversions on destination pages, demonstrating the tangible business value of machine learning-driven personalization.
- The discussion covers Switchfly’s philosophy of incremental enhancement: maintaining a familiar interface while gradually introducing new, user-friendly features such as natural language search—designed to keep users engaged and lower friction at every stage.
- Broader industry topics are addressed, including the evolution of AI infrastructure, data limitations, energy considerations, and responsible innovation, with context for artificial general intelligence (AGI) and the continued need for explainability and compliance.
- Ravneet highlights Switchfly’s ongoing cycle of experimentation and measured rollout, using A/B testing and operational focus to ensure new AI features steadily deliver value to both travelers and partners.
Transcript
Brandon Giella (00:01.538)
Hello and welcome back to another episode of the Travel Buddy Podcast presented by SwitchFly. I have with me as always the wonderful Rachel Satow and Ian Anderson. Welcome back to the show. And we also have a special guest back for his second time, Ravneet Guman, who is basically head of AI at SwitchFly. Your title is so illustrious that you're the head of tech. What is your title actually?
Ravneet Ghuman (00:30.04)
It's
Brandon Giella (00:30.146)
How do you describe your role? Because I want to make sure that listeners understand that you know all the things that there is to know about tech and AI, loyalty programs at Switchfly and all that. You're leading engineering teams in that regard. So tell us a little bit about what you guys are doing at Switchfly.
Ravneet Ghuman (00:46.808)
First of all, Brandon, thanks for having me again. Pleasure to be here. My title is Head of Data Science and Machine Learning. So essentially, I kind of hit two things. One is try to drive data-driven decisions across the company. And the other side is AI or more machine learning, wherein I try to help with improving efficiencies or operations side of things using
Brandon Giella (00:54.68)
Same thing.
Ravneet Ghuman (01:15.764)
using machine learning as well as customer facing site, more personalizing the whole shopping experience to the degree possible.
Brandon Giella (01:27.214)
Amazing, amazing.
Brandon Giella (01:32.536)
So when it comes to AI that...
Ian Andersen (01:32.69)
It's also an extremely understated explanation.
Brandon Giella (01:38.104)
That's what I'm getting at. Yes. Exactly. Yeah. So when it comes to AI, are, I mean, you are one that is literally thinking at a very macro level and a very micro level where we're seeing these trends or seeing things develop across the way that companies are using AI, particularly in the travel industry, particularly for loyalty programs.
And then you're actually implementing those models and those teams and building product to serve those kinds of trends. So that's what I want to get across because generative AI has been a huge topic trend for years. It's a buzzword, you essentially everybody's an AI company at this point, but you guys at Switchfire actually doing pretty amazing things with, with AI and the way that you're rolling that into the product and the customer experience. And so since.
AI is such a buzzword, I wanted to get really practical. What are the ways that you guys are actually incorporating AI into your product? How's that impacting customers? How's that impacting travel loyalty programs in general? And then we'll talk a little bit like future state and trends. So first question, Rob Neat. What are the ways that you guys are actually using AI at Switchfly?
in real tangible ways where you're bringing in these LLMs into the product, into the customer journey. What does that look like? What are kind of things are you guys developing there?
Ravneet Ghuman (03:13.258)
Sure. I think ever since LLMs became very popular, my take has been there are three ways in which GEN.AI helps. One is productivity improvements, which applies to every single or almost every single person who uses a computer. second way is more, second and third are more
time to market. So if I were to build a traditional or machine learning model using traditional approach, the life cycle is find the data, label it, build a model, deploy it. With Gen.ai, that labeling exercise can be done very quickly. so labeling and automation, the time to market, that has sped up quite a bit.
The last use case I like to think of JNI is more prompting or prompt engineering, where in the model itself is the large language models itself are so great, you can give it some specific instructions. And sometimes those instructions could be 50 to 100 lines long or maybe even longer, but really complex instructions and it can do specific tasks for you. Or you can...
fine tune a large language model to solve a specific problem for you. So at a high level, three things. I mean, of course, when I talk about the first one, productivity improvement, if I were to think from engineering point of view, it helps us write code faster. It helps us automate things like, say, unit test cases.
If I were a finance person, it can help me do some data analysis and forecasting. I could upload it small Excel and ask it questions on predicting future trends or identify some gaps in the data. One of my favorite ones on productivity is almost having a PhD level expert available to you that you can ask any questions and
Brandon Giella (05:39.266)
Hmm.
Ravneet Ghuman(05:42.304)
I like to almost brainstorm ideas and try to use that specialist lens to examine ideas from different perspectives. But I think overall, I think productivity improvement is valid across any industry, not just travel. But those are high level use cases where we use JNAA everyday.
Brandon Giella (05:53.208)
Hmm.
Brandon Giella (06:15.118)
Awesome. That's helpful.
Ian Andersen (06:16.146)
So Brandon something I was thinking about when we were getting ready for this Those I went back and kind of looked at the the podcast. We did a year ago last July with revenue and Yeah, yeah
Brandon Giella (06:28.526)
This is episode 13, by the way, last July, it's called AI and Machine Learning and Travel. So go take a listen to that.
Ian Andersen (06:35.89)
Right. And it's really fascinating to see, like, I think at the time, you and I, especially, we were still struggling to get our heads around what exactly, you know, this kind of new AI.
world is like and what it can be used for. Revneet I think had clearly a much firmer hold on what it's been and I think over the last year what's really dawned on me is that I was so much in a...
thinking about it from the user perspective, from the end user perspective, what can I go do in chat GPT or Claude or whatever, where I didn't fully grasp what the ramifications on the sort of enterprise side of things. And I think that's where Revneet has been.
really focused on is that before it even gets to the end user, there's so many different touch points that AI can kind of help speed things up and sharpen the accuracy and productivity. So is that fair, Reveneet? Am I starting to finally grasp where you might have been 10 years ago with this understanding?
Brandon Giella (07:43.348)
Mm-hmm. Mm-hmm.
Ravneet Ghuman (07:59.768)
I think fairly accurate. I think when we look at the whole business life cycle, there are just so many pieces that go together. What Geni doesn't solve yet is knowing where those pieces are and building that data integration pipeline. So that's still a manual task. Hopefully in maybe not near future, but some state in future.
those large language models would be smart enough to know how to find the right information. yeah, I mean, starting from operations all the way to user experience, there is still that dirty data problem that needs to be solved, finding the right data set, giving it to the right model and surfacing it to the user.
Ian Andersen (08:55.474)
Because it really is just garbage in garbage out, right? mean, if...
Ravneet Ghuman (09:00.118)
Yeah, definitely. I mean, over the past year or two models have definitely become faster. They have become more accurate. They've become cheaper, fewer hallucinations. But some of the problems when we talk about production grade applications, we don't want someone to think they're booking a certain hotel.
but they're actually booking a hotel across the street. the accuracy, for sure. So accuracy is our importance of a production grade application. think user impact is paramount and that's what drives what features we feel confident about rolling out to users.
Brandon Giella (09:32.75)
God be the worst.
Brandon Giella (09:51.438)
So what I'm hearing is AI is not going to immediately eliminate all jobs for white collar workers. that fairly accurate?
Ravneet Ghuman (09:58.744)
I do think it is a few steps away until...
Ian Andersen (10:01.084)
Knock on wood,
Brandon Giella (10:04.632)
Yeah, okay.
Ian Andersen (10:07.922)
Um, we talked a lot last time too about, and you've, you've helped us out with, you know, we've written some, some blog articles and some other, other stuff on like user data within the system and privacy and, and transparency of how we're, we're using AI. Um, one, one question I've had and, Rachel, I'm sorry, I'm totally monopolizing all this, but one question I've had is, uh,
Regulation sort of globally still seems to be an issue, right? Everybody's sort of struggling with where this is going and how do we put guardrails around it? And how are you thinking about the privacy piece and the transparency and that aspect of it when so much of that is still up in the air?
Ravneet Ghuman (11:01.25)
For sure, I think at the heart of all we build at Switchfly is I try to put privacy and explainability first as in any feature we build. So for example, we have this feature called AI destination recommender where a person can say, me cities that have a beach or cities known for culture and history.
When we make a recommendation, we give an explanation of why we are recommending something. If it's the first ranked city, why is that better than the second or third? So we provide explanations. I mean, explainability goes beyond gen.ai, so any recommendation to a user, even things like we have this algorithm that finds hotel deals, when we...
preference a deal, we explain it to the user. This is a price drop that has happened for the same check-in checkout dates in the last 30 days. This is the lowest price, which is why we are recommending something. So that explanation goes a long way in building trust and having people keep coming back to the platform. As we have seen through data, about 20 % of our users for certain clients are repeat users. So we want...
people to have trust in the system in not just the data, but a system that values their privacy and preferences in an inclusive way.
Rachel Satow (12:42.643)
kind of on that note, you
From a marketer standpoint, we think about things like GDPR, CAN spam, et cetera, all the time. And I think when we are talking more about the software side of things and the engineering applications there, we have to be cognizant that there's a very fine line between being helpful and being creepy. And for us, when we think about that user first interface or that user first mentality, that means being like super
context aware and to Ravneet's point, there's, you know, being very transparent and explaining why certain things are being served or being, you know, very upfront from a marketing perspective from Ian and my side, we need to make sure that we are being open with sharing how we are using Gen.ai or how certain features function.
We want to ensure and instill that it's based on behavior. This is something that we're learning from users actually utilizing the platform and not just from overall surveillance or that creepy side.
Ravneet Ghuman (13:59.074)
sure we don't tap into any data brokers to get additional insights about users. It's the data we use. mean, our systems are fully compliant, GDPR, CCPA, and we look at user behavior in an anonymized way. if a lot of people click on certain hotels, maybe that is a popular hotel for a particular season in that city. So worth
Ian Andersen (14:23.826)
you
Ravneet Ghuman (14:28.106)
recommending that to someone we know nothing about.
Ian Andersen (14:32.718)
So does this get into where the line between AI and machine learning is as far as AI, the way I understand it, please, please tell me if I'm wrong, that you have these large language models that synthesize just ungodly amounts of data to.
to then come up with variations on whatever the prompting is. But machine learning is the system actively learning about you, right? And where you're clicking and what you're doing. I know those terms are used so interchangeably, but there really is a difference, right?
Ravneet Ghuman (15:17.416)
For sure. Yeah, I think in the past few years now, Gen.ai has become equivalent to AI and machine learning is AI. So at a high level, the way I describe is AI is this superset or umbrella that includes systems or processes that are intelligent. They could be rule based with simple if-then-else conditions.
the weather is sunny, it's not going to rain, even those simple ideas. So AI encompasses all of that, but machine learning is a subset within AI that becomes more tied to pattern matching or pattern recognition using a large data set or a data set that it can draw those patterns from. And it uses mathematics and statistics to draw or identify those patterns.
GenAI or or large language models is a subset of machine learning itself. It's not a branch as such, but it's a type of machine learning in a way. And a lot of it is based on transformers, if we were to get technical that came out in 2017. Of course, a lot more advances have happened since and and now
We use LLM, GenAI, machine learning and AI interchangeably, but there is some technical difference when we get down to the nuts and bolts.
Ian Andersen (16:58.606)
I was reading something about what always strikes me and I know I specifically remember us talking about this last time of like how old this like the technology or at least the sort of mathematics and theories behind the technology is that like there was a guy that built a very very simple computer in like 1952
that taught itself to play checkers, you know, just by repeating and learning the rules and of, you know, those before my parents were born and we were getting into machine learning. And is it just that like we have now finally come to this point with computer processing that we can, that we've seen this explosion? Is that kind of what's going on?
Ravneet Ghuman (17:53.496)
Yeah, you're exactly right. A lot of the systems and algorithms used today are very old. for the longest time, there wasn't either enough data or enough compute to sort of apply those old techniques on a larger scale to make them seem intelligent enough or superhuman level intelligence.
I remember there is this researcher that I don't know how long she spent. vaguely recall about two or three years of effort in labeling images. The data set is called ImageNet, is millions of labeled images, whether an image has a human or a cat or a dog and labeled to the details. her...
her effort was creating that label data set, but it took 20 years from the point she created that data set to someone using a lot of compute and efficient systems to create a image recognition model that could identify images better than humans could. are old. And of course, now, I don't know if...
You guys have heard this that we are sort of running out of data again in that when someone talks about using all the internet to train the data, I mean, a lot goes into what I just said, but let's say all that data is utilized. So then where do GPT-5 or the next version of the model get the data from? So there is this side of generate synthetic data, but also can we improve?
Ian Andersen (19:52.338)
It seems really scary. It's just like, we're just gonna make fake data just to keep this thing going.
Brandon Giella (19:54.134)
Yeah, that's wild to think about. Yeah. That's crazy.
Ravneet Ghuman (19:58.232)
Yeah, I mean, because I think one limitation is at least the data that's available in the public domain has already been consumed by these companies, or the companies that lead frontier research models. And so either it's synthetic data or enterprises letting their data be used in some controlled way. But
at certain point it becomes efficiency of algorithms and being able to work with less data than looking for more data.
Ian Andersen (20:35.09)
So I wonder that it makes me think of something that's been kind of on the periphery of the news lately as far as where AI is going. You hear so much about the power requirements and obviously if it continues to ramp kind of in this exponential pattern that it has been the past few years, the power requirements are gonna pretty quickly
Over take over, you know overcome what we what we can do so there's been talk of in the future of Is generative AI going to be more generalized and we'll see that kind of the AG the what is the AGI or Will it become more specialized and focused? To have kind of fewer power requirements where where do you see that going?
Ravneet Ghuman (21:32.952)
I personally see it has to be a balance at some point, in we can only, or mankind or humankind can only produce so much energy given the infrastructure that's available. So at some point it's a balance of what's the energy available and then do we make our processors or chips faster at the same time reduce their energy consumption.
And then comes the next aspect, which I hope to see in my lifetime, artificial general intelligence. A lot of research is going on in that space. I don't know if you've seen one of the documentaries from Google DeepMind researcher Demis Hassabis.
He won the Nobel Prize recently for solving the protein structure problem or protein folding problem. So there are a few labs across the world that are working towards AGI, but it's hard to assess whether is it a compute problem for now or given enough compute, does it become a data problem?
Or does it become just the algorithmic efficiency or power problem? I think those are just different parameters these research labs are playing with.
Brandon Giella (23:10.306)
I mean, think we can figure that out on this call, right? I mean, this is kind of...
Ian Andersen (23:10.354)
And yeah, right, we got this. Yeah, for sure.
Rachel Satow (23:15.996)
I was about to say this is all barring a third party extraterrestrial tesseract coming into play to power.
Brandon Giella (23:17.208)
you
Yeah, I was saying, I was thinking in my mind like, well, I'm really excited to build my cabin in the woods, you know, powered by, you know, some solar and I have one lamp, you know, and I just read books. No, want to, I want to shift gears a little bit and think more practically in terms of like, there's a lot of research going on. There's a lot of trends, there's any energy usage, there's privacy and data and,
Ian Andersen (23:28.754)
Yeah
Brandon Giella (23:49.134)
regulation and compliance around these issues. But I also want to mention and bring to bear that these things actually do these systems, the people using them, this whole world we're talking about.
It does impact business in real ways and it impacts and drives conversions for loyalty programs and things like that. And so I'm curious from the data that you've seen and maybe even anecdotally things you've heard or things that you've read. What are some ways that you're seeing AI have a tangible business impact, particularly in the travel industry? Even if it's just your own data, you're seeing X percent increase in conversions and this and that on the platforms. Is there anything like that you can give us some insight into?
to like how these big abstract concepts are what feels to a novice like myself, an abstract concept, but now I wanna implement it or I wanna invest in it and I wanna make a business decision on these kind of concepts. Like how do I even begin to think about that? What are you seeing to help me make that decision?
Ravneet Ghuman (24:49.496)
I'm not a business first person, but my perspective is those business problems are still the same. People want to find the right information in as few steps as possible and be able to trust that it's reliable. now, now the workflow that we have behind it, we can surface new features. For example, if you look at a hotel,
Brandon Giella (24:58.286)
Hmm.
Brandon Giella (25:06.008)
Yeah, makes sense.
Ravneet Ghuman (25:17.972)
search results page where people or I would look for a hotel, I would put filters on a certain star rating, I would put filters on certain prices or certain amenities to find exactly the hotels that I'm interested in and then I may review those details manually. But surfacing features, like one of the things we working on is give a natural language text option to user where they can
ask questions or type some text and find hotels that match exactly that criteria. For example, if a person is interested in laundry or laundromat services, I haven't seen other platforms offer or any platform offer an amenity filter saying you can check a box and it would give you hotels that have that service. But now with
Brandon Giella (26:13.794)
Hmm. Hmm.
Ravneet Ghuman (26:18.486)
Gen.AI, we could create this natural language text feature. A person can say that and find exactly the hotels that match that criteria. So it basically makes unstructured data or natural language text searchable in more variations than we can imagine.
Brandon Giella (26:40.866)
Hmm. So for example, let's say I'm planning a trip to Paris. We talk about Paris on this show a lot. I like Paris, but let's say I'm going to Paris and I'm thinking, okay, and I'm typing into like a chat bot maybe. And I'm just saying like, Hey, I'm looking for something in the first arrondissement and I want to be near a restaurant like this. And I kind of have this in mind for the actual hotel. I want to make sure there's a pool. I want to keep it under this, you know, price point. I'm going to be there for five days.
You can say all those things in a chat and it's interpreting that information. It doesn't have to be a drop-down filters is what I'm hearing. Is that kind of how it works?
Ravneet Ghuman (27:21.496)
Yeah, exactly. So we're not making that big a shift yet where users just interact with a bot to make their booking just yet. But we want to keep some consistency in how people were using or entering a destination, dates, hitting a search, and then looking at the results. But now they can interact with the results using a natural language text box.
Brandon Giella (27:35.82)
Okay.
Ravneet Ghuman (27:49.676)
But hopefully in the future, as we see users get more comfortable with using natural language text, we could shift to either a voice activated or text based chatbot.
Brandon Giella (28:00.206)
Cool.
Brandon Giella (28:04.334)
Yeah, it's awesome. I like that. That's cool. Are you seeing any impact on loyalty programs specifically and how they might be interacting with these platforms?
Ravneet Ghuman (28:14.658)
For sure. I think overall, if you look at, again, the same business metrics that we measured before, so in terms of customer engagement, we do see the moment we make our platform better, even if it were give search results to users quicker, we see people doing more searches. We see people engaging with more content. And almost
Brandon Giella (28:39.201)
Ravneet Ghuman (28:43.256)
It is a statistically significant difference for all the A-B testing we have done. so we do see improvements in conversion. We do see improvements in search engagement rates, the number of hotels, or average number of hotels a person clicks in a day.
Brandon Giella (29:03.8)
So going back to the top of the conversation where you were like, there's productivity improvements and there's obviously like technology improvements. You're really seeing in hard statistical data, those kinds of productivity improvements on the platform and the way that AI is kind of filtering and searching and labeling. There's real business impact, more conversions, more clicks, more hotels searched. That's pretty amazing.
Rachel Satow (29:26.865)
Yeah, and just to chime in here, Reveneet, since we chatted about this earlier this year with the launch of Neighborhood Insights, which is powered by our models, the quick stats that we had chatted about was an increase in users, a 4 % increase in user searching. So going back to what Reveneet had mentioned, we're seeing more activity and engagement with actual search. then
from those viewing destination pages. And then a 1.6 % rise in total conversion, so actual booking from those destination pages, all with the implementation of our models.
Ravneet Ghuman (30:08.148)
Exactly. That was a cool feature as well where I haven't seen any other platform do that yet. Wherein, when a person starts exploring a city, I think one of the first questions I ask myself is which neighborhood should I stay in that city? Because cities or different neighborhoods have different vibes. And so we created this feature where
Brandon Giella (30:08.302)
Amazing.
Brandon Giella (30:27.65)
Yeah. Yeah.
Ravneet Ghuman (30:35.596)
we could make neighborhood based recommendations within a city and explain why we are recommending a certain neighborhood. For example, a neighborhood is known for its monuments and history. So we recommend that neighborhood and hotels under or within that neighborhood. And that led to some statistically significant improvements across the board.
Brandon Giella (30:59.906)
That's so cool.
Rachel Satow (31:00.202)
And just, just go ahead again.
Ian Andersen (31:00.208)
Yeah, I'm... no, I looking at... I cannot... I just looked up because I couldn't remember the exact statistics, but I cannot believe this was two years ago that we were talking about this, the similar hotels thing you put together over two years ago now. And even just the short time we were doing the use case examples.
You know, we saw like 20 % increase on conversion rates, on booking rates, you know, what's that, 30 % increase on searches. I mean, just, was startling how drastic it was in such a short period of time. I mean, this was a matter of weeks that we looked at this. wasn't, you know, a whole giant data set. So I can't imagine it's just gotten even better since.
Brandon Giella (31:42.382)
huge.
Rachel Satow (31:57.383)
Yeah, and anecdotally, the neighborhood insights has quickly become one of my favorite features. So I'm actually planning a trip to New York in January to see a Broadway show. And I grew up hours from the city, so I'm familiar enough, but it's obviously changed since I've moved out of state. And I found myself really leaning into neighborhood insights to try and refresh what may have changed since I've been gone.
and to try and find the perfect hotel spot for me and a couple of friends to go see with the show, because there are some people who wanna stay in the middle of Times Square and there are some people who wanna stay outside of all of the hustle and bustle. So I found myself leaning into that feature quite a bit.
Brandon Giella (32:45.686)
Yeah, that's great. I was about to ask.
Ian Andersen (32:45.967)
So what show are you going to, Rachel? Yeah, yeah.
Rachel Satow (32:48.723)
I have full plans to see multiple while we're there, but we are specifically going to see the Cursed Child because Tom Hiddleston, or not Tom Hiddleston. Yeah, he's reprising his role of Malfoy and we're going to see him.
Brandon Giella (33:07.074)
Having seen that show, it's fantastic. It's a fantastic show. I loved it.
Ian Andersen (33:07.25)
instead of known, would be Harry Potter related.
Rachel Satow (33:09.385)
Yep.
Ian Andersen (33:14.34)
I wanted to see Hamilton when I was there last year with my wife and she wanted to see Moulin Rouge so we saw Moulin Rouge of course. It was a lot of fun.
Brandon Giella (33:24.334)
That's great. That's great. No, I love that feature. It's almost like vibe check, know, like, okay, I'm going to, let's, I'm just picking Paris again, but say we're going to Paris and I want, I want a quieter neighborhood or I want a neighborhood of more shops or I want a neighborhood that's got a lot of, let's say Michelin star restaurants or a lot of museums. Like where do I, where do I go? And to be able to, kind of put some language to that. Cause if I'm not that familiar with the city, especially that's such a fantastic feature.
Ian Andersen (33:51.194)
And one thing to highlight of just bringing it back to how quickly this all moves is, a couple of years ago when we were really first talking about AI getting integrated in with SwitchFly, it required...
a little bit more from the user as far as specificity, right? Like the similar hotels feature, you had to pick a hotel before you got some sort of specificity, right? And it's just gotten more and more, I don't know, what's the word I'm looking for? It's predictive as far as leading you rather than,
than being reactive, right? It's being more proactive. And it is maybe a subtle distinction when you're looking at it sort of side by side, but as far as usability, I mean, it's game changing, right? Being able to go in and not have to explain necessarily what I'm looking for entirely to give very subtle kind of interactions and let the program.
really tease that out of you and get you to that point. think it's a big deal.
Ravneet Ghuman(35:14.808)
For sure. I think more than making huge changes that changes or gives a new look and feel to how people search, it's these consistency of the features that people are used to. How can we do some minor adjustments? A person has already chosen a hotel that they reading about, so why not recommend some other similar hotels so the person doesn't have to go back?
in the shopping flow and sort of find another hotel. So it's these smaller incremental improvements that are making a big difference in delivering value.
Ian Andersen (35:56.562)
That makes sense just from a I don't think I've thought about that but the like if you Say you booked on switch fly a month ago and then you came back and it's completely different No matter all of the like upgraded functionality. It's probably not as impactful Then if you come back and it's pretty similar, but there have been some subtle upgrades and you're like, hey last time I couldn't do this or last time whatever like it
Even if it's not as like broad and sweeping it probably is a little more impactful per user
Ravneet Ghuman (36:33.848)
Yeah, and which is why when I was talking about the natural language search, we want to keep existing UI or UX and slowly move users to a natural language search. But at the same time, learn about. Sometimes I wonder the way people use a platform. It's more of a symptom of how the platform is designed, which is consistent with every other platform. Before we start.
thinking of a personalized concierge, because practically how many people use a concierge today? And the other part of that is the data that they're willing to share. So it's a trade-off, and we want to make sure we take the baby steps and deliver value where users see value.
Ian Andersen (37:30.128)
and which will get them more comfortable with sharing more data, right? Like it just is a natural kind of progression. But yeah, it feels right.
Brandon Giella (37:30.392)
Great.
Brandon Giella (37:40.526)
To me, listening to you guys talk about this, it's, Romney, you kind of use this language, a moment ago where it's like changing the relationship to how we're searching for travel. And it makes me think of, I've, kind of had this, this thought on my head and Ian as a history nerd, you'll appreciate this, but it's kind of like the, printing press, how it changed, changed the relationship we had to text and information. And, and then, you know, the, the, relationship between church and state and education.
and all kinds of different things that we won't get into, literally everything. But what I find fascinating is that's kind of where we're going is more closely to how the human mind works, which obviously is the underpinning of a large language model and that it's mapping a neural network between our brains and information, things like that. But what's interesting is like,
Ian Andersen (38:12.402)
Literally everything. Yeah, yeah.
Brandon Giella (38:32.07)
Ian you kind of mentioned like you when you're searching for something you just are you don't know exactly what you're looking for and it's the same kind of thing when you are speaking as I'm speaking now I don't know exactly what I'm thinking but I'm kind of searching it out as I'm speaking and a conversation a dialectic or writing itself is thinking and so I love that you guys are This is very meta but you're just kind of going in this direction of like how people are interacting with travel itself and this kind of like there's
kind of like vibing their way into the trips that they want and using AI to get there. Is that, is that, am I directionally correct in how you guys are thinking about this?
Rachel Satow (39:11.571)
Yeah, I mean, this was something that, Ravneet, you and I had talked about way back, but it was really about the consolidation of the effort of planning your journey, your, your, your, way we search. So a lot of the intention behind some of the features that are going into the Switchify platform are with the end goal of making it simpler and removing friction points so that you're not having to go.
to five different other websites to find out this information and you can do it all within one platform that you can then eventually book from. So, Reveneet, correct me if I'm wrong, but like, yes, is all with that goal of being able to streamline the whole traveler journey.
Ravneet Ghuman (40:00.28)
That's one point, Rachel. think overall there are so many travel platforms, consumer facing or reward, loyalty rewards focused. each external platform has their own special features. So rather than having people do research and discovery outside and come to our platform just to make a booking.
We have these features where we want people to be able to stay within our platform right from the point where they decide to travel and find the right destination, the right neighborhood, the right hotel, activities, or whatever the itinerary looks like, and be able to do it in one single place.
Brandon Giella (40:51.822)
Cool. I love that. Okay, last question for you just in the last two minutes or so.
What are ways that you're thinking about the future of AI and travel? So we talked a little bit about kind of vibing your way maybe into travel and the way that you're searching, but what are some other maybe technologies or trends that you're paying attention to that will really impact your work in the next 12 months? Because we can't predict any further than that. Literally every month, AI is totally different is what it feels like. So the next 12 months though, kind of where are you guys pointing your ship?
Ravneet Ghuman (41:26.36)
A big focus is we have done some experiments over the last 6 to 12 months. A lot of it is focused on operationalizing those. Our not star is our business metrics. So what drives conversion, what drives customer engagement and the features that help move that not star, we are focused on that. So a lot of experimentation, A-B testing new features and
making decisions, how far do we want to go on a certain feature where it drives business value, but at the same time, it's useful for the users. So essentially, that's the driver. then tying it back to GNI, more productivity improvements. How do we improve the quality of our deliverables? How do we identify issues earlier? How do we...
maybe automate code reviews or write better test cases. So a lot of productivity improvements and business value driving features both for users and the business.
Brandon Giella (42:44.014)
That's great. It's a great place to start.
Ian Andersen (42:45.948)
So I do have one final one, Brandon, kind of triggered by your meta wanderings is, Rebne, from a data scientist perspective, what is intelligence? And is there a way for a machine to be intelligent the same way a human is?
Brandon Giella (42:49.26)
Okay.
Brandon Giella (43:03.79)
Ravneet Ghuman(43:12.792)
That's a great question. I mean, I personally like to think of when I see my three year old son, since he was born, I'm seeing him more like a machine in training for that time in that he makes an action or says something and then
Brandon Giella (43:12.863)
In 60 seconds, go.
Ian Andersen (43:14.204)
Yeah, no bullet points.
Ravneet Ghuman (43:41.504)
My wife and I, nudge him in a certain direction, kind of supervising him and his neural network, which is maybe far sophisticated than a machine network, but his neural network is in training. And so at least in theory, if there is right data set, right algorithms and enough compute, then a machine should be able to exceed human intelligence. The question is not.
if it's a question of when in my mind. And purely speculative, that when could be five years from now, it could be 10, 20 or 30. But it's more of a question of when to me.
Brandon Giella (44:27.726)
scary, hopeful, I can't tell. But if you have any training data on my eight week old to get him to go to sleep, I would love to just like plug that one right in because I'm very tired. That's great. That's a great analogy. Awesome. Well, team, thanks so much. I love being on the show with you guys because there's so much wisdom and insight and just fun.
I enjoy it. So, Ravneet, thank you again for your expertise. I am...
I'm super excited in the way that you guys are kind of thinking about AI generally, know, conceptually abstractly, but then how it really impacts businesses and makes travelers lives better. I mean, that's the whole point. And so as a, a, uh, abstract user on your platform, I appreciate you guys and what you're doing. Um, but, uh, if you haven't yet, this is to you listeners. If you haven't yet listened to episode 13 AI and machine learning was Robney the first time, please go back and listen to that because it'll kind of get you up to speed on the philosophy.
philosophy or the approach that you guys at Switchfly just think through AI and machine learning in general. And then this, obviously this show we talked a little bit more practicality and kind of future facing. And it's really interesting to see just in the last 12 months how this kind of technology and thinking has developed and how you guys are approaching it. So please go listen to that. with that, we'll see you on the next show. Thanks guys.
Rachel Satow (45:54.111)
Thanks, Brandon.
Ravneet Ghuman (45:54.136)
So thanks for having me.
Ian Andersen (45:55.09)
Thanks, Brandon.
