David Shim: 0:00

I've been in the startup scene for about 20 years. I've never seen something this compressed. Smartphones were nothing compared to AI and the level of adoption and the mainstream adoption as well.

Joeri Billast: 0:10

Hello everyone and welcome to the Web3 CMO stories podcast. My name is Joeri Billast. I'm your podcast host. Today I'm excited because I'm a Web Summit and I'm joined by David. Hey David, how are you? I'm great. I'm excited to be here. Thanks for having me. Good to have you from Read AI, all from the US.

David Shim: 0:27

That's where you live? Yeah, live in the US, live in Seattle, Washington.

Joeri Billast: 0:31

Okay, guys. I live here. Actually, I'm Belgian, but I live here. And uh, you know, David from Read AI, I could not miss the opportunity to get to speak to him. David, you will dive straight in. Uh, Read AI. You call it the system of wreck of wreck of productivity. It's so what gap did you see in the way people work that's that convinced you this had to exist?

David Shim: 0:54

Yeah, I think the first thing is with meetings. We're in meetings every single day, uh, but it's all about memory. Or the notes that you write on a piece of paper, or maybe you write something on a notepad. And the problem that you run into with that is it's very ephemeral. It's almost like a Snapchat where it's like, I kind of remember it, but I have to, if I forget about it, if I have another meeting, it disappears. So you need that system of record for productivity to go in and take that meeting information and store it. You may or may not use it, but having that storage in place. And then if you stack those meetings, you can start to see you have meetings about AI, you have meetings about startups and start to contextualize that. Then if you add your emails, your messages, your files into that as well, whatever you're comfortable with, it'll start to get more context. So now you've got the system of record that not only has your meetings that are ephemeral, but it has all that static data to go in and say, these are the things that represent me or my organization and find more content that's relevant to me. So by having that system of record, it's really a storage of knowledge.

Joeri Billast: 1:49

I love that uh I learned something already about Sweet AI that you can do that, put it together because I use meeting notes of source too for myself and productivity. If not, I forget things. And sometimes you can also learn from it if you do a meeting. How can you be better in a meeting in a meeting scene? So I do it for that.

David Shim: 2:06

And the big thing there is like it's for you, but longer term it's multiplayer. So you want to be able to share the report with other people. So over half of our meeting reports now are shared with somebody other than the person who requested read. So because you don't want to create a report, send the meeting report, send the notes over. So people are getting more comfortable with that level of automation. And now imagine with that data where you've got all the knowledge that you have about AI that I have, if we combine those two things together, it can actually generate some unique content. And so that's where multiplayer comes in for AI. That's and that's what we're really excited about. It's like once you have a system of record for productivity, you could have those systems of record talk with one another and actually can produce content to generate insights. And that's something that might not have been available to either one of us as a standalone basis.

Joeri Billast: 2:51

Exactly. Uh, you may be heard I've written a book, but I was able to write the book because of my podcast conversations. I had them transcribed, and then I had so many knowledge from different people, and AI is really good in finding patterns in it, and so it helped me to create that.

David Shim: 3:08

And where we're seeing, like, so we've got uh two podcasters, one specifically uh in the US, they're a big podcaster, and Tiger Sisters. And what they're doing is they uploaded all of their podcasts, and now they actually let people actually access that fully available. So just like you used it for yourself. Now imagine if you flipped a switch and said, Any of my listeners, you want to ask me a question, you can take my entire knowledge of podcasts, all the files and blog posts that I've done, and now I can actually give you the answer. It might not be the perfect answer, but 80% of the time it's gonna be pretty good where people are like, oh, this is great. It's like I'm having a conversation.

Joeri Billast: 3:39

Yeah, and that's something uh I've been building also with AI because I every podcast is on a blog. Yep, but it's every it's different pieces. And I understand if you have one tool that does it, that's uh that's incredible. So uh let's talk a bit also about your background because you have been building and selling place to snap, then leading Foursquare, I remember that time, to profitability. What are the mindset shifts actually that did bring you to to found Reads AI?

David Shim: 4:09

I think it's about what are the outcomes that you're generating. So it's placed the look, it was a location analytics company that Snapchat acquired. Uh the focus there was to go in and say, hey, can we understand where people are in the physical world? And you might say, like, okay, that's that's a hard problem. But also, two, how do you get that location data without the user's permission? Uh what we did that was really novel was we just gave them money. We said, hey, I'll give you $5 a month if you just let me measure where you go in the physical world. And people can decide to accept that or not. But then we were able to build models on top of that to go in and say, if you hear an ad on Spotify, did you actually go to the business that was being advertised? But that was an outcome-based solution. It's going in and say, we were solving a problem that people didn't know if they were going into the store when they saw an ad. When we were at Snapchat, we were able to go in and build things where it able targeting. So being able to go in and say, don't copy people's data, but actually build a model against that to say, people who go to McDonald's, are they more or less likely to go to Burger King versus Walmart versus Costco and getting that information? So then you could build these lookalike models. Uh when we're at Foursquare, it was about actually being that ground truth, that system of record for location. So when you're able to kind of give people those outcomes, the adoption is incredibly high, even for things as scary as location. When you think about AI right now, people are still nervous about what are you gonna do with the right data? How do you do it? One, people have to opt in. So that's our goal is like everyone opts in. But when you opt in your data, the return is so much bigger. I like to use this example is if you use Waze or if you use Google Maps or Apple Maps, your data is going into a big database to say, is there traffic at this location? But you're okay with giving it and you're okay with actually listening to it because it saves you time. And so that's the value proposition here is like if AI can make you more productive, if it can save you time, people are going to be more open to adopting it and actually giving more and more data because they're gonna say, can I give you more? And that's the same with driving right now. Now we've gone from physical maps that people used to use 20 years ago, and that it wasn't that long ago, but 20 years ago, people were using physical maps to now we've got autonomous cars where they're driving us to the location where we say point A to point B.

Joeri Billast: 6:14

Absolutely. And it's you know, you at Read AI, you mentioned already, system of practice, a lot of raw data, a lot of data coming together, but how you move then to true understanding of the data? How does that work?

David Shim: 6:26

Yeah, so the understanding comes in where if you think about the outcomes, everything that we do at work, even in life, there's a certain outcome that comes into play. So you want to sell more of your product. You call clients, those clients respond, you send them a quote, they actually signed a contract. So there's these steps that occur. Well, all of that is essentially training data. So if you go and everyone talks about training data using YouTube data, using books, magazines, et cetera. Well, honestly, the best training data is who you are and what you do every single day. And so if you've got all that date system of record digitized, then you can actually apply models against that to actually drive outcomes where it's customized to you. So a lot of it is really about your training your data sets and actually enabling that. But there's a you have to be comfortable. Right now it's generic models where we're going towards it's your personalized model. And people are gonna get more comfortable. Because let me give you one example here. Let's say you're on vacation, let's say you have a child, you go on paternity leave or maternity leave. A lot of people are worried about their jobs when they go on paternity and maternity. Hey, people are gonna forget about me, the products that I'm working on are gonna get stalled out, they're gonna slow down. Perfect world. You can continue to work, but also have your child. How do you do that? Imagine if there was a concept called the digital twin where it access your store system of record for meetings, for productivity of intelligence, and it can fill in while you're gone. Hey, why did we sign this contract? Hey, what was the customer feedback the last time around? And rather than waiting month, two months, three months for you to come back, the project continues to move forward. By the time you come back, it's further along and you can step right into it.

Joeri Billast: 7:58

Yeah, that sounds incredible. Imagine that of people then maybe scared that this that this AI can take over their work, or maybe I always say it's assisting you, it's not replacing you. 100%.

David Shim: 8:09

It's assisting you and it's amplifying. Yeah. So it's taking that ground truth, which is you, and it's saying, where else can I put myself? Can I put myself in multiple places so I can do more of the jobs, then I can actually spend more time with my family, my friends, my hobbies, because I'm able to actually use AI to amplify myself.

Joeri Billast: 8:27

Yeah. Another question, if I mentioned I'm Belgian, I speak a couple of languages. English is not my native language. Yes. So I speak a lot of Dutch, French, English, German, sometimes Portuguese. So meetings are often in different languages than English. So um, and then this can be a problem. I'm wondering how does Read AI uh work with that? How is it is it transformed into kind of different sources or is it translated to in kind of sort of intelligence? How does it work?

David Shim: 8:58

Yeah, so we support 22 plus languages. So all the languages you mentioned, we actually support today. So if you use it, we'll automatically detect it. So what we figured out was people don't want to check a box for a setting because if you're talking in four different languages, you'd have to go in and change the setting every single time. So we'll automatically detect the dominant language in the conversation and we'll use that as the primary language. So when we create the meeting notes, the summaries, the action items, we'll pick that up. Uh, we we actually have measured different cultures where we can understand uh are people more hand movie? Like I'm moving my hands a lot around a lot, but we're actually one of the number one meeting note takers in Brazil. And there, people are very emotional. Everyone's super happy, they're talking in a positive tone. Now that they're not really happy, but they look like they're happy, they're they're excited, etc. So we actually had to build new models for different markets where we were able to go in and say the baseline of the US is 75, but in the Brazil, if I use that same model, it looks like everybody's excited in the call. So I need to build a new baseline model for Brazil versus Belgium versus France versus Germany. You always want to adjust for the cultures. And so the sentiment is measured. Uh yeah, so we'll measure sentiment two ways. Not just by the words that were said, but based on how you react to the words. Like, so we call it a multimodal model that goes in and says, if you're nodding your head right now. So our models will pick up David said this, and you actually nodded your head. So now we're able to go and say significance goes up.

Joeri Billast: 10:20

Okay, right.

David Shim: 10:21

Versus like, let's say I'm trying to sell you something and you get really annoyed and you're like, oh, not this again. And you start looking around, your face frowns a little bit. We don't fingerprint your face, but we look at where are your eyes. Are you smiling? Are you not smiling? Are you looking away or not? And we've built models to detect things like second monitor. So you talk to your camera here, but then you look at your big monitor over here. So we've built all these things to understand are you paying attention or are you happy?

Joeri Billast: 10:45

Are you looking at your phone or doing something else? I love like I've seen these uh sometimes I tend it uh, you know, as a third party uh uh conversation between two other people and I see the body language. You see it better when you form a distance when you are in the conversation itself, sometimes it's a bit harder.

David Shim: 11:01

100%. And that's we call it the narration layer. So right now, if you think about meeting notes, a lot of it's just transcriptions and then it summarizes the chat GPT. But what's missing is how did people say it? If you ask me a really tough question, I started sweating and talking really fast. The transcript would only say David answered that question really well, versus like if I actually have a narration layer to say David was nervous, he went from 150 words to 225 words per minute. Uh, he started to say, and um uh like uh the filler words more. We'll pick those up and we'll actually adjust the actual scores and also the continent.

Joeri Billast: 11:34

Well, this sounds actually an extra added layer that you add to a conversation because I also did it. You take a transcript, you put it into ChatGPT, and you ask it, you know, get out, or do I get out? But it gives some feedback, but it doesn't see the sentiments, of course.

David Shim: 11:50

No, and we've had like uh people that have autism, they're actually use this pretty heavily because they can't understand the reactions in real time because they're kind of going into a talk about X, Y, and Z. They're talking about a specific product, a feature, etc. But they want to know and how to deliver that message better. So they'll look at the coaching afterwards and they'll actually look at sentiment and engagement, not for them, but for the audience. So it actually helps them learn and say, you know what, I talked for 10 minutes straight, the engagement went down. When I only talk for two minutes, it goes up.

Joeri Billast: 12:19

I love it to do that because I already do this for my podcast, of course, to see you know how long stay people and you know what at what moment they lose attention. Now, um, yeah, with all the data that you already gathered, uh, what has been the most surprising, surprising insights that you have seen in the data so far?

David Shim: 12:36

Uh I think how quickly if people have adopted it. So not just read AI, but AI in general. Like I've been in the startup scene for about 20 years. I've never seen something this compressed. Smartphones were nothing compared to AI and the level of adoption and the mainstream adoption as well. So this is where we have one to two percent of Columbia's population using RAID on a weekly basis. Whoa. And it was college students that members measured every single class, they started to take it to work. Now we've got, you know, enterprises there, government agencies, et cetera, reaching out to us and saying we want to get set up. And it's because it gives them value right away. They don't need to know how the models work, they don't really care that much. They just want to know, like, hey, I don't have to take meeting notes. That's a great solution. Hey, you're gonna remind me when I need to follow up on something because I don't have an assistant, but now this AI is an assistant for me. Hey, I'm able to follow up on a question that I didn't know to answer for because the AI is actually able to say after the call, hey David, this is what you should have said, and you should follow up with this. So people are actually seeing real value all the way from like tech people to social workers out in the field that they're recording conversations with their patients and then going back to the office and the notes are all done.

Joeri Billast: 13:42

Yeah. And it starts, of course, with one meeting, but then at the end of the of the year, or you have so many meetings, so many knowledge, which is gathered in there, which is Oh, yeah.

David Shim: 13:51

And we've got like ad agencies that are using it today where they go in. So if you've got an ad agency, you have a team, that team interacts with the client. Uh, they actually are looking at sentiment and engagement for the client now to say, hey, last three weeks, it hasn't been going well. The client's been really annoyed. Your reports might not tell you, but you're like, hey, maybe I need to sit down on that call, help out a little bit. What can we do to make it better? So it's not to get people in trouble, but it's actually to catch issues before they actually happen. So to go in and say, hey, client's not happy, let me join the call. Oh, you're unhappy about this, let me solve that for you. And now the client sticks around and they spend more money versus they cancel the service.

Joeri Billast: 14:27

So many things happen in AI, it goes so fast. Um, if you look a bit into the future, how do you see companies evolve like intelligence that you see that we are getting today from the I think anybody who doesn't have a storage of intelligence plan from an enterprise perspective is going to run into significant issues, both publicly traded markets where companies are going to be, why is this knowledge not a moat for you?

David Shim: 14:51

Why is this not a differentiator? I think that's gonna be a big question that will come up. Like, are you storing all this information that comes into play? And then two, I think is the digital twin. I think there will right now you've got the storage of intelligence, but how do you apply it? How do you actually put it to action, put it to work? And that storage of intelligence needs a digital twin to actually represent it. So if you or I, let's say go on maternity, paternity, vacation, your digital twin can step in and cover for you. But longer term, as the enterprise continues to grow, it goes in and says, Hey, welcome to the company. I've assigned you a digital twin that has all the tribal knowledge, all the historical knowledge that in the old days we wouldn't record, but we would just train each other one-on-one. Now that's immediately available to you and can be your sidekick. And so I think what you're gonna see is significant ramp up times uh sped up. And I think employees are gonna love it because it's like, hey, new job. I don't have to spend three to six months to learn how everything works or who to talk to. My co-pilot, my twin, is going to actually accelerate things. And I think that is something where everyone talks about agents today and AI agents. Those are good, but the problem is they're not solving specific problems. The digital twin takes the knowledge that's available and helps you solve specific problems for Octa.

Joeri Billast: 16:00

Amazing. Now you you see really excited about your what you're doing what you're building, really passionate. If there is one thing I would ask you, what is the one thing that you're now the most excited about today in a thing that you're building? Uh what would it be or in AI?

David Shim: 16:16

Yeah, I I think it's really that digital twin concept of like making that applicable to the general population. Because I believe that digital twin, whenever that comes out for consumers, will get rapid adoption, if not faster than an open AI, a cloud anthropic, where it's solving a real problem and it's going in, it's gonna instantly work. And it's not gonna change your workflows, but it's gonna learn where you work, where you communicate, and it's gonna participate in that area. So it's just gonna fit right in. So you don't need an AJ, you don't need a train A. All the things that we talk about today, that's gonna be automated where it's just plug and play.

Joeri Billast: 16:51

Amazing. Now, guys, we are here at WebSmith and I know our time is limited. Luckily, David, you are really passionate and you can uh talk very fast, so that's really good. Now, at the end of this podcast episode, people, I guess they aren't excited about everything that AI delivers about meeting us, but you are telling where can people find you or how can they sign up for readai.

David Shim: 17:15

Yeah, so readai is a free service. So read.ai. So read.ai, uh you sign up, create an account, use Google, Microsoft, email address, and then you create an account. You're live within five minutes. Uh it can join whenever your next meeting is, and you can see how that works. Then you can decide to connect your email, your Slack, your Teams, your files. And the more we made it so you don't have to deal with that all up front. So you can get comfortable with it and then start to add more data and see the value. But super simple to use. We've got a free plan that 80% of users it works for them because not everyone has meetings every single day.

Joeri Billast: 17:48

No, yeah, I'll have a lot of meetings, a lot of podcasts too. So, yes, I can try it, you know, upload our conversation and see see what comes out of that. Uh well, David, it was a real pleasure to have you on the show. This is great. I really appreciate the time. This is a lot of fun. Guys, this was a really short conversation, but I this I know people around you, a lot of people that have their business or that work in a company, they can benefit from this episode. So be sure to share this episode with them. If you're not following the show, this is a really good moment to hit the subscribe button. If you haven't given me a review yet, if you give me these five stars, it will help me reach an even bigger audience. And of course, I would love to see you back next time. Take care.