The following is a rough transcript which has not been revised by High Signal or the guest. Please check with us before using any quotations from this transcript. Thank you.

===

hugo: [00:00:00] Hi, I'm Hugo Bowne Anderson, and welcome to High Signal. The goal of this podcast is to help you advance your careers in data science, machine learning, and AI, by speaking with experts in the field. Today, I'm speaking with Gabriel Weintraub, the Amund Professor of Operations, Information and Technology at Stanford Graduate School of Business.

Gabriel is an expert in data science operations and market design with deep experience studying how digital platforms like Uber, Mercado Libre and Airbnb reduce friction and optimize matches He was also part of the Chilean team that won the prestigious Edelman Prize in 2022 for improving COVID 19 response strategies through analytics.

In our conversation today, we dive into the challenges and opportunities for companies to adopt data driven strategies, focusing on how non tech native organizations can build the foundations needed to leverage data science and AI. 

Gabriel [00:01:00] discusses the gaps he's observed between C level leadership and technical teams, the importance of starting with practical, high ROI projects, and fostering a culture of experimentation where even negative results drive learning and progress.

Although the challenges in Latin America are shaped by its unique context, they reflect broader struggles faced by non tech native organizations globally. Gabriel's insights offer valuable lessons for companies across industries in both developing and developed economies, from legacy businesses in the U.

S. to startups in Chile. We explore how generative AI is reshaping innovation by lowering technical barriers, enabling smaller teams and less technical users to create impactful solutions. Additionally, Gabriel shares his perspectives on the AI and data science landscape in Latin America, highlighting the region's immense talent pool and the need for greater investment in corporate R& D.

He discusses the role of local innovation systems in fostering startups, the potential of domain specific language models [00:02:00] like a Chilean LLM, for example, and how AI can address critical challenges in areas like government, education, and healthcare. This episode is a mix of practical strategies for organizations and reflections on the global implications of AI and data science. Before we jump into the full episode, here are three clips that highlight some of Gabriel's most compelling insights.

To start, Gabriel highlights the foundational steps organizations need to take before diving into AI or machine learning. This includes building basic data pipelines and focusing on simple high ROI projects. Here's what he said.

gabriel: there's this gap that I see between the C level and the people on the ground trying to implement these tools, where like the C level doesn't really understand what they need to do, what's the value,of using data and then just things get lost in, in, in translation.

the other one is.what is the starting point? I think sometimes you talk with [00:03:00] executives and tell you, well, I need to, we need AI. We need AI. And then you ask, basic questions. It's like, where is your data? If I ask you a simple,question about, say, your sales, can you give me, can you access data to give me a quick answer?

And in many cases, the answer is no. So it's just like very hard to think about building like machine learning models, like AI models, if you don't have the basic data structure in place, like the pipelines in place and understanding that and that this is going to be a journey that there's, you know, you need to start with the basics.

I think that's that's really important, and I think it's many times in the discussion. It gets overlooked,start with the basics and the basics are typically, building the pipelines of data pipelines. You cannot analyze data if you don't have structured data. Um,and maybe let me add a third thing, which is, think it's important to start with like low hanging fruit.

hugo: [00:04:00] Next, Gabriel discusses how data teams must be integrated into the business rather than working in isolation. Solving impactful problems requires collaboration between data leaders and business leaders. Take a listen. 

gabriel: And I think data teams like data. If there's a data leadership, I think they also play a role because.

When you have a technical team, there's a tendency to solve like interesting technical problems. And maybe those interesting technical problems aren't the most relevant for the business. So I think there's a responsibility also for the data team to engage with the business, to understand the business, to find partners.

business partners. So they, so that they actually solve important business problems that they're not solving some like second order, technical challenge, but they're really involved in the business and we, there's like a bunch of things we can also follow up at a more like tactical level that I think are also relevant.

hugo: Finally, Gabriel explains why [00:05:00] experimentation is crucial for any data driven culture and how even negative results are valuable for driving learning and progress.

Here's what he had to say.

gabriel: It's incredibly hard in my experience to convince people that experimentation is important, and that running an experiment that has a flat or negative result is actually helpful because you are learning that what you thought was a good idea is actually not a good idea. And I found, in my own experience, again, in companies that maybe were not 

born data driven,with an experimentation culture, which is like now very typical in Silicon Valley, of course, many tech companies in general,to see the value of experimenting, and again, then it falls to, who has the. The strongest voice who says, who in the meeting is more aggressive.

And, you shouldn't make business decisions like that. You should tell, you should let the data speak. But I think that's, that's another big challenge. And going back to your initial question, I agree with you. I think there's like,More [00:06:00] like maybe low hanging analytics, an analytics function that could be,if a company doesn't have it,that by itself, having dashboards, just the basic analytics that by itself would be, incredibly valuable.

hugo: Those are just a few highlights from my conversation with Gabriel. 

Before we jump into the full episode with Gabriel, let's take a quick moment to check in with the fine folk at Delphina who helped to make High Signal possible.

Thanks for joining me here, Duncan, it's great to have you here, and I can't thank you enough for all the support you've provided to make this podcast happen. can you tell us a little bit about Delphina and why you're supporting a podcast focused on AI and machine learning?

duncan: Yeah, at Delphina, we are building AI powered tools for data scientists, and we come across a lot of very skilled practitioners in our work. And so with the podcast, we're looking to identify and share the pi signal.

hugo: Awesome. And we've just shown a few clips from, my [00:07:00] conversation with Gabriel and I'm wondering what resonated, in there with you?

duncan: Wow. There is so much goodness in the episode with Gabriel. He's such a deep and eloquent thinker.the thing that I wanted to highlight was his discussion around the gap in expectations from really the C level kind of operators to the folks on the ground. And I think a lot of this is because the power of data and data science has grown 10x, 100x in the last decade.

And most leaders can't keep up. and so my mental framework for this is the levels of data literacy, and there's a level one in my mind where you're first getting data organized, getting the basic analytics off the ground. Level two is starting to show value from,predictive analytics and machine learning, basic forecasting, customer segmentation.

Level three is where you start to multiply the use cases. You have maybe real time use cases for the first time [00:08:00] delivering value in the product. And level four is this kind of self learning systems and bandit style frameworks, reinforcement learning live. And the reality is that is a multi year journey, even for the best and most advanced teams, as you need to evolve the skill set of folks on the team, the data engineering needs, the ML infra needs, the business understanding.

And yet, I think the mistake that many executives make who aren't exposed to enough of this stuff, and frankly, many data leaders make as well, is they try to jump too far ahead too fast. And try to expect themselves to operate a level four autonomy When they're really actually just learning to crawl and walk and so I think you need to be honest with yourself About where you are and how you build the momentum of the freight train And not operate a herky jerky go kart as you try to scale the levels.

hugo: I couldn't agree more. And I'm so glad that was your series of takeaways from it. Because I think a key point was when executives [00:09:00] and stakeholders come and say, Hey, what's our generative AI strategy? Gabriel made the point, Hey, do we have access to particular types of data to even begin this? So it is this hierarchy of needs.

And I love how you framed it as these differing levels. so perhaps without further ado, we should jump into the conversation...

friends, viewers, listeners, just a quick note that if you enjoy this podcast, please do share it with friends and colleagues. If you're watching it on YouTube, please do like and subscribe. If you're listening to it on an app, give us five stars and write us a review as well. That will help us keep on releasing the podcast you love.

Now let's jump in. 

hugo: Hi there, Gabrielle, and welcome to the show. 

gabriel: Great to be here.

hugo: So great to have you here. I'm so excited to jump in, and talk about the types of things you've taught with respect to data science and how it plays into our conceptions of how data machine learning AI work more generally. So as we'll get to, you're in Chile, currently, again, but you've taught courses on data science in a lot of places, including, [00:10:00] Stanford on, on, on platforms, online marketplaces, these types of things.

So I'm wondering if you could walk us through how. US based platforms, multinationals like Uber, for example, and Mercado Libre in Latin America, leverage data science. And why these are powerful examples of data driven organizations.

gabriel: Sure. Sure. Yeah, so that's right. I'm in Chile now. I'm on leave from Stanford but At Stanford, I teach, classes as you mentioned on data science and particular data science for platforms, for online markets. and the way I think about the role of data science in, in, in platforms has to do with, what is the business model for two sided markets?

What are they selling? So I think of two sided markets as selling the reduction of frictions. If you think about like Airbnb, they're not really selling lodging. They're selling a reduction of friction. And what do I mean by that? An important friction, is search and information. Like how would you ever know that there was, I think [00:11:00] you're in Berlin now, a one bedroom, a nice apartment in Dowtown Berlin for you to rent.

And that's what Airbnb is selling you.the search costs. It's also reducing information costs. You can look at the apartment and see the quality, try to infer the quality and so on. And ultimately this reduction, enables matches, enables, matches between buyers and sellers in this case,guests and hosts.

and if you think at a very high level, what a, marketplace, a two sided marketplace is doing, is enabling, it's enabling matches between buyers and sellers. Then it's learning from this matches, it's learning from market participants to further improve this matching process.

Like a very high level, you can think of a marketplace, going through this. Through this loop. Now, when there's this matches, as I said before, this platforms collect data, to further improve matches. How do they do it? They will, use this data to change their ranking algorithm.

What they show you. [00:12:00] That's it's very critical for MercadoLibre, you were mentioning for Airbnb and others,to increase, conversions, increase matches, they will decide what information to share with you, what kind of review system, they, they will build. And in all of this data science is key.

how do you use data to improve the product, enhance the functionalities, with the ultimate goal of, increases, increasing matches and quality, the quality of the matches. 

hugo: I really appreciate all of that context. And I do think, The establishment of online marketplaces is one of the most important and matching use of matching algorithms is one of the most important aspects of, data and the online, environment more generally, even getting the right information in the hands of the right people is, a matching problem.

And I think, I recently interviewed. Michael Jordan from Berkeley for the podcast. And we, talked a bunch about Amazon actually, and how they were one of the really first early examples of really helping reducing friction for [00:13:00] between, supply and demand of, originally book sales. I,

gabriel: Yeah,

hugo: I am interested in what you've been seeing in, Latin America and Chile in particular, since you've been there.

And we discussed this last time we spoke, but I think, I have a friend who says not everyone needs to do, and he calls it jokingly, Jeff Dean machine learning. He's not everyone has to, be the Google and these FAANG companies and that type of stuff. And so I want to approach the conversation about what we may call reasonable scale, machine learning, reasonable scale, AI, how a lot of different organizations who may not be Google.

Can adopt these tools and I actually think you having been, Return to Chile recently, you have a unique perspective in people. I have conversations with, with respect to challenges in developing markets and people adopting, data driven machine learning, driven AI driven mindset. you've observed that some companies are behind.

Quote unquote behind in the use of data in some sense. I'm wondering what this actually means in practice and what do you think some of the biggest barriers for business businesses in developing [00:14:00] markets are when it comes to adopting data and artificial intelligence? 

gabriel: yeah. And I think that I love that question. whenI arrived here, I'm from Chile, but, recently we've been spending some time after not spending significant amount of time and like after 20 years in the U. S. So it was like a big, initially I observed is like huge gap between what I was just discussing and how data is used in this companies like Uber, Amazon, Airbnb, and what I'm seeing in Chile, even in large corporations.

Okay. These are like big corporations. and I think one, I think there's like a few big differences and barriers. One is most of these corporations, are not. Tech companies. They were not born as tech companies. they're not like data science native. It's not something they started doing from the beginning.

And now maybe five years ago, 10 years ago, they started realizing. oh, shoot,the world is moving in that direction. We need to do something, otherwise the Amazon of the world [00:15:00] are going to eat us. and I think I see a lot of challenges like there's this, transition period for these companies and there's, technical constraints, finding the right talent.

But I think maybe the main challenge that I've seen. really thinking of data as a key component of a company strategy and to build products. And that's really, that's, it's related to changing the culture. of a company. It's like a big cultural change. I think a lot of these large corporations are quite conservative.

They have a lot of inertia. Sometimes they operate in markets that aren't super competitive. so they get complacent. and they, a lot of the decision making is the highest paid person in the room. Uh,the, the, maybe the person has it's being the longest in the company. and in data, driven organizations.

hugo: It's not like that. Like you run experiments, you look at the data and making that shift. I [00:16:00] think it's a huge challenge. that's that has been probably my main observation, which has to do with the culture, how to build a data culture in companies that were not born.as we with data, in mind to make decisions and to build products, I love it. I love that you framed it in terms of, non data native companies or companies that weren't born in the quote unquote data revolution for lack of a better term. That was my term, not yours. I do wonder. How this relates to companies in the U S that have, be predated. this entirely new mindset.

So I can give you an example. Years ago, I spoke with, an executive at John Deere, trying to convince them, certain things around adopting data strategy, this type of stuff. And his response was, and he was. Being a contrarian, but his response was we've been operating for years without data, Hugo, without the type of data you're talking about.

And we're doing very well as a [00:17:00] business. Why would we change that? so I think there's even in the U S, outside of tech companies, there's a slow adoption mindset. So I suppose my question is, how do you think what you've seen in Chile and Latin America relates to non tech companies?incumbent and non traditionally data driven organizations in the U.

S.

gabriel: that's a great question. So I think a basic question, a company needs to ask isif you think about this, like data revolution as a meteor, As are you like in the middle, is the meteor going to crash? Cross you. Are you going to be, it will touch you or is it more farther away?

And I think that really depends. In which industry you are some industries, maybe like data, at least maybe for now, played like not a core critical role and in others. it does. So I think that's the 1st question The world will move there. The question is, at what speed in the industry you are.

It's like, how urgent is this? And given that, I think, suppose you are in an [00:18:00] industry where data is becoming increasingly relevant, and maybe there's like startups coming up, perhaps are doing things much better because, they use data. So you either, like react or peril, right? You need to do, something to compete.

And, and I think that's, I think I see a parallel, with some of what I've seen in Chile, some of the,concerns you were mentioning, Hugo, if this company, the company has, is. is very conservative and it's in the path of the meteor, it will probably not survive. and one of the things that maybe is more particular, more relevant now with this, all this now AI revolution, where there's so much hype, is to really ask yourself,what is the problem I'm actually solving?

when there's all this like technological hype, It's very easy to fall in love with the tech,and forget that you're still solving a problem and the tech is a vehicle to solve that problem. And that's something I see both in the U. S. and in Chile. There's like friends say, okay, we need to do something.

We need to do something without really understanding,how AI [00:19:00] can help solve our business problems, like the pain of our customers. so I think in this path, us. Companies that maybe were not born either were not born, doing data science, keeping that in mind. I think it's critical. 

hugo: totally agree. And I'm interested in performing some form of, for lack of a better term, gradient ascent. Not gradient de descent. But, what I want to do is, imagine that, whether we're a company in Chile or one in, in North America or somewhere in the West that hasn't necessarily, isn't a FAANG company, hasn't been data native since the beginning.

what type of initiatives do you think data leaders or leaders in these organizations can take to establish more data driven cultures, and get to where we'd like to be? 

gabriel: that's a great and complex question. I think there's so many dimensions, to that question. And I think probably like the first thing which is necessary is to havea [00:20:00] leadership team that believes in this. I think it's it's just hard. to build a data culture, where, maybe you have a data team in, in, some office somewhere, and it's doing some of these like obscure things that no one understands, and it's not really like in the table, but say the C level is not really making.

business,involved in important businesses. So I guess I'm going back. We can also talk about like more tactical things. But first I'm going back to this

hugo: No, I think the cultural things are incredibly important.

gabriel: think it's incredibly important and it's necessary. And I think data teams like data. If there's a data leadership, I think they also play a role because.

When you have a technical team, there's a tendency to solve like interesting technical problems. And maybe those interesting technical problems aren't the most relevant for the business. So I think there's a responsibility also for the data team to engage with the business, to understand [00:21:00] the business, to find partners.

hugo: business partners. So they, so that they actually solve important business problems that they're not solving some like second order, technical challenge, but they're really involved in the business and we, there's like a bunch of things we can also follow up at a more like tactical level that I think are also relevant. Great. maybe we can follow up with that, but take a slightly different tangent on it in the sense that I did want to ask if you've noticed and the answer may be no, but if you've noticed any unique challenges in developing economies, like Latin America. So my, I think my specific question is.

gabriel: from your experience in Latin America, where do companies face the most difficulty when trying to build a data driven culture? Is it more technical, tactical, cultural, or organizational hurdles? I think probably all of the above. and maybe I'll let me just emphasize like. Two points. one is related to what we've been discussing, that sometimes there's [00:22:00] this gap that I see between the C level and the people on the ground trying to implement these tools, where like the C level doesn't really understand what they need to do, what's the value,of using data and then just things get lost in, in, in translation.

the other one is.what is the starting point? I think sometimes you talk with executives and tell you, well, I need to, we need AI. We need AI. And then you ask, basic questions. It's like, where is your data? If I ask you a simple,question about, say, your sales, can you give me, can you access data to give me a quick answer?

And in many cases, the answer is no. So it's just like very hard to think about building like machine learning models, like AI models, if you don't have the basic data structure in place, like the pipelines in place and understanding that and that this is going to be a journey that there's, you know, you need to start with the basics.

I [00:23:00] think that's that's really important, and I think it's many times in the discussion. It gets overlooked,start with the basics and the basics are typically, building the pipelines of data pipelines. You cannot analyze data if you don't have structured data. Um,and maybe let me add a third thing, which is, think it's important to start with like low hanging fruit.

With simple things. there's now we have all these AI tools that are off the shelf, and there are actually simple things that maybe with not too sophisticated data engineering, you can make a lot of progress. for example, if you think about streamlining, processes, reducing costs, speeding up some tasks, there's a lot of things you can do with this, AI tools, that aren't Very hard to implement, given all the tools we have and could have a, a big impact.

so maybe I would summarize it with,having a clear leadership that understands the value starting with the basics, including, structuring your data, and also, start with like high ROI [00:24:00] projects, something that is simple and potentially high impact.

hugo: Yeah, I do think all those things are incredibly important. I'm glad you mentioned the last point because I've seen companies try to Essentially AI data driven ML capabilities and tried to go with moonshots initially and failed at them. And what that does culturally is no good at all. You want to know, you want to start with something you can deliver in a short timeframe that can deliver value to the organization. And.

Essentially across the board. I also liked the point that you made on one end where, and this comes back to something I'll share in the show notes, what Monaco Rogati built the data science, the AI hierarchy of needs, which as we've discussed before, it's like Maslow's hierarchy of needs for those who haven't seen it, but AI and machine learning are at the top and like ETL is at the bottom.

So the whole. Her whole point was trying to get a pithy summary to let people know you gotta learn to store, manage, transform, and count [00:25:00] before doing anything else. But, what She did that before we had APIs as well. So now there's this interesting dance happening where it seems like maybe we can build more things more easily.

but of course, once you see the cool generative AI demos, you realize you then need to build, get them in the software development life cycle and get them. Integrate into your entire infrastructure stack. So I'm wondering what changes here for you and for things you're seeing in Chile. Now that generative AI has exploded. 

gabriel: so I agree with you. I think we have tools. first, if you think about like copilots where the barriers to entry to develop, some of these things have gone down and potentially, you can build cool, cool products with like much smaller teams. and I thinkthat's something that.

could, should help, in this transition to becoming more data driven. it's still, I, given, even though the tools [00:26:00] are there, I think there's all these other frictions that we've been discussing that, needs to get, resolved, before, before that happens. But I do think, There's like a, there's a huge opportunity, by using these tools where, you know, someone that perhaps with not such a technical background can make, like significant, like reasonable progress on, solving a technical challenge.

and I think that. That should be definitely helpful. it's also, I think having, you know, you mentioned what is the cost of, investing in, in, in this projects and not having a win. and I totally agree with you in organizations. And I've seen this in many places, organizations that are,they don't have a culture of data showing quick wins is key.

More in most of these organizations. Most people do not understand the value. They do not see the value. And for people I ask that have worked on these things and see the value, have experienced the value,it's, we don't, we shouldn't give that for granted. We need to convince. [00:27:00] the organization that this, tools,are valuable and, having quick wins is key.

And perhaps with this gen AI tools, it may be easier to have quick wins where, you know, maybe you don't have, you don't need such a long cycle, to build something that, that, that's helpful. and I think companies should definitely leverage. This off the shelf tools, before doing this, like super ambitious, like this moonshot projects that you were, you were also talking about.

hugo: Absolutely. Some of the best initiatives I've seen early on for people adopting, more data driven cultures is Just lead qualifying for your sales team. So on Monday they get an automated email, which is this is the spreadsheet of people to reach out to in the next three hours. And you see higher conversions that way.

Also, I am interested in this is slightly tangential. The ability to use generative AI in a less generative sense, but in context, learning and non technical people to be able to speak with it in natural language to build machine learning. Machine learning models. I am interested though. So for [00:28:00] companies in places like Chile that are beginning to adopt machine learning and data science, as well as companies in North America that are beginning to adopt these types of things, where do you think they'll see the most immediate return on investment?

Would it be in dashboards for decision making or automating decision making processes? 

gabriel: that's a good question. I Think it really depends on the industry, and in which, going back to this hierarchy of needs like where the industry is there's some industry. I don't know think of Construction. That's, an industry that hasn't evolved, like the technology hasn't evolved that much, and I think that industry, like very basic things will have a huge ROI, just like managing inventories, managing purchasing processes.

And there's also off the shelf tools for this and more and more companies are adopting this. I think that just that,you can think about more sophisticated things, but just that will have an impact. Now, if you think about like more customer [00:29:00] facing,digital experiences, if you think about like retailing or banking. I think those, companies need to go a step further and really use data to improve products and not, and perhaps I'll streamlining the process, but improve products, improve the interaction with customers, the offerings following a customer journey and see whether they can improve it.

hugo: what tech companies, do and. I think that's actually existential for those companies. If I think about big retailers in Chile, banks in Chile, if they don't, you're asking about ROI. And I think it's just more than like large ROI. I think this is existential, if they don't move in that direction where they're using data to build products, as we were discussing at the very beginning, your first question is how to use data, like in online marketplaces, I think it's just going to be hard for them to survive. So this is incredibly important, right? And [00:30:00] incredibly important to you. I can't imagine the journey in being from Chile, going to America, doing all the work you've done and returning and seeing this, this gap, which is a technical gap, organizational gap, skills gap. but this gap between countries in the global South, like Chile and countries in the U S in terms of AI and data science capabilities, particularly, Now we live in a globally connected, world, where big tech companies, they don't stay in Silicon Valley.

gabriel: They make pretty serious plays in a lot of, we don't need to get into the nitty gritty of, of those tactics. But what steps do you think companies in developing,markets can take to start closing that gap. yeah.I think, we, touch upon, a lot of these things, but,if it were to, summarize some of these ideas and what would be maybe more advice thinking of a path, to closing this gap to building, A company that uses, data, to [00:31:00] improve their business, again, like I, I don't want to be too insistent, but I think it really starts with a vision, like commitment of leadership and really think that this is like a long journey.

Okay, this is going to be, if you think about companies that have done this successfully in the US, think of Walmart. But I at some point realized, okay, we need to become like omni channel. We need to have a digital experience, that it's also integrated with the physical experience. That was like a long journey, and you need a commitment to do that.

and then, start with the basics. You don't start with,the most sophisticated things and the basics, as you were saying, it could be, collecting, data in a smart way in an Excel file, which is, what are your like largest customers? What may be the customers that will have the higher, highest chance of, converting.

and and for some of these projects, you may not need Super sophisticated like data engineering, and I would start with those, like start with like high ROI projects, the low hanging fruit, show quick wins, [00:32:00] and then move to, to more sophisticated,data engineering, data products, where again, you can also speed them up with the use of, off the shelf tools,but I think companies really need to understand that In at least in some industries, this is going to be existential, companies that have that vision, have a leadership, have a much higher chance of surviving,and really, I think what's also key, some of these companies have already invested a tremendous amount of money in what's called like digital transformation they just haven't been successful.

it just has not paid off. and I think it goes back to this idea of having the silos, it's this people doing, technical staff, data staff that is not really integrated with the business. And it just doesn't, it just doesn't make an impact. Into break those silos, bring those technical people inside the business, engage them with business problems.

and let me just mention maybe one more thing is, [00:33:00] it's very common also in these companies that internal organizations own their data. And they don't, it's not like a culture of like open data and they, that's what makes internal, some of these internal orgs valuable is, okay, this is my data.

And if someone else would come and ask you, okay, I need that piece of data. I said, no, well, I'm not sure I can share it, which like makes absolutely no sense, so that there's like a repository of all the data in the company, and these data sets don't talk to each other. So also like building a culture of Oh, of Openness of open data.

I think it's also critical,

hugo: I couldn't agree more. And I actually think open data and, building internal platforms that allow people to leverage all the, because all the data you can have. And then, it was at places like Uber where they really started revolutionizing, not on just data stores, but feature stores, They realized people in different departments were generating the same features or similar features all the time. So making sure they were, Accessible to [00:34:00] everyone. And I do think this idea of making sure the data function isn't siloed as we're both seen across a lot of tech companies. One of, there are pros and cons, right?

But one of the. I suppose most effective models has been the data science in team embedded across different organizations, right? Where they're like, you'll have a data scientist as a, on the growth marketing team, one on product, whatever it may be one with sales, but they all have dotted lines to a centralized data team as well.

gabriel: And it means they take ownership. and I've, it's rare for me to meet a data scientist who wants to be treated. Like a service center or be given the orgs where data scientists are responding to tickets are some of the least successful ab Absolutely. and it's very, and it's very easy. I've seen it like many times. It's very easy to become that. it's, and basically once you do that kind of, you lose like old perspective of prioritize, like of priorities. You're just like answering questions like all the time. as opposed to thinking about, okay, what would really add value to the business?

[00:35:00] and I think also doing that shift, I think it's really important and I think being a data leader, I think has a lot to do with that. Really like prioritizing, what are the, the big business opportunities. And that's why, like when I talk to companies, not just in Chile, but in the U S I think it's really critical to have data leadership that, that, understands the technical aspects, but also the business aspects.

I can actually really combine this to, to, make the right choices, the right priority, like work on the right priorities.

hugo: Absolutely. And I found also having data scientists and MLEs who are have more of an entrepreneurial spirit and they're there to solve business problems, not just technical problems, as we've discussed as well. But they're interested in what they're actually solving for. I am interested in, let's say we're in a situation.

gabriel: It's clear that Really executive level buy in, if not board buy in to,Yes,

hugo: right. And shareholder buy in, but let's say I'm a data leader in an organization and I like [00:36:00] my teams, I'm using a cliche here, but I've got a data team that works in the basement and they're big bearded guys like me and, I'm a data leader and I want to get more executive buy in.

what type of strategies can I use to make sure that. My bosses and the executive and leadership, I can get as much buy in from them as possible to be data driven. 

gabriel: That's a great question. so maybe like the first advice I'll give is try if you're a data leader, like try to work in a company that has already like leadership buy in. It's just it's just very, I, what I've seen is that it's like very hard to change that like bottom up. Uh,and it has to do with also what you were saying is if you don't have buy in, not just from the C- level, but also you brought a good point from the board and what I've seen in Chile.

Also, what I was mentioning before is the gap between the sea level and sometimes, the people working in the problems that also applies to the board,which happens quite often. Chile and maybe other [00:37:00] countries like America are pretty like conservative. And so you have this kind of old guard of, that's typical, like in, in boards, that may not understand,what a data scientist,how a data scientist could add value.So maybe first, if you can try to work in a company that has, buy in, and I actually have learned that the hard way sometimes,and I think if, okay, let's say like you are faced with this challenge. think you need to be incredibly entrepreneurial. this is what you were saying here.

You really need to,go out, find partners. Like the, probably the main thing I would do is find the right partners. So if maybe like you don't have buy-in from leadership already, but maybe you can find. The right business partners within the organization that are willing to try things to you that are impactful to the business.

hugo: And if you can do that and you can show that to leadership, at least that will probably give you the best shot.so I think it's really a combination of being [00:38:00] entrepreneurial and find the right partners. and it doesn't need to be solving complex problems. It has to do with solving relevant first order, problems. There's the other thing that I've seen happen whereby you can provide analytics as a new data function that gives people so much information about what's happening that they have their hands full then as well. And they're like, Hey, the first two insights actually are all we need currently, because they will solve a lot of things for us and we actually don't need a massive function.

Right. 

gabriel: Yeah, I, I think that's also like a helpful, starting point. So maybe I was, talking more about the data science, function,but you're absolutely right. There's basic analytics,that just, that would be helpful. in my experience in this large companies, so maybe the interfaces are not great, or maybe they're not immediately available, but there's some analytics already.

But what's really lacking is how to use data to improve products, to [00:39:00] really, embed data in, say, a customer journey,in how you improve the offerings to customers, reduce churn, all these things thatdata science would do. I think the other important aspect that really lacks typically is experimentation.

It's incredibly hard in my experience to convince people that experimentation is important, and that running an experiment that has a flat or negative result is actually helpful because you are learning that what you thought was a good idea is actually not a good idea. And I found, in my own experience, again, in companies that maybe were not 

born data driven,with an experimentation culture, which is like now very typical in Silicon Valley, of course, many tech companies in general,to see the value of experimenting, and again, then it falls to, who has the. The [00:40:00] strongest voice who says, who in the meeting is more aggressive.

And, you shouldn't make business decisions like that. You should tell, you should let the data speak. But I think that's, that's another big challenge. And going back to your initial question, I agree with you. I think there's like,More like maybe low hanging analytics, an analytics function that could be,if a company doesn't have it,that by itself, having dashboards, just the basic analytics that by itself would be, incredibly valuable.

hugo: Absolutely. I'm glad you mentioned, the importance of experimentation and the importance of negative results as well. as you may recall, I used to work at my backgrounds in cell biology and biophysics, and I always wish there was a journal of negative results, because we all do the same negative results.

Can then you meet someone at a conference and they're like, Oh yeah, I tried that and it didn't work, but there's no incentive to publish negative results on, A tech company and data driven side. I actually recently interviewed, I think your colleague Ramesh Johari

gabriel: Yeah. Good friend.

hugo: the show, [00:41:00] man, he's cool. Isn't he?

gabriel: Yeah. He's awesome. Ramesh is awesome.

hugo: and we did a really an episode around the importance of experimentation, how to develop a self learning or organization as he put it, but also, To your point, making sure there's a culture in which people not only feel comfortable getting negative results, but recognize that getting negative results may not impact the business in the short term, but in medium to long term learnings, particularly if you try to gather insights across longitudinally across experiments, what you can learn as an organization then, and reshifting to that mindset, I think,

gabriel: Absolutely. and creating like shifted to that mindset is not obvious. It's pretty interesting becausewe come from Silicon Valley and this is like obvious, like experimentation is, this is just how you do things. And then you come with an environment where people have never run experiments.

And it's what I found in my personal experience, just so hard to convince people that [00:42:00] this is valuable. It's like, if we think it's a good idea, why don't let's just implement it. Why would we spend time, running an experiment? They say, because maybe it's not a good idea, roughly, 80 percent of experiments run in 10 companies are either flat or negative.

a lot of the things we try are not great. So, finding the value of that process, I think, is super important.

hugo: And what's the blocker there though? Because I think, business leaders historically and current good business leaders have a deeply experimental mindset. It may not be a quantitative experimental mindset in the way we're talking about it and the way implemented places like Uber, AirBnB. Amazon, and so on, but if you're, if someone tells me Henry Ford wasn't a crazy experimentalist, I'd say there's no way he didn't experiment every day with new ideas.

and same with, we look at something like, alcohol, right? sorry, it's Nearly 4pm and I'm in Germany, so I think I, people, I feel people are drinking MassKrugs out there. But, so the Guinness, factory, it [00:43:00] was William Gossett there who published a paper about statistical tests.

He couldn't publish under his real name because he worked at Guinness, so his pseudonym was Student, right? So he published the student t test, which came out of Guinness, right? so historically we've had a lot of businesses that embrace experimentation, maybe not in the most quantitative sense. So how do we convince business leaders that this is the way to do some forms of experimentation now? 

gabriel: I think that's a great question. which I personally struggled with and in my experience, it also has to do actually like even let's say, like a business leader say, okay,we should like experiment. and that's a message. It's a message to the company. Okay. But we should like,we have this, I don't know, Stanford professor is trying to help us build that experimentation function.

Why don't you guys be nice and like help him but I also think that's like valuable to the business. And I think what I've seen also, sometimes you get resistance from the actual [00:44:00] teams and, And the resistance comes, I think, from a few reasons. One is, why do we want to spend this?

We just, we need to practice. we need to,launch. we don't have, time, to experiment. the other one is, you don't have an infrastructure, it's an experimentation infrastructure. it takes time. and you may not, you may need to invest engineer resources that are scarce that are typically this curses in,in, in companies.

and People are just not willing to do that. so I guess it's not just about willing to experiment. I think there's also a commitment to a minimum level of resources, to have a basic infrastructure to enable the experimentation. And I think that transition, it's not that simple.

And again, what we're discussing Before probably if you are involved in this efforts, like the most valuable thing you can do is show wins. [00:45:00] and again, like wins. It's interesting how you define wins because a win may be like realizing that what you thought was a good idea, it's not a good idea.

So that's just harder to show,to, to a CEO. but then if you start like aggregating results, maybe like over some period you run like. 100 experiments and then you should listen like 20, 30 percent were positive and this is the lift. I think that's very compelling. and so I work in, in, in a startup, in LATAM, building their experimentation infrastructure.

I have, we have all, all of the above, like all the challenges I just mentioned, it wasn't easy, but what I thought was compelling is, to, to the leadership is actually showing this effort of, Not just running one experiment, running tens of experiments and how eventually by iterating you get to the right answer.

Like maybe you won't get to the right answer. First experiment, not even at the first 10 experiments, but eventually by iterating this process,you're learning and [00:46:00] getting, to a better answer. And I think that's compelling. So if you find the right partners, like a use case where you can do that, I think that can, be a game changer.

hugo: Very cool. And it's great to hear of a particular case in which, you're working with a company on the ground there. and to your point about, you need to invest in engineering and platform stuff,I've worked in startups. I've never worked for a massive, company. It's not really my type of thing, but I've been in startups for a decade.

And the story I've seen, even at companies that have become huge, They'll hire their first data scientist or two, and those people will become essentially data engineers for 12,

gabriel: Yes.

hugo: 18 or 24 months. And they'll have issues with AA testing, right? Like it's, the infrastructure and we could do a, I'd actually love to do an episode or a panel on the importance of AA testing sometime.

But the point being that if you don't get the infrastructure set up, the experimentations don't even work. And there is initial, A non trivial, actually, yeah, a not insignificant initial investment that [00:47:00] Needs to be put into force. Historically products like optimizely have helped us a

gabriel: That's right.

hugo: getting over these hurdles,

gabriel: That's right. That's absolutely right. And I think what's alsohard is how, how to use experimentation, to really change valuable parts of the product. So something that is not, Hard is like run experiments, to, test different like messaging strategies. do you email?

do you send text messages? Like how in what frequency? That's easy. I think what's much harder is let's use experiments. We have all these ideas to actually change the product. How do we run experiments to do that? Because that requires engineering resources. So you need to have an organization that has enough flexibility and sa and willing to spend, engineering resources, to, there's this like treatments that you wanna test.

Okay. But those treatments require some engineering resources to get built. So you need the willingness to actually, do that. And I think that's another challenge [00:48:00] I've seen when building an experimentation,culture.

hugo: absolutely. So I am interested in bringing the conversation back to your experience now in Latin America, and you have said that if, these big companies don't get serious, They will fail. And I'm wondering, I suppose I haven't necessarily got a well formed question here, but I think it revolves around current incumbents, perhaps.

gabriel: So companies that are, Latin American, but not necessarily a data ML, AI native, then American companies that maybe are trying You know, make movements into, into that region versus local startups. So you've got this triangle of how to, how do you see these dynamics playing out and what's your optimistic take? Yeah. No, I do have an optimistic take. so let me start with some data. So there's, this, Latin American AI index that an organization here in Chile actually builds. and what you see in the data, and that's basically. Trying to figure out [00:49:00] what is the state of development of like technical tools like AI in, in Latin America, across different countries.

So what is that, for example, a country like Chileis number one Latin America, but it's like way behind the US. Way behind the north. Okay. And if you double click in a country like Chile, which is the most advanced Latin America, you have the talent. There's a lot of research coming out.

There's a technical capabilities, but there's like very little investment and sophistication in large corporations. So for example,when you look at the, the R and D,in developed countries, two thirds of the R and D is done in companies, one third is done in universities, in, in the developing world, countries like Chile, Latin America, it's the opposite. The ratio is reverse. So there's like very little R and D in companies happening in companies. And I think like with this large companies, maybe. Because of all the challenges we've discussed, maybe that's just hard to change, even though, [00:50:00] some will and some are.

And for that reason, I think going back to your question, like fostering an ecosystem of innovation where startups can thrive, I think it's critical. I think that's what's going to maybe start closing the gap. and if I think about, policies,I would think, how do we reduce the barriers to start companies, to hire people quickly, to develop infrastructure, just, help fostering an ecosystem of startups where we can also attract funding.

because we have the talent, in, in, we have a great engineering schools in Chile, in Argentina, in Brazil. but sometimes we're, lacking other, other things to build the ecosystem. And that's where, I have an optimistic view, you know, how, and that's my hope to build, which already started like 20 years ago, there was nothing in Chile.

there's absolutely no entrepreneurship in Chile, no tech ecosystem, [00:51:00] relative to the US is still, it's still small, but it's like way larger. and there's like a few unicorns, way larger than 20 years ago. And I think that's the direction we need to move. and relative to, maybe companies in the US, I think, Which, some have come, To LATAM with success others with less success.

I do think there's a space for Local startups like really knowing the culture like, you know speaking the language in You know a lot of this, customer facing Applications is importantUnderstanding the local institutions and so on. It's important. so I think there's like a huge opportunity.

There's a lot of exciting things happening in tech in Latin America. And I just hope and I think that will, that will continue to grow. And I think that's critical going back to one of the previous questions you asked.in terms of closing [00:52:00] the gap, the technological gap, between, Latin America and say the U.

S. 

look, all of that is so exciting and I appreciate your optimism and it gives me a renewed optimism. I also love the focus on enabling local communities and local people to do things that are important to them and build software, build companies, build Yeah.

hugo: I lamented to him that we were having, local elections.

I was in Australia at the time, and our papers were filled with more information about the upcoming American election than the local election, which is arguably significantly more important to us, right? So I am excited about a renewed focus on, on, on. On local communities. Also, I'd love to see more,Spanish language, language models.

For example, the fact that, my friends who want to interact with language models in Spanish will go to ChatGPT, which was trained on web scale corpus, which is 70, 80 percent English language, but it has enough translation [00:53:00] capabilities to go back and forth and do that type of stuff, right?

Imagine if the people of Chile had their own model, which incorporated without all this external noise, the stuff that was important to them.

gabriel: Yeah. That's a really interesting question because I was just like last week, talking to a researcher here in Chile that has a research center on AI and they're building the first Spanish LLM. and I was asking him like, And initially that didn't make sense to me, it's like, what, we have chat GPT and, I use it in Spanish, it seems to work pretty well.

Um, and basically the point he made was, not about translation, like these models are, could be pretty good at translation, it's about capturing local culture. that, you know,ChatGPT is trained on a corpus, that is, maybe 90 percent in English,and what he claims is that doesn't capture,local knowledge and local culture.

So basically they're training this model with very specific,very specific, like corpus of knowledge. that is [00:54:00] localized to Latin America. so this would be documents that are like, getting from, the national libraries of the different countries, things that you would not find in the web, but constitute really important, like local knowledge.

So I'm like really curious to see,how does it work and like, how does it compare with chat GPT? I think it's a fascinating, project. Yeah.

hugo: with ChatGPT is it's just a bit for an Australian. It's just a bit American. Like it has an American vibe and it's just, it's a bit polite in a classically American way. My German friends actually like, I can't stand how polite it is.

I wish it would just say to me, you're wrong. Not of course I'll do this and that, and I don't necessarily want an Australian LLM that's G'day mate, go down to the blah, rah, rah, rah, rah, throw shrimps on the barbie or all of that, but something that would capture, more about Australia.

I definitely find interesting. The other thing you mentioned about building You know, we're talking about organizational culture. We're also talked about building a more like city based culture [00:55:00] and country based culture of startups and technology and this type of stuff. And that does remind me of what we've seen happen across parts of North America in the past, past decade, right?

It used to be very much localized in Silicon Valley. I lived in New York for the best part of a decade. During that time, I saw a lot of data science, ML, and AI shift over there. we've seen Austin, Texas grow a lot. Nashville has a lot of Python startup stuff. happening now. And similarly, Miami has a bunch.

Miami is a different story because everyone from web three pivoted to AI. So you can't even tell what's happening there. But I do get a sense. There does feel like there's a sea change happening. So I do think, your optimism isn't only warranted, but I think it's pointing in a very good direction.

gabriel: And I think what's critical is to provide a government in my view should provide the right incentives to create these ecosystems. and so my hope is that, in Chile, that. Startups are built here and they want to stay here. I, of course, they'll go to Chile is too small, right? So they need to like operate in Brazil and Mexico, [00:56:00] like the large LATAM markets, but that the talent,the teams stay here.

And for that, I think you need to provide the right incentive structures. so that's my hope, for, for Chile and Latin America more broadly.

hugo: Amazing. So I would, I'd love to chat some other time about, the incentive structures we can set up from that particular level. but I think too. To wrap up now, I'd just love to get a sense of there's a, there's not only a skills gap in a lot of places, there's a literacy gap and something I've seen, you mentioned there are a lot of good engineers.

I'm interested if there are people who can think in a data sense, because I think the engineering skills can be picked up in some ways, but for engineers to start shifting to a data mindset of building software that incorporates the entropy of the real world where you need different types of monitoring and maybe not unit tests, but more smoke tests and these types of things.

gabriel: I'm wondering, just given the cultural barriers in many companies, both in LATAM and elsewhere, what strategies you've seen when it comes [00:57:00] to fostering data literacy among business leaders and decision makers and even in government, Yeah, no, I think there's like, when you think about business leaders, I think there's, there's a huge gap, and I've seen, I was recently, actually, taught a class in a program, that a research center did for, I don't know, hundreds, like maybe 200 C level executives in Chile. And it was basically about teaching, the basics of machine learning and AI.I think they're evaluating the results like is like a long, like a few months program and it was like serious content. they're evaluating the results. I don't know, maybe for a next episode, I'll tell you how effective it was, but, absolutely super important, to have this type of, training programs. 

hugo: I am interested and we've talked around this a bit, in wrapping up with where you see key areas for future development. So essentially where you see the biggest growth opportunities for data science and AI in Latin America and what businesses and governments, I like [00:58:00] the. Important aspect of shifting incentives to make sure everyone's incentivized to move in this direction, but what businesses and governments in the region can do to become more competitive in the global data economy. 

gabriel: So the biggest, opportunities, so one, I actually thought in government. Like governments are tremendously inefficient. in Chile, we have a huge issue about permits. So when you want to do an investment project, permits can take years to process. And that's just like pure inefficiency. And there's some political barriers as well, but there's a lot of, inefficiency that can Be solved and speed up with AI.

That's something I'm trying to look at, actually. so I'm really hoping that governments will modernize and become more efficient and serve the needs of the people. the other area I'm like, really excited about, this, education and health. So the access to education and health, in countries in Latin America.

can, that not everyone has access to [00:59:00] good education and health. and if you go to remote regions, for example, in Chile, maybe there's going to be one teacher for 50 kids from K to high school, all in a room. using AI tools to aid these teachers, having like more personalized,education program for the different students.

I think this is just a tremendous, opportunity there, in terms of health is similar. You go to remote locations,in Latin America, and there's no specialists. maybe there's one, general doctor, but no specialists. So using, AI tools to aid doctors with more specific needs of patients will, have That's, a tremendous opportunity.

So I think maybe what I'm mentioning, these are,opportunities with a large positive social impact. of course there's, bunch of other opportunities in retailing, in banking, access to, financial products, that are also super exciting and already, like in this.

ecosystem of startups are, there's [01:00:00] a lot of things happening, in that space. So I think it's like a really exciting time. we're all,super excited about excited. And then also, sometimes, looking, Maybe with, a bit of hesitation, like how a I,will pan out and I'm a tech optimist.

So I see a lot of opportunity. Of course, there's challenges that we need to manage. But maybe just to wrap up, I'll probably like finish with the message that.there's like many different paths that AI can take and this is not like exogenous. It's not something that will just happen. It's really a function of what policymakers, business leaders, will do, like what decisions that they will make.

hugo: my hope for the future is that we'll make the decisions, so that these tools, AI tools, will for the most part have a positive, societal impact. I'm excited for that future as well. And we have a lot of work to do, but we're doing it. And I appreciate your optimism. I want to thank you to wrap [01:01:00] up for not only your time and wisdom and generosity, but also a refreshing and different take and new knowledge from a different part of the world, as you know, and as our viewers and listeners know, I live in Australia, I honestly still know more about the U S tech scene than the Australian tech scene.

For a variety of reasons, I'm in Europe. At the moment, as you know, I'm in Berlin and there's so much exciting AI, data science, ML stuff happening here, but we don't hear about it as much as we do about what's happening in North America. We very rarely hear about what's happening in places like Chile and Latin America.

So for you to come here and bring this wealth of knowledge and experience, very much appreciate that Gabriel.

absolutely. It was, really fun for me.Fantastic. thank you once again.

gabriel: Thank you.