Transcript
[00:00:11] Matt Welle: Hi, everyone. Welcome back to another Math Talks Hospitality. And in this episode where we discover the technology behind hospitality, I invited someone who people on previous episodes of this podcast kept referring to. And I said, what are you saying? What is this D3x? And, Jason is one of the founders of the company and I thought, come on, you tell us about it and we'll go deep on the AI rabbit hole. But Jason, do you want to maybe give us a short introduction to who you are and what you do?
[00:00:39] Jason Noronha: Absolutely. My name is Jason. I am originally from India, but I studied in Singapore, and I worked as a financial services consultant for a couple of years before realizing, yeah, life didn't feel very good. So I quit my job. I traveled the world, and I went back home to India and opened a backpacker hostel. And that's how I entered the world of hospitality. It was just, like, random and unexpected.
[00:01:05] Matt Welle: And no experience?
[00:01:07] Jason Noronha: No experience.
[00:01:08] Matt Welle: You just, and then you just started running it back in hostel?
[00:01:10] Jason Noronha: It was a little more dramatic. Like, I was traveling through Goa, which is where I was born and grew up. I was staying at a guest house where they charge $10 a night. And it was cheap, but I didn't expect much. But I was taking a shower. No hot water. Of course, no Wi-Fi. The power went out, so I was in the pitch dark, and a frog jumped on me. And I said, this can't be true. This is not happening to me. So I went down the road, rented a building, four rooms, and started what was back then the first backpacker hostel in India.
[00:01:41] Matt Welle: And how did that, like, your first experience in hospitality that leads to now being a tech entrepreneur?
[00:01:47] Jason Noronha: Oh, man. That took ten years, I think but, yeah, it was a hostel with four rooms. And I should add, guest number 7 was Laura, who's now my wife, and we've been together for fifteen years.
[00:01:58] Matt Welle: Wow.
[00:01:59] Jason Noronha: She's from Germany. She was volunteering in India and traveled through, and she realized there was no hostel, and she came to stay with us. And I'd like to think the rest is history, but, a little more, a few twists and turns, if I put it that way. But we eventually ran five hostels, three coffee shops, and two bars. And we kind of grew everything organically, had a team of 60 people working for us. But our dream was to live in New York. So we packed our bags, moved to New York, but we're still managing the business in India, and we realized the systems didn't do what they were supposed to do. And we said, okay, let's just build a little bit of tech, and it's gonna fix everything. And that's how it all started.
[00:02:42] Matt Welle: Amazing. So what was the problem that you were looking to solve from a hotelier's point of view?
[00:02:48] Jason Noronha: We didn't know who was buying stuff at our bars and restaurants. So we didn't have an integration between the PMS and the POS. And now I know 羞羞视频 has launched a POS and that all works, but that didn't exist in 2017. We had a POS, and it was just anonymous. And it was just like, are these Australians spending all this money at the bar, or is it the Germans? And we had no idea.
[00:03:11] Matt Welle: Interesting. So is that what the parts that you've built are today, or has it worked over time?
[00:03:18] Jason Noronha: No. So that product, what we started off as a small integration between what we were using back then, Cloudbeds, and another POS system grew into a PMS, believe it or not. And it, kind of, grew into a PMS for backpacker hostels. And at some stage, we partnered with Hostelworld and started serving the industry. We onboarded around 1,200 hostels onto the system, and then Hostelworld took control of that system. But that kind of freed up our time to ask ourselves, what do we do next?
[00:03:50] Matt Welle: Yeah
[00:03:51] Jason Noronha: What do we want to do? And I think we were just looking at ChatGPT coming out, and we said, we understand the operations world. We understand the PMS world, and now this shiny new technology is coming out. Can we combine all those three together and build something?
[00:04:05] Matt Welle: And what does D3x stand for?
[00:04:08] Jason Noronha: Yeah. Nothing, really.
[00:04:09] Matt Welle: You mean like, just
[00:04:11] Jason Noronha: It's three letters but I think in our minds, what it stood for is Direct3x. Yeah. But I think the way we like to think of it is, our name doesn't matter because what matters is the hotel, and what matters is the hotel's relationship with their guests. So, we firmly want to see ourselves on the hotel side and the acronym D3x is just there as a very robotic technology solution that's gonna hopefully help them grow their direct relationship with their customers.
[00:04:41] Matt Welle: Nice. So, so far, I hear you've got hospitality experience. You've got deep experience on the PMS side with all the different data flows. And then you were a first mover on this massive wave of AI combining all the experience that you've had. So what does D3x do differently? So I've spoken to many different suppliers of different chatbots, but what is the unique thing that you do with D3x?
[00:05:05] Jason Noronha: I think the way we are looking at it is we're looking to build an engine. That's the middleware that connects all these disparate technology systems together.
[00:05:15] Matt Welle: Yeah.
[00:05:16] Jason Noronha: And roughly speaking, what we're looking to do and the breakthrough technology is we are looking to convert language into API. And that is not necessarily a straightforward task. Yeah. It involves a lot of complexities, a lot of gotchas, a lot of exceptions but to give you one small example, someone might say, 鈥淚 need a towel.鈥� So, our system will look at those four words, and say, 鈥渨hat is happening here? Is this a booking modification? Is it a cancellation? Is it an invoice request?鈥� Well, guess what? It's a housekeeping request. Cool. So we now know what system we need to connect to. Who's I? Can we figure out who I is? Can we go to the CRM, use the phone number, or maybe they're messaging on an unknown channel? So we try to figure out who is I. Go to the PMS, figure out which room they're in and then now that we know it's a housekeeping task, we have to actually go to the housekeeping system. We've detected that they want a towel, so the item is a towel. And the action that we have to perform is a request for a towel. So we attempt to call an API in the housekeeping system, creating an action request for a towel. Remember, it could also be that my AC is broken, so the action might be 鈥渞epair鈥� or something else.
[00:06:29] Matt Welle: Yeah.
[00:06:30] Jason Noronha: And once we get a response back from that housekeeping system, we can go back to the guest and say, you know, we've requested a towel for you, and that should be on its way to you shortly.
[00:06:41] Matt Welle: That is such a great way to explain this end-to-end journey because I think you described one of those scenarios that is become truly end-to-end from the moment that something happens to someone, a human normally taking a note, picking up the phone to another department who then notes that back into their task management system. And there's so many breaking points and so many human touch points, and things break down in that process and I think the way you described it, you've gone beyond just being a chatbot that sits on the website that can answer FAQ questions to customers to talking to all of the systems in the hotels and then leveraging what those systems allow you to do and taking it to the to the next level. So what I'm hearing you say, and you may correct me if I'm wrong, is you've gone from, like, basically a chatbot to a true autonomous agent that absorbs the information and then does something with it rather than just responds to the question.
[00:07:35] Jason Noronha: I wouldn't call it an autonomous agent because I prefer agentic workflow, but we're seeing a lot of autonomous agents come up where agents have free reign over all your APIs and they're calling it an MCP server. But, actually, you don't want to give free reign over all your APIs to a guest. Right? What if the guest says, what's the occupancy tonight? What's revenue for next month? You don't want the agent to answer that question. You don't want an agent to tell the guest, who's staying in the next room? Give me their email address. So it opens up a lot of security problems if you have an autonomous agent that anyone has access to. So you actually need to box that agent into certain, like, a road.
[00:08:14] Matt Welle: Yeah.
[00:08:15] Jason Noronha: There's certain guardrails. So it's only connecting to that system when it's absolutely necessary to. And even when it connects to that system, it can only do a, b, and c.
[00:08:24] Matt Welle: But does the hotelier have to build the workflows? Or they just have to say, okay, have access to these systems, and then you look at what the scope is of those systems and the extent of automation. So the towel, for example, that you gave, does a hotelier need to say, okay, if someone asks about a towel, then connect it to this system, or you do that for the hotel?
[00:08:44] Jason Noronha: We do that for the hotel. So if you would use a horizontal customer support AI, you might have to build flows yourself where you're kind of going through these if then kind of scenarios. We built out those 鈥榠f then鈥� scenarios ourselves, and it's specific for hospitality. So, we are very opinionated on how we are using these AI agents to say, we're building for this hospitality use case. The hotelier doesn't have to worry about any of this. All they need to do is go to the 羞羞视频 marketplace, connect D3x, they need to go to the housekeeping software.
[00:09:16] Matt Welle: Yeah.
[00:09:17] Jason Noronha: So they could go to Flexkeeping, connect D3x there, maybe connect their CRM, go to Bookboost, connect D3x, and there we go. It's connected, it's talking to Bookboost, 羞羞视频, and Flexkeeping at the same time.
[00:09:28] Matt Welle: So you don't actually replace the CRM. You are a partner to the CRM. Yes.
[00:09:34] Jason Noronha: Yes.
[00:09:35] Matt Welle: Got it.
[00:09:36] Jason Noronha: We don't need to replace the CRM because people have been building this for many years. They're quite sophisticated with how far they are with those softwares.
[00:09:44] Matt Welle: Like, it's clicking in my head why so many people were talking so positively about what you're building because every time I step behind the reception desk for you know, I do cross-exposure with our customers a lot. I always ask, can you show me your checklist? What are the things you have on a piece of paper that every day you have to go and check? And it's like the trade says, like, the activities on bookings that they have to manually call to the housekeeping department to add an extra bed in a room or some things like that. And it feels like you're solving for the checklist, that paper thing that they have behind the reception desk.
[00:10:16] Jason Noronha: Yeah. And the checklist could be dynamic as well because the guest might be messaging at two in the morning, and it doesn't get picked up till nine in the morning, but it should be handled on the spot.
[00:10:26] Matt Welle: Yeah.
[00:10:27] Jason Noronha: Ideally. Yeah.
[00:10:28] Matt Welle: How do you deal with voice? So if I am in my room and I wanna call reception, are you able to already pick up voice or not yet?
[00:10:36] Jason Noronha: Yes. We're able to pick up the phone. And the way we do voice is, firstly well, there are two ways to look at voice, and we're kind of going down both directions. And, again, this is slightly more technical. But, again, this is the tech that we are dealing with on a day-to-day basis and trying to abstract it from the hoteliers. But we are looking at two possibilities for voice. One is we break the voice down into text, and then we process the text, and then, we have to process that text very quickly because we need a response back in a second or two seconds. Right? And in order to get that voice back quickly, we also start streaming. So we start constructing the sentence. And while it's speaking the first bit of the sentence, it's thinking about the rest of the sentence.
[00:11:19] Matt Welle: Right. And just one of the things I always see is latency. Right? So, when you call someone, they will immediately start responding even though in their head, they're still forming the thoughts. Whereas with a lot of AI voice products, and we've looked at many to help with our support, there's, like, a three, four second delay, but that is unbearable, like, it's this pause. And, like, the way you're handling it is basic saying, well, two seconds is acceptable, and then you start talking. And then whilst you're talking, you basically figure out the rest of the sentence. I think that's a really smart way.
[00:11:52] Jason Noronha: There's a challenge though, because when we convert voice into API, which we can do, someone calls up and says, do you have rooms available tonight? We need to go to 羞羞视频 and we need to check rates and availability in 羞羞视频. We need to make an API call and those API calls can be quite expensive. So they might take some time. So in that case, we might try and add fillers in there to say one moment, please, or we might add some kind of filler words in there. Hang on, um, that kind of stuff. Yeah. So, those are the kind of challenges that we have to deal with voice.
[00:12:21] Matt Welle: And are there hotels today that use end-to-end voice for reservation handling, for example?
[00:12:29] Jason Noronha: We don't do end-to-end voice and the challenge with voice is, again, that AI might not be able to interpret a person's email address correctly. So look at my name. Imagine spelling that out on a phone in a noisy train station, and it messes up the spelling and sends an email to the wrong person. That reservation is lost to me, and that's quite a poor experience.
[00:12:51] Matt Welle: Yeah.
[00:12:52] Jason Noronha: And the second thing with voice is you're kinda boxed in. You can't show people nonrefundable rates, refundable rates. They don't retain that much information. So what we're looking to do for now is kind of funnel people to the 羞羞视频 booking engine. So we're just like, let's construct a deep link. Let's assure them that a room is available. We can support their query. We have this kind of room type available for them. But then when they have to make that booking, we send them a WhatsApp message or a SMS to say, please complete your booking here. And they also have that compliance and security in mind when they provide their credit card information. It's dealt with in a secure manner and not over the phone.
[00:13:29] Matt Welle: Yeah. Fantastic. So has there been a time when the AI has really surprised you in what it could handle in a workflow where, like, I didn't even think that it would do that before?
[00:13:39] Jason Noronha: Oh my God. Yes. I had one of, and I think I've seen a lot of conversations at this stage, but every now and then I see a conversation where I just share it with the team, and it was like, what happened here? How is this possible? And the particular scenario was, we have a hostel chain based out of Dublin called Clink, and they're very, like, forward
[00:13:59] Matt Welle: They鈥檙e just down the road from me actually
[00:14:00] Jason Noronha: Yeah. Right. And Diogo was on your podcast as well a while ago but I think what we noticed with Clink was a guest was messaging saying, 鈥淚 made this reservation. I can see my reservation. It's only got one person on it鈥�. And the AI realized that it had to authenticate the person because this chat happened on the website, so it was an anonymous person's messaging. So, it asked them for their reservation number and last name, which they provided. The AI then decided to call 羞羞视频 and do a reservation lookup in 羞羞视频, and it did that. And it noticed that the person actually had made two separate reservations for one bed. And it went back and told them, hey, maybe you wanted to book two beds, but you actually booked two reservations with one bed each. So I hope that works for you. And the guest was like, yeah, that's amazing. That's great, like, that's what I wanted.
[00:14:51] Matt Welle: I think a lot of the time, what I see with AI bots is that the provider doesn't inject context, and context is everything, which you get the context through API embedding. Right? So if you just put a bot on your website and feed it some FAQs, it might get it right. But if you tell them, this is Jack who's been with the hotel 20 times and this is his history, like, it's going to get the answer significantly better than answer it from an FAQ sheet. And it sounds like that's what you're doing. You're feeding it context constantly. And the more context you get, the smarter the AI gets.
[00:15:30] Jason Noronha: Absolutely. Yeah. You're absolutely right. And so we need to connect to those underlying systems that have up-to-date context about the customer. Did they cancel? Did they add another guest? Are they in-house? That's the biggest one. Are they in-house? Are they not in-house?
[00:15:44] Matt Welle: And you integrate with how many systems today, and what's the plan? Or how fast could you integrate with a new solution?
[00:15:49] Jason Noronha: Well, it depends. So if we were to add a new category of solution, it takes a while. So for example, what we're working on right now is table reservation. So that needs to go really deep because we really want to reserve the table for the guest and figure out which room they're in and all of that stuff so that we're passing all that information through without asking them to repeat themselves. So, that's what we're working on in terms of a new category. When it comes to things like we've got two housekeeping systems, we need to add a third, that can go quite quickly.
[00:16:20] Matt Welle: Yeah. Yeah. If it's just a copy paste of the same MPN endpoints, kind of, it's easier than a new workflow, I guess.
[00:16:26] Jason Noronha: Yes. But every system has its own nuance, and every system gives you a few more things to do or a few different things to do, so we have to keep that in mind as well.
[00:16:36] Matt Welle: Can it also do upselling? Like, one of the hardest things in hotels is always teaching the humans at the reception desk how to do a nice upsell to customers because it's just not a natural thing for humans. I've actually seen the kiosk, for example, the 羞羞视频 Kiosk does an incredible job but how can you do upselling through your solution?
[00:16:53] Jason Noronha: So we're currently doing intent-based upselling. So if a guest comes to us with intent, we upsell them. In terms of outbound upselling, we rely on our relationships with other partners to handle that upsell because maybe and a good example would be a system called Bookboost, where they have really cool audiences and segments and rules because it's a CRM. They know everything about the guest, so they can kinda send the right message at the right time to the right guest. But maybe the guest replies to that message and says, yes. So our job is how do we take that original marketing promo message or upsell message that's gone out to the guest? Take the yes, and then figure out which reservation, which room? Do we add it? Do we confirm the price with them? Maybe they've already agreed on the price. How many people, that kind of stuff and then actually execute on that whole task. And a good example would be, and this can be quite complex, so for example, with the best westerns in Sweden, at the Stockholm Alanda Airport, they kind of offer parking, but parking can be quite complex. They have one price for seven nights, one price for fourteen nights. But if you're actually staying at the property, parking's free. So the AI needs to figure out what kind of parking are you talking about?
[00:18:13] Matt Welle: Yeah.
[00:18:14] Jason Noronha: Is it while you're staying at the property? Or if it's Park and Fly, then it would go into 羞羞视频, fetch the price from 羞羞视频, and then actually confirm the reservation and add that Park and Fly booking to their profile.
[00:18:27] Matt Welle: So I'm sure a lot of people are listening to this, and you鈥檙e like this sounds too good to be true. This is the sales pitch. Like, is this reality, or is this how smart it is to get all that context?
[00:18:38] Jason Noronha: This is reality. I've seen it sell a tour of the Grand Canyon for $180 because it's kinda figured out, like, this is the tour. This is the description. It even knows the stuff is, like, you'll get breakfast on the tour, but bring your own sunscreen and then does some kind of pattern matching where it goes to the PMS and adds that to the guest reservation.
[00:19:00] Matt Welle: And how do you deal with security? Like, it's so hard to identify customers or safely share information. Like, how do you think about protecting the identity of customers through GDPR?
[00:19:11] Jason Noronha: Well, we have to firstly think about data residency. So we ensure that all of the customer information is sitting on a secure server. In Frankfurt, we have end-to-end encryption. We're currently going through our SOC 2 Type 2 compliance as we speak so that we can also provide an external audit of our security. So it's just essentially ensuring data is encrypted during transmission at risk, at rest. We've got role based access control, so the right people get access to the data based on their security access levels. And the last thing is we don't try to access data more than what's necessary. So, again, if you're chatting with me, I'm gonna go to 羞羞视频, extract your reservation information from 羞羞视频. I'm not gonna save your 羞羞视频 reservation information on my system. I'm only gonna access it when I need it to talk to you, and then it's just forgotten.
[00:20:05] Matt Welle: Great. If you think about the resistance of hotels towards AI, how do you tackle that? Because it's real. Right? We've been trying to convince hoteliers to move into the cloud already for twelve years, and even that's a conversation. How do you convince them to lean into some of the amazing innovations that you've done? Like, how do you convince them?
[00:20:26] Jason Noronha: Well, I think, yeah, I've been talking to a lot of people now over the last two years, obviously, and it's been, like, people ranging from different backgrounds and different countries and different styles of hospitality. And I think what I'm largely seeing is two kinds of objections that come up. And with the luxury and the boutique hotel crowd, it's more about human touch. I don't wanna replace the human touch with a robot. I'm afraid that it might degrade my service, and it might not be as good. And to that, I think the numbers that we're measuring tells us a different story. Because what we do is at the end of every conversation, we send out a CSAT request to the customer to say, can you rank this conversation from 1 to 5? And across all the properties on our system, the AI score is 4.6 on 5, and the human score is 4.37 on 5. And we track that, like, quite rigorously and at the same time, we are seeing the AI increasing every month. So I can see that going higher for AI, but I'm not sure it could go higher for humans. You know? Like -
[00:21:30] Matt Welle: Yeah.
[00:21:31] Jason Noronha: What's gonna happen next month? So that's the way we are thinking about it. And another thing, like, from a resistance point of view, we're seeing another type of persona, and that's the large enterprise customer who might have 400 people out there. And they don't want to train 400 people and give them access to a new system with another username and password. So their challenge is how do they bring AI into their organization, bring AI to their team without necessarily replatforming that team and retraining all of the team and essentially managing for change so that they can get the benefits of AI but without the chaos.
[00:22:08] Matt Welle: Last question. What's up next? What's something that you get really excited about that you want to build?
[00:22:14] Jason Noronha: It's a really tough one, but I think it's gonna be really important. What we're working on right now is invoice modification in 羞羞视频. So someone comes and says, hey, there's no VAT ID on my invoice. My address is wrong. My company name is missing. So, we want to firstly authenticate that person so that we know that this person is connected to this reservation. The way 羞羞视频 treats it is once people check out, it's actually a legal document. So we need to actually kind of void that invoice by issuing a credit note. And then we need to create a new bill, tag that new bill to the invoice, and then generate a new invoice that's then linked to that bill. And that's a whole workflow, but we want AI to handle the entire workflow. And I'm very excited about it. It's very boring. But one of our customers, it鈥檚 one of our customers genuine problem.
[00:23:07] Matt Welle: Like, people listening to this, oh, invoice, like, this is one of the most painful things that happen in hotels. And often, this happens after they've left for their home. They're submitting it for their expenses, and they realize that the address is wrong. They then call the hotel, and you can imagine how painful it is now to get an invoice from two weeks ago, reissue rebated, reissue with the correct address. And this happens so much. And not only does it happen in the hotels, often it does, like, create support tickets on our side. So, this solves a massive problem, not only for the hotels, but also for us at 羞羞视频.
[00:23:39] Jason Noronha: Yep. So it's a big challenge, but we're up for it and that's why I think it's more about workflow automation, and that's definitely a workflow that needs to go.
[00:23:47] Matt Welle: I understand why everyone is excited about what you guys are doing. Keep on doing what you're doing. I'm watching closely, and this is, like, such a good, I get so excited about AI because I understand the possibility of it. But seeing someone already bringing this to life today in hotels and hoteliers loving it, it's really, really powerful. Congratulations.
[00:24:09] Jason Noronha: Thank you. Thank you very much. And thanks for offering us a great API, an open ecosystem, amazing documentation, really great support as well when we email. And that allows us to do what we can do because AI without API does not work.
[00:24:26] Matt Welle: That's a very well spoken sentence because I had told us to be brief. It was honestly why we started 羞羞视频 because we're told, you know, by the hotel that we were working for, well, you just gotta buy this industry-leading solution. I'm like, great. What's their API? And there wasn't one. And that became the reason why we built 羞羞视频 with an open API that is literally on our website. Like, if you want to see the documentation, it's on the website. It's not hidden behind some documents I have to sign. It is literally open. And today, I think a lot of the industry players have now opened up their API to the point where you can almost find it on websites. You can almost find it ten years ago.
[00:25:01] Jason Noronha: We're still applying for access, and we are, like, chasing a lot of people. So I wouldn't say the industry has moved just yet, but yeah.
[00:25:09] Matt Welle: We're working.
[00:25:10] Jason Noronha: Definitely. We're getting that.
[00:25:12] Matt Welle: Thank you so much.
[00:25:13] Jason Noronha: Thanks for having me, Matt.
