Revolutionizing Customer Service with AI w/ Mattieu Detaille

Mark: 00:56

Hello everyone and welcome to another episode of the CTO Compass.

Today we have Mattieu. Mattieu is the founder and CEO of Elora. Elora builds AI voicebots and he's going to tell us a lot more about it in this episode.

So, Mattieu, tell us a little bit more about yourself and how you got to starting as a company.

Mattieu: 01:20

Yes, sure. So yeah, I'm Mattieu, CEO and co-founder of Elora.

So at Elora, we are building that platform, you know, for businesses to be able to, I would say, set up the AI receptionist in less than an hour, so super fast, and answer, I would say, Level 1 questions that the users or the clients might have when calling them. We transfer calls and book meetings with those AI agents. About a year ago now, really to implement this - Thoughtful? And so everything is based on Jenny I saw. The core, I mean, AI is really in the core of our product.

Mark: 02:14

And I love that. And yeah, I love that you're building everything an AI first company. What is it that. Got you the idea to start Elora.

So what's, what happens that you thought, this is what I want to build. This is a good idea.

Mattieu: 02:28

Yes. You know, so at first when we saw, I would say, the Gen AI technology models, right, we saw that the I mean, we're pretty good for chat. And I mean, the chat market was quite saturated very fast. And so that's when you, We were feeling that you had to reinvent something or go higher than that. Otherwise, you would just be a company that does chat and that's it.

So we started by doing chat at first. But then I was talking one day with my co-founder, Raya, about her job. She was a sales director in another company at a point. And she told me the point that she had to handle And she was bothered by all those calls that she was getting during the day. It was taking, I would say, a lot of time from her and clients.

So their clients and their users were sometimes waiting, you know, more than 15 minutes just to get simple questions answered. And that could really, I would say, make them sometimes. Lose the offer, lose the job in itself that they were selling, and so make the client switch providers.

So that was really an issue. Then we kind of observed afterwards in the market that for small companies specifically, that's often, I would say, executives or important people in the company who are answering those calls. And so this takes a lot of time for them. And this time could be spent, I would say, somewhere else for sure.

So that's really where the idea comes from.

Mark: 04:09

Okay, and it's quite interesting. So is this the idea that you had at the start, or did you... Pivot anywhere along the way. When did you feel you achieved product market fit.

Mattieu: 04:21

Yes, indeed. We started at first because we wanted to be, I would say, even more innovative and we wanted to begin with outgoing calls, right?

So meaning that the AI kind of calls the people, right? So we did some prospection campaign. We did some debt collection campaign for some clients. But we noticed that it was not. Working properly, the results were not there. Meaning, for example, when you do prospection with an AI, I mean, it's already hard enough to do cold calls, but imagine that you are called by an AI out of the blue.

I mean, today, now it's happening all the time for some people. But you are just fed up and you hang up, right? And you totally feel... Comfortable to hang up to an AI, even I would say, in an easier way than with a human, right?

So the results were not quite there. Then, debt collection.

So we tried to do debt collection for some companies. A few companies, we tested it and honestly, It was kind of hard because we had to reach the right departments and it was already hard enough to make the clients pay, but when it's an AI calling, they recognize the voice, they hang up, whatever.

So to get the notification was not super nice. And the results were not quite there as well.

So that's why we decided to pivot and really have received those incoming calls where people are trying to get an answer. So they're already ready to wait a bit and to, you know, get, I would say, an answer and they can get a faster answer with the AI, which is the point here, right?

Mark: 06:00

Yeah, no, and it's true. I mean, my, most of us don't have really good experiences in calling customer service.

So, usually spend 10, 15 minutes on the wait, and then still don't know if you're going to get a good answer. So I think, Yeah, so it sounds like the right place.

Mattieu: 06:15

And most of the times the worst is that it's just about transferring the call to the right person in the company. And so you have only someone who is picking up the call, right? A receptionist.

And then he or she needs to transfer it to the right person and that's it and they do that all day and so they have plenty of other stuff to do in the company. So that's a good. I would say, tool for them to.

Mark: 06:39

Use. Now, and in the past, and it's probably still there, you have all these office collection buildings where you had like multiple companies in one office and they used to have one receptionist just indeed what you said, picking up the phone, transferring to the right office, the right company. And it's, yeah, it's absurd. Nice.

So taking back a little bit, your backgrounds before, I mean, you have a computer science background, you have a lot of, you worked in bigger companies before starting your own company. What made you decide I'm no longer going to work for these large companies, but instead I'm going to do my own thing, my own startup?

Mattieu: 07:14

I mean, I think I really... Started I mean the idea came out directly when I was in university, right? I tried already a couple of things there. Then for whatever reason, I guess I let them. Myself be convinced by other people or my family that I should first try, you know, to go in a corporate environment.

So I started during three years. So I worked three years for a company here in Luxembourg, but I noticed directly that it was not really my type of thing, right? It was too stable, too...

So yeah, I didn't really like it. Just because of me, right? Not because of the company. But then... I kind of decided then to try to go for freelancing. And so As I have a computer and cloud development background, I tried directly to go freelancing. I did it for more than two years for some clients that I found out. But it was kind of a first step as I had kind of to sell, I would say, what I was doing to kind of find the right pricing and so on.

So that was kind of the beginning for me to really go freelancing. I would say to the startup and create my own company.

And then the next step, I would say, after the two years, was really for me to try to create my own startup because that was the ultimate goal behind it, right? So that was, I would say, gradual. I like always to go gradual, like going directly and try it directly. For me, it's not like the best thing.

So I need to go gradual. Yeah.

Mark: 08:54

Okay, cool. And then you started Elora last year.

So one of the things that I really like about Elora is it's not actually English because most of these are not English first. I mean, the website is in English and of course you support English, but you are from Luxembourg and you actually support a lot of other languages as well, like French or German and even Luxembourgish, which is a language. That must be quite hard because the models they work, I know I'm Dutch, they work really well for English, but other languages it's often less or less good.

So how has that been for you guys doing multiple languages that aren't English?

Mattieu: 09:31

Yes, I mean so for now we support English, German and French. We are working really on the Luxembourgish version. I would say that yeah, the goal was really to provide a way to switch the language inside the conversation, right? When the AI pick up there is always a default language, for example, French or English, and then the user can switch. I would say it's The hardest part is really for the part where you are translating what the user is saying, right?

So you do speech to text. And that's where the model has really to recognize the language that is spoken. And especially if you have some languages that are very close to each other, such as, you know, for example, English and Dutch, some words that are common, I would say there is a bit of like accent that is common.

Like it's. It's kind of hard at a point. You can have some words that I even recognized in the language of the other.

So that's really, I would say the more. I would say voice that you record, the more you will be able to choose the language and the model will choose the language. That's kind of really the challenge here. This works really well between French and English, but for example, English and German, it's still like getting from time to time, like trying to go to one language or the other. If you kind of persevere and ask the model, I would like, for example, to speak French now, then it's like directly switching or like German. But so we are getting there. But what we notice as well is, especially in Luxembourg, you know, A lot of people are speaking a lot of different languages.

So if we choose, for example, French as the default language, most of the people don't even switch. They just stay in French.

So that's kind of... Good for us, but I mean, the goal is obviously to be able to switch in the conversation easily.

So yeah.

Mark: 11:37

Yeah, I'll tell you my experience. I use these note takers all the time. And of course, I'm native Dutch.

So sometimes I have a meeting and. We start with two Dutch people speaking Dutch, and then we translate. Because somebody else joins, we just switch to English, which is fine. In the notes, it just doesn't work.

So the first part of that whole conversation is gibberish because it's trying to transcribe what we said in Dutch to English and it's nonsense. And then the English part is good. And it's so strange for me that it can't pick up that it was Dutch at the beginning and then English.

So I can imagine how hard that must be indeed.

Mattieu: 12:11

And maybe for the Luxembourgish part, so we're really working on a model, a specific model, because for now, there are some, I would say, text-to-speech, so voices in Luxembourgish, so you can translate text to Luxembourgish and it's working pretty okay. But the speech-to-text is not really existing. And so you have to collect the right data, you have to train the model, you have to, so that's really a whole process. But we are getting through it.

Yeah.

Mark: 12:40

So you're training your own custom model for Recognizing Lux Mortgage. Exactly. Wow. That must be quite a lot of work and quite hard.

Mattieu: 12:51

Yeah, I mean, but that's really a game changer in Luxembourg, right? So that's why I'm getting it.

Mark: 12:58

I can imagine. So, and adding to that, in the beginning you said you, well, when you started, there were all these chat-based companies and they do that because chat is easier, I would say, because you have a message, full message input and a full message output. It's a lot easier to, you don't have to use transcription. You've chosen voice. What have been the hardest things that you've had to deal with in just the whole voice, speech to text to speech model, apart from everything you already said?

Mattieu: 13:26

I would say the hardest thing in this product is really the real-time constraint, right? Because you have to answer in less than, I mean, That's the theory, right? Under 500 milliseconds, then the conversation is smooth if you answer that. But You are using so much, I would say, external APIs that have to really integrate that it's hard to have really a stable product and you have to optimize a lot in case. One of the API fails, or there is a bit more delay than another day.

So that's kind of hard. So because imagine you're in the telephony area, so you have already, I would say, a platform in telephony. You have then really the speech-to-text where you translate the voice of the user to text. You have to generate via an LLMU answer and then go back, I would say, to the text-to-speech to generate the voice again. And here I'm not even speaking about the rack component, right? In Gen AI, we are kind of getting the data of the specific client in order to like formulate your answer, right?

So what we notice is today, most of, you know, the products and the platforms that exist in telephony and do, I would say, Gen AI agents are kind of telling you, Okay, we have this. Super nice delay, right? It's less than one second or whatever, but they don't even use a rack behind.

So you can just have a script of whatever you want to do. Everything is in the context of the LLM, right?

And then you don't really go and check, I would say, the specific data of the company in itself. So that's kind of... Different and we are trying to do it in another level.

Yeah. For that.

Mark: 15:10

Then you can't book a meeting or take any actions, right? If it's only the context within the RAC, you can't go and book a meeting somewhere else in another tool or check availability of someone.

Mattieu: 15:21

I mean, you can if you connect to external, I would say, providers, right? But here I'm saying about the RAC. Or data context of the company that you have to have somewhere, you know, an agent to be able to answer those questions.

Mark: 15:37

Yeah. Okay. Makes sense.

So, and still, you've been at this for about a year now. You've been quite successful. There's a very, you've had a very large first funding rounds already.

So congratulations for that. And especially, I mean, I see quite a lot of seed rounds here in the Netherlands and Luxembourg is even smaller. That's quite a substantial seed rounds. What happened? How did you make that happen?

Mattieu: 16:02

I mean, it's a lot of, I would say, networking events that we attended, you know, to find the right person, I would say. So here it's really a combination of three sources, right, for the fund.

So we have first an accelerator in which we were selected in Luxembourg and provide free equity opportunities. Funds.

So that was the fit for style, right? We can name it here. I think it's a bit of publicity for them.

So that's good. The program was really super nice. Then it's an R&D grant from the state that we got in order to, you know, for example, train specific models.

So the Luxembourgish model and so on. So that's really... Here linked to the R&D that we are doing.

And then it was a business angel that provided the rest. So really, it was really via events that we kind of met, you know, this business angel, and then we kind of talked about, you know, the topic, we kind of, Then we began to know each other, we met him in several events in a row and so on.

So progressively we kind of then really found a common ground and this happened. Really like a process.

Yeah. You have really to think it through, think it ahead. Usually they say that it takes six months for us. It was approximately that.

So yeah, you have to think it through. Let's say, really, before you started and you have to come, yeah. Those kind of six months. That's interesting.

Mark: 17:48

Still, congratulations on such a first round. So it says a lot about the future and what is possible for you.

So I think adding to that, you're based in Europe, you're actually based in Luxembourg, a small country, but still you're competing globally. Do you think that's an advantage or disadvantage being based in Luxembourg and why?

Mattieu: 18:08

I think it's an advantage because it's really First, a country where they speak multiple languages, right? So as we said, that we can kind of even if they kind of can adapt to whatever language you provide them, that's kind of a good way, I think, to test it because you can still ask them at a point to speak different languages for testing our agents. But then it's a super nice country because you are very close to the authority, right? And so if you need something, they're kind of really listening to you. They're very close to you. They can adapt easily.

So that's really helping. And they know that actually Luxembourg is small, right?

So they really kind of... Tell you from the beginning will help you go international and so on.

So they are kind of very supporting. To the startups and to the ecosystem. Even for corporates to kind of go international because they know about it and they recognize it.

Mark: 19:09

Okay, that's kind of a cool story. I don't think we have that here, at least not from the government. It always seems very far away for a startup. No, it must be really cool to have. Okay. From all the customers, I mean, you already have some customers live on the platform. What's the most exciting?

Well, maybe not live, but what are the most exciting customer cases that you've seen come by already?

Mattieu: 19:32

Best of my case. Interesting question. I would say that What is exciting is, I mean, at least for me, is when you kind of, So I would say this product, right?

So the receptionist on demand, and then you discover plenty of other use cases that they could use your platform for, but that you didn't think about, right? So that's the case with the first few clients when they say, okay, but we could use your product there and there. They're even asking now if we support outbound again, calls for them, but it's... In some ways that we didn't think about, and it could be like really replicated in another in a lot of companies and so on.

So this kind of really make. Think and evolve the product. You know, so that's really good.

Yeah. Because when you kind of work and you have not yet, I would say a client that is using your product, you're kind of. Working in a theoretical environment and you cannot really enhance your product at the end of the week.

Mark: 20:44

Maybe you're going back to outbounds again.

Mattieu: 20:48

Yeah, maybe at a point, yes. Maybe. Who.

Mark: 20:51

Knows? So, and I was reading about your co-founder, you mentioned her, Rea. It's quite a classical. Combination you have, I think you're more the technical side. She's very much on the sales and the business development. Side how does that work for you guys i said they work well together must be quite a.

Mattieu: 21:11

Yeah, I mean... You can, I would say, find often the normal conflicts that you see everywhere between the sales and the tech, right? Okay, the product is not yet ready.

Yeah, but we have to sell it, you know, so that's kind of always this kind of conflict for sure. It's really, I would say, what you want to have in your company, right? Different skills, but like some people that think differently than you are. And that's really what makes us evolve.

Yeah. The fact that she's a woman obviously, so she doesn't think like me, like a man.

So that's a good way of making the company evolve. Like see all the different options.

Mark: 22:02

Right? - Still, you're the tech guy. And if I look at LinkedIn, I still see you attending or attending a lot of different events.

Like you were in CES in Las Vegas last year, you've been in other events. So where do you find the time to do all the technical developments, oversee the tech team that you have, but still go to all of these events? How do you make that happen?

Mattieu: 22:22

I mean, that's a lot of work. I don't sleep that much anymore. That's for sure. But no, that's part of a, I would say a strategy we have. And, you know, I'm not alone in the company anymore, especially to code the product. Right.

So when you get this investment, it allows for sure to enlarge the team. And now I'm more like I would say CTO, I would say in terms of tech, right? CTO types of role, at least for part of the product. And I'm delegating a lot, you know, we have, I would say, because I said CTO, but we have a person, I would say, playing the role, part of the role of the CTO in San Francisco, Andrea, and he's helping me like, you know, in that area to, I would say, be part of the development of the product.

So I would say that's really helping me to step a bit out of the coding in itself and so to add an event, develop strategies and so on. So, but that's how it, I would say, has to go because if you're always in the development and there is no one kind of, you know, trying to have the vision and see where we are going and so on, you kind of not know where you're going and that's Can't be problematic.

Yeah. But I would say the short answer would be that I'm not alone anymore.

So that's why I can do it.

Mark: 23:53

Which is good so spending less time behind the keyboard actually coding these days Cool.

Mattieu: 23:59

Yes, I would say concentrate more on the part where I'm needed.

Mark: 24:05

You're building out a team, you're hiring people, you've built a tech team, you're in the AI space, which is incredibly competitive. If I see some of the things that are going on in the US, it's completely absurd. Where do you find the talent? How do you approach hiring in such a competitive space?

Mattieu: 24:23

At first, I would say when I had to hire the first people, it was kind of hard for me because I was really a cloud developer before and I never really even managed technically people, right? So that was really like going from being a developer to kind of hire and manage people. I would say it's first about surrounding yourself with people that did it, right, to have the learned lessons and so on. For example, Andrea, like, so our CTO, like, I managed people for like now 10, 20 years.

So he knows how it goes and he could kind of advise me there and that was really helpful. I would say secondly is to have the right tools. For example, for us, now we use... You must know maybe I do. There are plenty of other tools like this, but we use it as a recruitment tool so you can kind of easily, you know, then put profiles of people, compare the results and so on.

So that's really easy, even for applications. And I would say it's about. Trying, right? You do the first interviews and you see, okay, no, it wouldn't. To go well or it would go so I would test. We had some, I would say, people that we thought would, you know, even with you, right? We had some people that... We thought would be a fit. And after a few weeks, you're like, okay, it's not.

So I would say it's about not hesitating to say, okay, we. We kind of tried, it's not going okay, so we have to cut the contract, right?

So it's about trying and say, okay, this worked, this didn't work, and then at the end of the process, you kind of learn what goes well, what doesn't. But I would say in the AI sector, It's very hard to find the profiles. That's the first thing. Even when you find it, for us, we found, for example, two profiles that would fit. We waited a bit of days before really telling them they would start. It worked. They signed the contract, everything.

And then at the end of the process, just one day before, they kind of canceled. When signing the contract, you know, and you're like, I spent now four weeks to just, you know, like, try to hire this person.

So what's going on? And so, We lost, I would say, a couple of months because of that. But at the point, you're kind of getting better, right? Because at first, when we found, for example, the right AI profile, we were like, okay, we found him, like, that's perfect. Let's wait for him to start. He didn't start and we kind of didn't continue in the meantime, the interview process.

So that's kind of something we didn't do after, right? We were still continuing the interview process.

So for the second, I would say, profile. You cancel, then boom. We kind of had someone ready already to take his place if he wasn't starting. I would say what really made the difference is that we used, so for the AI profiles hiring a startup, in our incubator, and, The startup was from, I mean, The people in there are from Romania, I can name them, remote labs. What really helped here, I think that he convinced the profile that it was the right choice to join us because of these reasons, it kind of I would say sold well Luxembourg as well, right? Saying, yeah, it's really a, his country and so on.

So I think that this kind of made the difference where, you know, This guy from this startup was from Romania and the profile who we hired was from Romania. And they kind of found a match and a trust, I would say, between them. And I think that's what made the difference between this and the other profiles that we tried to hire.

So I would say it's very important to go sometimes with. People who were local there before, that can help. It depends I would say, on the sector that you are targeting, right?

Mark: 28:55

Cool. That's a cool story. I think that, yeah, a lot of people can learn from that because it's still hard to hire the right people in this industry today.

Mattieu: 29:04

Yeah. I mean, in the AI, yeah, because...

Yeah, super expensive profiles, but they can choose wherever they want to go, right? I mean the really the best profiles I mean I.

Mark: 29:17

Of course. And you said you've now started to lead people as well.

So that's new for you and Elora. So what have been your biggest lessons about leading people and what's your... Your beliefs right now about leading people in tech. What have you learned?

Mattieu: 29:34

I mean, I think, yeah, I've learned a lot of stuff. I would say that first, You know, using, I would say the right approach and tools to lead them can really make the difference, right? I would say that With my experience, I... As a freelance, like I saw in, I would say, big corporates how they were doing it, right? And so that's kind of processes that you are putting in place, daily meetings, this kind of things that can really set the tone. But it's about Again, surrounding yourself with the right people. For example, Andrea helped a lot with this to be more efficient, saying, okay, daily meetings, it's good. Meetings in general that are fixed, it's good, but not...

Like we don't want them to lose the time in meetings, right? So let's make it more on-demand meetings, this kind of thing.

So we're kind of trying different approaches that he was trying that I learned and so we're kind of making this mix of For example, if we speak about meetings between fixed and on-demand meetings, but trying to reduce as much as possible the fixed meetings, I think it's about.. Trusting, I would say, people, especially we are in a remote environment, right?

So you never know at a point if the person is not relaxing in his couch or like really working. And so that's like trusting the developers at some point. But if you're not going to be able to trust them. You feel that something is odd, like I never hesitate to kind of, you know, ask and establish the communication and say, okay, this is wrong.

Like now we need to change this or whatever. And I feel that if you are transparent with them, they will be transparent with you. And at a point, I would say, we began with freelancers in our startup, right? With hiring freelancers. And I think it's the good way to go because at that point you have someone that is responsible like that knows you know, That's...

You know, he has to be there, work, provide the best of his abilities, because otherwise the contract can just stop. Right? And so this really helped. Then Yeah, I think that's-- That's what I learned the most. I think that I can say doing one-to-one, I think is important as well.

You know, because Even if you are in a startup and you have to go fast and so on, sometimes just spending 10-15 minutes with the different person in the company, even if they are freelancers, is a good time. Thing to do because you speak about other stuff, you kind of then notice what they like, what they don't like. And that's where they kind of feel, I would say, the possibility to talk to you and to give you even feedback that you wouldn't have had.

Like in another, in a work environment, basically. And so I think it's really super important to take the time, like even if You don't have it, let's be clear, but you have to take it.

Mark: 32:55

Make time for the people in your team and give them the right attention. Exactly.

Yeah. Nice. Okay, I think then if we move back to the AI space, because of course, I think that's probably the most interesting space for a lot of people want to learn about. How do you guys work internally?

I mean, you're an AI first company. Do all of your engineers use AI tools like, I don't know, Replic or Cursor or something else in their coding? What's your experience been with those tools? Are they overhyped or? Are they actually useful for.

Mattieu: 33:28

You? Yes, I mean, you know, it's evolving so fast. That's crazy. But yeah, they are using, we were using at first GitHub compiler, right? But I would say you would see even... The developer we just hired in Luxembourg, he was using ChatGPT to get the code and to enhance his code, which is good. But it's not yet, for example, integrated in your IDE, right? Such as GitHub compiler, the cursor.

So there's really a big difference. My goal was really to give him some time to kind of test those new things and include it in its IDE. And so now, everyone is using either GitHub Copilot or Cursor in the team. But I know that and I've heard that cursor is I would say, better in a way, meaning that it's more efficient in providing more, I would say, useful code when we need it.

So that's why I'm kind of I told you know this new developer, okay now you take whatever time you need during a week, you watch videos, you kind of watch tutorial, you make yourself comfortable with cursor and we'll try GitHub Copilot later, you will see what's the best because even if I heard that and I'm kind of using it and I feel that it's this way, you never know. And like GitHub Copilot can catch up, you know, so you never know.

So that's why I'm kind of forcing, I would say, the new member of the team to use this or who didn't use it to use it. But... That's another level, yeah, that's for sure. I've heard in other startups where there was a developer who was not using it at a point, so those kind of texts, right? They were producing, I would say, hundreds of lines during the week. They kind of started using cursor and using it, used it in a proper way. And certainly it was like thousands of lines, you know, but Then it's another... Kind of type of functioning because you have to tell them to review the code. You are not in a I would say environment where you code everything, you have to review everything. And so that's kind of very different... But I feel that's the way to go. Definitely because it reduces definitely the time you spend on different features when you kind of control it in a nice way. But I would say, just to end with this, you know, in events that we attend, like Trachos and so on, we kind of learn a lot about those new tools and so on. And for example, for the last Copilot versions, they are now providing you a way to prompt on GitHub to add a new feature to your entire repo, then you have your agent that is like kind of working, can work, you know, hours and hours behind to produce it.

And then you can put your review. Directly in his code and then he's working again to produce, you know, the code according to your comments.

So it's really like, you don't have a developer anymore. You have just like the tech lead directly, you know, so that's, why I was saying you have kind of jobs that are evolving and all those kind of developers will have to switch and be more in a review mode, like kind of a tech lead mode than, you know, just developing.

Yeah.

Mark: 37:16

It's it means, I mean, trying different tool, right? And then cursor is better at one moment and the next minute get a copilot is better. It means your developers need to.

Yeah. Constantly switch or adjust or...

Mattieu: 37:29

Yeah, it's really about trying but and a At a point, you don't want your developers or the team to switch to tools and trying so many tools. That's the trap that you can have in AI, where there are so many tools or so many models that at a point you have just to... See, okay, does this tool, this model satisfy what I want to do with it for the moment and then Yeah, you kind of try later, but not like all day or all week long.

Yeah.

Mark: 38:00

Yeah. Is it good enough? Then it's good enough.

And then stick with it. Yes. And you're in the midst of the AI space. There's a lot of hype. You talked about the outbound tooling, we just talked about development tools. Where do you see the biggest gaps between the hype in AI and what is actual What's reality in AI?

Mattieu: 38:26

I think that the problem is that at a point, People feel it's magic, right? And they kind of ask everything, you know, and you're like, no, that does not do whatever you would like.

So, for example... We could have companies that are asking us, okay, I would like now to have my agent that is plugged to my whole SharePoint and be able to answer whatever question I want and so on. And you're like, no, like that's, you know, there is a model like behind. We have to know what type of data you have. We have to know what kind of scenarios you want to do with it and so on.

So that's still, I would say at the end of the game, it's a model, right? So you have to choose the right model. You have to choose the right.

I mean, scenarios for the company, you cannot do everything, right? In the company, you have to start with something.

So that's what we try to investigate first, if our platform is the right fit for them or not. Yeah, so I would say that's about telling that it's not Magic?

You know, it cannot answer everything and that you have to be careful with it and careful with the data you give it, right? Yeah.

Mark: 39:51

Yeah, and there's actual limitations in what you can and can't do.

Mattieu: 39:54

Yes. Yes. I would say that, yeah.

Mark: 39:59

And still, it's going incredibly fast. If we dream a little, you're really working on this communication between businesses and consumers or their clients at least. If you look at like five years in the future. Which is very far in the AI space. I got that five years ago. GPT wasn't even out. But if we look, let's say three to five years in the future, What do you think it's going to look like? Where do you think your services will be?

Mattieu: 40:26

So... My vision is always to have more, I would say, like... Helpdesk tool rather than a receptionist tool, right? Meaning that now we really provide a level one, I mean, service, right? In terms of telephone you ask, I would say that you answer the generic questions that the users might have or you transfer. But the vision of the product is really to have, I would say, helpdesk where you can speak with the end users, the clients about this specific file, this specific experience with the company, this specific project. But there you have a lot of things like that you have to think about, you know, refers to scale and have a possibility for the AI to have a conversation with each client, right? You have to authenticate them, you have to, I would say, have a lot of features on top of this and Even then, you have to provide, I would say, an easier way for the company to be able to integrate its data with the tool, right? It cannot maintain manually every file for every client, right? I would say, and in terms of model, you have to be able to ingest a lot of data and choose really the right data for the right client and so on.

So that's really, I would say... The end game that I see for this product where you kind of set up super fast with the CRM, the tools that you have in the company. You integrate them with our tool and then it's providing instantly an AI that can answer to whatever client is calling and just be responsive. Secured, right?

So we don't want a client to have the information from another one. So that's definitely a challenge there. But yeah, so that's really the end game I see in terms of telephony. But I would say in terms of having a conversation that is super realistic, Right? Because now, even if you see that the models and the text-to-speech of the voices are very good, you kind of still feel that you are talking with an AI, right? Not that we want to impersonate a human, we are always announcing that we are an AI, but I would say I think I would say that I will think that we are successful if one day we have a user that is calling our AI and it doesn't, I would say, it has a conversation like he would have with a normal human, right? And even maybe doing this Turing test, right? You just call and you cannot make the difference between the two. And I think even with the best products today, it's still possible to make the difference in terms of voice and telephony.

Mark: 43:21

Yeah. No, absolutely true. Even with things like ChatGPT Voice, it's You know, after the first two words, you know, it's an AI and it's not a human. Behind it. Okay, cool. Thank you very much. To wrap up, and this is our last question. You've learned a lot, I think, over the last couple of years. If you had to give some advice to the people that are listening or the biggest lesson that you've learned over the past, maybe the past year starting Elora. What is it that you would like to pass on?

Mattieu: 43:51

Sure. I mean, you know, I read a lot of books about startups, like the Lean Startup book, right, where they tell you to... Test a lot before launching your product. And I agree totally with this. That's the strategy you should have. But at a point you always... Come to a place where you are like, okay, Do I do the leap of faith and now I invest and I go for the product or not?

Right? You can test whatever you want.

Like, you're never sure it will work. I would say that, You have to test for sure, but if you feel that with your guts, with what you have seen so far, it's the product. And your product will work, then you have to go for it at the point, right? You cannot be always behind the... Do I have tested enough? Of course, you have to have it in your mind, but at a point you have just to go for it and There, the theory won't help you.

So it's really to kind of be sure of yourself, of your product and go for it. So that would be, I would say, my point because, you know, I always tried to tests.

So much the product and at a point I was like, yeah, but you know, like the theory, like here doesn't help me, what do I have to do? And so at a point you see that gradually you have to do this leap of faith and kind of Go for it. Of course like You have to have the funding for it if you want to develop your product. But at a point, yeah, it's the first step to go for if you feel that your product is the one.

Mark: 45:34

And to trust your gut feeling along the way as well. I Okay, thank you very much.

Mattieu: 45:38

Think intuition is a good... Big part, yeah.

Mark: 45:44

If people want to know more about Laura or yourself, where can they find you online?

Mattieu: 45:50

Yes, so our website is elora.lu, right? So E-L-O-R-A dot L-U, and you can just sign in and Try our platform. We have our own reception. It's done with Elora, right?

So on the website, you can just call it and talk with it. So it's possible via this website, yeah.

Mark: 46:13

Cool. Okay. We'll put the links in the show notes as well. And Mattieu, thank you very much. Thank you for being here.