Episode thumbnail
Episode
29

In this episode, we sit down with members of the MIT Low Income Firms Transformation (LIFT) Lab: Director Josué Velázquez Martínez, Postdoctoral Researcher Sreedevi Rajagopalan, and Doctoral Student Fabio Castro.

We discuss the LIFT Lab's work empowering micro retailers and nanostores in emerging markets to lift themselves out of poverty. These retailers, while making up an overwhelming majority of retail business in their regions, are at a significant disadvantage when dealing with large suppliers and competing with large retailers—some 30% of these firms fail within 5 years of opening.

Using generative AI, the LIFT Lab is helping these retailers enhance their business decision-making and supply chain capabilities to help them survive and thrive.

Transcript

- Welcome to MIT Supply Chain Frontiers, presented by the MIT Center for Transportation and Logistics. I'm your host, Benjy Kantor. Each episode of Supply Chain Frontiers features center researchers and staff or experts from industry for in-depth conversations about supply chain management, logistics, education, and beyond. Today, we're happy to be joined by members of the MIT Low Income Firms Transformation Lab or LIFT Lab, with director, Dr. Josué Velázquez Martinez, postdoctoral researcher, Dr. Sreedevi Rajagopalan, and doctoral student, Fabio Castro. First, MIT CTL offers a variety of educational programs for graduate students, seasoned industry professionals, and anyone at any level looking to learn more about the supply chain and logistics domains, including new online courses on sustainable supply chains and humanitarian logistics. To find out more about all of CTL's educational offerings, visit ctl.mit.edu/education. In many emerging markets, micro and small enterprises, or MSEs, make up an overwhelming majority of all retail business up to 90%. Think mom and pop shops but smaller, but in the face of competition from larger firms and economic damage from the COVID-19 pandemic, MSEs survival is increasingly under threat. 30% of them do not survive the first five years of operations. The MIT LIFT Lab or Low Income Firms Transformation Lab aims to help MSEs lift themselves out of poverty by enhancing their supply chain management and capabilities through applied research and emerging technologies. And today, we're going to find out how small retailers can leverage those tools, including AI, to survive and thrive. Josué, Sreedevi, Fabio, welcome to the program. Thank you for joining us.

- Thank you so much for the invitation, Benjy.

- Just to get it out of the way, I do know that we're also celebrating Sreedevi's birthday today, so happy birthday to you, and thanks for joining us. You probably have celebration things to do.

- Thank you so much, Benjy.

- If you were to give your elevator pitch to somebody, and Dr. Velázquez, we'll start with you Josué, about what the LIFT Lab did and how it started, how would you tell the layperson what you do? Micro firms, just in Latin America and other emerging markets, represent 99% of the firms in the world. In comparison with the largest firms, they actually just have a fraction of the productivity. In this comparison, when you say the majority of the firms are actually having low productivity and low survival rates, this actually implies a huge impact in the economic environment of the countries. They actually, by far, explain poverty, opportunities that rise also in the regions, and challenges that are faced also because of the low level of education that all these people actually that manage these micro firms are facing. So the project aims to tackle particularly this problem, not just those micro firms, but also the consumers of those micro firms that are what we call the bottom billion or the base of the pyramid. By looking at the whole ecosystem, including also the largest CPGs, we are convinced we are gonna really make a difference and an impact in what we believe is probably something that is going to transform the future of the economy in emerging markets.

- And to get, I guess, into the heart of things, what are the specific challenges that these firms or these micro firms are experiencing? And then, to kind of glom onto the term of the day, how is it that generative AI is gonna affect them or help them or perhaps hinder them?

- Yep, absolutely. From one side, imagine that you compare these micro firms, these, what we call nanostores, because, as you said in the intro, it's not a mom and pop site. It's actually smaller than a mom and pop in the US for instance. So these nanostores are serving the low-income market, most of them. But in comparison with a modern channel, and you can think of, you know, a Walmart or a Carrefour or just a big retailer. Usually a big retailer actually gets 60 days, 90 days to pay suppliers, for instance. They usually can bundle and provide economies of this scale because when they place orders to either Coca-Cola, PepsiCo, Dannon, they usually can agree on larger terms that will provide better price per product that they are acquiring. And this, of course, translating to offering products cheaper to consumers. Now, when you compare now the context with the micro retailers, with the nanostores. What happens is that, first, how many days do they have to pay suppliers? Well, the answer is they have in average minus two days. So they actually have many times to pay in advance. One, 1.5 days in advance. In some cases, we have observed that in countries like Brazil, for certain restaurants, some companies were actually paying one month in advance. Now, they actually seem to be financing the larger retailers. Now, in addition to this context, they also place orders that are way smaller. So they actually don't get any quantity discounts. Therefore, the products that they acquire are actually more expensive than what a big retailer might get. Now, once they get the products, they still try to, you know, divide, split these products to smaller versions, because many times the consumers do not have credit cards. They actually pay with cash. When you split this, it means you're actually increasing the price per product. And who is actually paying for that? The person that has no money or at least difficulties, you know, in the cash availability. SO because of this operations, it makes it so expensive to save this market. And if you look at the retailers, when you look at Walmart, you see, well, plenty of trucks go to a warehouse and then there is a single that goes to store. When you see what happens with the nano retailers, you actually have all the different trucks delivering to a single corner nanostore. The question is why this market or this business has survived that long. Well, the reason is this is the largest retailer in the world. There is estimated a number of 50 million nanostores in the world. In Latin America, for instance, they're actually bigger than Walmart. The sales of Dannon, PepsiCo, Coca-Cola, they actually represent from 40 to 70% of all the sales of these big retailers.

- In the conglomerate as a whole?

- Exactly. And also, if you put that, that you say, well, yeah, we know that they actually don't manage to survive. You mentioned a figure of 30% in the first years. Yet, what happens is that there is very low barrier of entry. So people do not have any other means to make a living. So what do they do? They open their window, they open their garage, if they have, and they just start selling whatever they can sell. And then, Coca-Cola will deliver. Pepsi, they will deliver, and then suddenly you have a small version of a nanostore.

- Well, I've got to imagine that with that many standardizing or codifying how the supply chain works, how things get to them is the real challenge, right? Like the person, the nanostore that's going to succeed is the one that figures out and works with vendors and suppliers, how they actually get things and then get them out their door.

- And consumers. So how we see it, just to illustrate this, is for one side, whatever efforts we can make for the suppliers, the largest CPGs, to improve the delivery of goods to nanostores, that's going to decrease the price. If we help also nanostores operate, because at this point, it's hard to imagine how they're doing demand planning, how they're doing inventory and management decisions. Like this is not even in the discussion, because at the end, they even are not having means to keep track all their records. So if you don't have data, you are actually working with your own intuition. And every time that many countries in many regions, including also some NGOs, like World Bank, IDB, have really tried very hard initiatives to improve the technology adoption of these nanostores to help them make better decisions. Because the contention is by having better decisions, you're gonna also improve your cash availability and get better prices for the consumers that really need it. Now, the studies that we are doing in generative AI are related to particularly answering this question. Can we manage to improve the technology adoption of the nanostores and the retailers? Can we find ways to leverage the growth on these large language models to improve not just this, but also the performance in the decisions they're making? And some of the experiments that we've done in recent months are actually tackling these type of issues, which is part of the dissertation of Fabio Castro and also the work that Sreedevi is co-advising.

- Well, I want to get into the work that you guys are doing as well too. And I planned to ask about the efforts for standardizing those processes and standardizing the data, and whether it makes sense from nanostore to nanostore to have that standardized data. So Sreedevi and Fabio, actually, if you don't mind talking about the work that you're focusing on for LIFT Lab, and what part of the country or the world that your work is focused on?

- So yeah, at the moment, we are focusing on Latin America, particularly Mexico. So if you look at the challenge that these micro retailers face is given that they buy on cash and sell on credit to their final consumers, because majority of the consumers that they cater to are from the lowest part of the society, and hence, you know, those consumers may not be able to pay, you know, upfront when they purchase goods. So because of this, and these are exogenous factors, these are some things that are not in the control of these micro retailers. And as a result, their cash conversion cycle is longer. So the only way they can tackle this cash conversion cycle is, you know, through effective inventory management. Now, if you look at, you know, how they manage inventory, the fact that they operate hundreds of SKUs, it's difficult for them to, you know, even if they use their intuition it is difficult, you know, to design on how much to buy, when to buy and what to buy.

- And very quickly, just to define SKU.

- SKU is a stock keeping unit. For example, you know, one of the top-selling product is Coca-Cola, so an SKU could be like 500 ml Coca-Cola bottle would be one SKU.

- And could you also comment on cash conversion cycle?

- Okay. So cash conversion cycle basically is about the amount of time it takes for them to get the cash back, or the return on investment. So it talks about the number of days it takes to pay their suppliers. The number of days it takes for them to receive cash from their customers and how many days it takes to sell their goods and get money out of it.

- And I've got to imagine that there's challenges with that too, because with some nanostore situations, you have a street where a truck can get down and park, and then you have nanostores where a truck cannot get down and park. And so, you know, like, that's got to affect where this conversion cycle, where, you know, if things aren't getting delivered because the delivery driver has to get to the next delivery stop before they can find, you know, like... How does that cash flow in and out?

- If I can just jump in on that, Benjy, very quickly. So, we use metrics like the cash conversion cycle because we are interested in making things fast for the shopkeeper to get money, right? If you see for instance, that I need to pay immediately to my suppliers or sometimes in advance. If, for example, I have no other option to take a credit card because if I take a credit card, then the bank is going to give me back the money in 30 days. So all those days make a difference whether I should adopt these type of payment methods or actually just focusing on the cash. Because if I lose the cash, as you know, this is the reason why all of them go to bankruptcy. So in parallel to the flow of goods, like in any supply chain, we are very much interested in understanding the financial flow and how that financial flow can actually speed up in all the metrics that we are doing so that actually more cash, more capital can actually be allocated to the shopkeeper to improve the chances for survival.

- So to put it bluntly, their rent is due and their landlord is expecting cash on the first of the month or whatever. They have to wait 30 days. And so Fabio, if you don't mind jumping in with how your work.

- I'll just build on this and why this situation of the cash conversion cycle is quite relevant for these micro retailers is because different than the big retailers, they don't have so much access to credit. So if they are not able to have their cash circulate back so fast, they'll not be able to buy the next products, the next round of products. So, they will run out of cash. Why large retailers like Walmart? Well, they have access to all the financial system to obtain credit or sell shares in stocks.

- Well, so Fabio, let me start with you on this question, which is what are the solutions that LIFT Lab is working on and, I guess, how do they work or how do you implement them? That's where I'm getting stuck.

- We are working on many different solutions, on many parallel projects. The one I'm going to focus on is it started when we realized that these small businesses, they are adopting management systems and different technologies. They're not adopting because of supply chain decisions, they're adopting to improve their financial management. So they're adopting these simple systems. And simple systems happen to have very rich data about transactions, sales, purchases, inventories that they're not using. So, what we started looking at, okay, these are rich data. How do we communicate this data with them? Should we use dashboards? Should we use charts? Should we use text? Then we realized that, well, all the shopkeepers, they actually use WhatsApp. They feel very comfortable talking with their customer, suppliers, and also their customers place orders through WhatsApp. They talk with everybody via WhatsApp sending text messages. That's the most used application in Latin America. Even though people in the United States are not familiar, don't like so much, they prefer SMS. And so we realized that, okay, at the same time, the most comfortable tool they have their hands is WhatsApp. Well, there is just, OpenAI just develop a new tool that allows people to talk in a very comfortable way with the machine. So we realized let's merge it together and use these large language models to actually communicate the analytics of the store's data to the shopkeeper in a language he understands. And that's how we started this project. I have the impression, many of the projects using generative AI, they have this hammer and they're looking for a nail to hit with the hammer, which means like, hey, let's do some research with large language models. In our case, we realized we had the problem we wanted solve and realized, look, there is this new tool that we can use to solve this problem. And that's how this project started.

- What are the stakes for the folks who are gonna be using this technology? I mean, as we said, like, you don't succeed, you go out of business. Are these also tools for the company, for the nanostores that are surviving three, four, five plus years? Are they already doing these things? Have they already learned this on their own? Are there additional tools that they're...

- No, they're not learning on their own. Basically what we are trying to bring to them are all the mathematical management models and data-driven decisions that are very already common in the large companies and that are taught here in the CTL SCM program. So there is a large body of knowledge in science coming out of business schools that does not reach these micro retailers. So, what we are doing, we are using their data to do analytics and communicate with them. Large companies like Walmart, they will hire SCM alumni to do their analytics and who will be able to understand the data, do the analytics and understand the charts. The small business, they don't have somebody who is able to do these analytics. So what we are building is a tool to use the existing models in the supply chain body of knowledge and communicate it to these micro retailers in a way they understand it. So, they also get benefits from these models.

- And from a practical standpoint, how do you implement those? How do you get these, thousands and thousands and thousands of nanostores and stakeholders to participate?

- So one of the things that we saw, you know, in terms of technology adoption, one of the barriers is these micro retailers, if they do not see the perceived usefulness of that technology up front, they may not be able to, you know, adopt the technology. And the second thing is, second challenge that we see, or the barrier that they face is, if that technology does not come from a trusted source, they may not be willing to adopt the technology. So, we had done some pilot studies in the past to be able to reach out to these nano stores without the app that our doctoral student has developed. But what we see is, like, given that they have to input the data for a period of time before they could see the result, it is difficult for them to even adopt in the first place. So going forward, what we are thinking is to collaborate with a supplier. For example, we are right now collaborating with a large wholesaler, you know, who is catering to around 3000 nanostores. So, we are collaborating with them, given that the micro retailers have the highest level of trust with their suppliers. So it's easier for us to reach out to these micro retailers through a supplier or through a sales agent.

- Yeah, because there's no database of email addresses of the nanostore holders.

- Correct. Yeah. So that is one way. And, you know, if you look at answering to your question on the stakes that these micro leaders retailers have, if they don't adopt technology. You know, the fact that inventory management is the only thing that's in their hands, given that they have to pay a friend to their suppliers and then sell on credit. So if they do not manage that inventory in terms of how much to buy and when to buy and what to buy, either they're going to end up having high levels of stock out or they're going to end up having huge amounts of inventory for some goods. You know, and that's going to actually block their cash and strain their already cash-strapped business.

- And let me also to build on what Sreedevi commented, and Fabio. We have the effort, we have from one side the effort of dissemination, how we are gonna reach out all these hundreds or thousands of nanostores so that they could make better use of the technology. But on the other side, we are also working on the lab. We call it the lab experiment. So that means we are developing technology and testing the technology in control settings to really learn what is the potential they may have. And part of what Fabio started, you know, with the dissertation is developing a chatbot, you know, built entirely by him in which users could actually submit some questions, inquiries, and then get to the prompt that we were designing. And then coming back, allowing ChatGPT to just interact in a more, as Fabio will say, human way. So the experiment that was done actually tested these three scenarios because as Fabio said, we know that there is not technology adoption in the nanostores. But there are some, right? The number, and correct me on this Fabio, should be around 30%. You know, one third of them probably are those that we studied, that we have already 15,000 in this first preliminary study. 30% of them are having some sort of POS system, point of sale system. So they actually keep records, as Fabio said, very valuable data. But the question is how many of them are using that to make decisions? The answer is another one third of them. So the question is why? And then we say, well, probably they do not understand all the things that we have already highlighted, low level of education, understanding. So what we did is build an experiment, a behavioral experiment in which we say, well, one group is gonna get simply all the records of the data, sales and whatever orders. Then the second group is gonna get that record of data plus dashboards. You know, the fancy charts that you can find in a common POS system. Then the third group is gonna get both. Like the data, the dashboards plus the chatbot. And the question is, what will be the effect of the chatbot in comparison with the other two? So, the preliminary results show two interesting insights. One, those that were using the chatbot, 5% increase in revenue. This is cash directly. This is an improvement in sales, as Sreedevi just mentioned, coming from the inventory management decisions. Now the second, this is usually the number that both of my colleagues here love to say. I love to say the 5% revenue because everybody understands. They hate me for this because they are scientists, so they like to say how close we are to the optimal value. Can you guys help me on this one?

- Yes, I can certainly help. We do identify that those use the chatbot or those who have the chatbot available for use, they changed their decisions in this lab experiment so that they make closer to optimal decisions. And it's interesting, I just like to mention that these decisions are not intuitive in the meaning that even experiment managers, even MBA students have been shown when they use their intuitions, their decisions are not optimal. And so what we are trying to do is you make the analytics to communicate to us the optimal decisions. And we do find that the users, they interact with the tool and they able to, from the tool, obtain better decisions and make decisions closer to optimal.

- So what they ask is, the decision that they'll have to make in that game, is in that simulation game, it's to decide on the optimal order quantity of the products. So what they ask the chatbot is, given that they already have access to the data, they ask the chatbot, like what is my average demand? How variable is my demand? Could you please tell me, like, if this is my optimal order quantity, what would be the outcome? So the chatbot basically helps them answer their questions and also direct them towards making decision on the optimal order quantity. So what we find is people who actually made use of the chatbot before they made the decision where their order quantity was closer to the optimal order quantity of the model, that the model predicted.

- So at its very base, if I'm a nanostore owner, am I saying to the chatbot, I ordered 20 cases of Coke, should I order 30? Or, like, how specific do they get?

- Yeah, we try to control in this. The idea is to have that possibility in practice. But the challenge is if you wanna test just interaction without knowing, because we know the future is uncertain. In this controlled setting, we know what is the optimal solution, right? But in reality, we will never know what's actually the right amount to order because demand is uncertain. There are many patterns, many factors may actually affect in practice. So what we do is to try to constrain that those questions will not get the specific answer how much to order, but just information and how to understand, read the data and already the context they already have as an organization. So some of this is still Fabio is working on understanding more about the type of questions they are rising and we want to study also what type of questions are actually leading to a better performance or not, right? And all those things are gonna be, you know, as insightful that can actually help us improve the prompt and the context. Help also feed a loop to improve also ChatGPT in general for the context of this type of potential solutions in the future. We're just in the tip of the iceberg on this one, Benjy.

- It's interesting how you brought up the point of sale and how a lot of the answers, they do have some formal point of sale.

- One third of them, yep.

- Okay. And we talked a little bit about this in a previous episode of the podcast actually with Melanie Nuce-Hilton from GS1, who produces barcodes and SKUs and things like that. We didn't talk as much about developing areas or about lower income places, but just getting folks on board and the challenges with implementation, I've got to imagine is really where... You can do as much study as you want to here and create these amazing tools. It's a matter of getting people to use them. And so I'm curious about that.

- I believe it's a totally fair question. Some of the things that we are doing actually Sreedevi just mentioned. We already identified the challenges, what's the value we're going to gain, right? How useful it is. Like, you come and ask me questions and you're telling me to use a technology, which by the way is gonna help me track information, but also somebody else that is having the technology.

- Well, if you tell somebody that if you implement this, your profits will go up 5%, like you said earlier, that's got to be a big draw, right?

- Yet there is always privacy issues, right? Probably one figure that we haven't commented on, but we are very much aware is that in these markets it varies from 40 again to 70% of informality. So from 40 to 70% are not actually paying taxes. So we know where they are. They actually identify as economic unit, yet they are not providing more information because of this concern of taxation. This is another discussion, but we know this is definitely challenge to actually overcome. So the way that we are targeting this, we have identified, we've run, I mean, the exercise, as Sreedevi said, we are focusing in Mexico now, but we work with nine countries with 20 plus universities in Latin America. We've learned a lot in seven years running all of this before, we actually are now developing these type of initiatives. But some of the things that we've learned through the focus groups and the workshops we've done with them is what are the things that they are mostly concerned? We take part of a special Facebook page in which we estimate there should be around 100,000 nano retailers. And we see what are the things that they are exchanging. What type of questions they are asking among themselves. And we actually discover one of the things that they care the most is actually the pricing. How much they should charge for specific products to consumers, so what others are charging. And they will ask, "How much are you charging? How much are charging?" So following this is another, by the way, research also with Camilo or another PhD student now in Mexico. By the way, footnote, we know that Camilo now had a little accident in basketball, so we're wishing him well. Just as a footnote.

- Get well soon, Camilo.

- But one of the things that Camilo brought is why don't we build a pricing app? And same idea, right? Working also with Fabio here, with Sreedevi, we develop a price app taking information from the government to have the prices of nanostores and just to tell you how it works. A nanostore owner or operator takes the app and says, "Oh, how much are my peers charging for this particular product?" Takes a look. And there is a map showing the locations of the other nanostores and saying, what is the price they're charging? Now this one was taking, oh my goodness, awesomely good. Like everybody got very excited about this app. Now what we are doing is this is a way to gain also the trust to show that we have good intentions. We are developing now in a partnership with Vintagium, an app that is going to be professional to help shopkeepers answer these questions. Our intention is to use that to later build capabilities as adding a chatbot. Maybe adding ways in which you can interact and ask questions and we may learn more and also develop new technologies for you. Now, we will find a way to also ask you other questions, but also maybe implement the app that Fabio is working with Sreedevi on inventory replenishment. We also have other projects that, by the way, Sreedevi, I would like also you to comment on the training program for instance. We want to also train massively. The exercise to, an effort to educate the shopkeepers has taken governments, thousands of consultants in the field, more than 50 hours per shopkeeper to actually be able to train them in certain practices. Can we actually use generative AI to improve that training and have same results? So some of these questions, what do you say, Sreedevi?

- Yes, Sreedevi, you know, we talked a lot about like how some of these are being implemented in Mexico. I know that you're working on projects outside of that area. Where else are you looking at? Because I've got to imagine that, especially when you're talking about pricing and that the price of somewhere in Europe or Asia is gonna dramatically be different from what we're looking at in South America and Mexico and the States.

- Yeah. So if you look at the need for a pricing app in Mexico or in Latin America, what surprised me was like the way they sell their goods. They do not have a maximum retail price on the product. Interestingly, when you look at India, every product that's sold in the retail sector will have an MRP, which is a maximum retail price. So that basically helps the micro retailers over there. They don't go look for prices like, you know, like, "What is the price at which my supplier is selling or what is the price at which I need to sell this?" You know, so they have a tie-up with their supplier, and then they decide, depending on the MRP, how much do they have to sell. But in a Latin American context where you do not have that price, a standard, you know, it's important that we provide something that will help them, you know, to price their product better.

- Are there MSRPs in other countries like the manufacturers suggested prices?

- I would imagine that, yeah, I'm not so sure if we don't have that in Latin America. I wouldn't dare to say it like that yet I believe we do not reinforce it.

- Professor Jan and I checked. Jan Fransoo and I checked, like, which are the countries that have an MRP. And interestingly we found that India and Sri Lanka are the two countries that have products with MRP. That's the maximum retail price. So when you know that this is a maximum retail price and it is not legal for you to sell beyond that price. So depending on, and then they also have different prices depending on whether the consumer is a loyal consumer or not. So if you are a consumer who frequently visits a store, then I might as a micro retailer, sell at a lower prices compared to somebody who comes new. So then I'll sell them at the MRP. So the problem of pricing doesn't exist much in a context like India, but it definitely exists in Latin American countries and other areas.

- And Sreedevi, you mentioned that working in India, what are the other sort of considerations or things that you're looking at there?

- This was before I joined MIT, I was working on around 25,000 women micro enterprises. My team and I, we were basically wanting to understand, you know, what are the factors that lead to their failure? These are enterprises that have support from the government in terms of training, in terms of, you know, startup capital and also in terms of market linkages. So they have, the government has set up an NGO that basically helps these micro enterprises with all this. So whenever you look at, you know, the requirements for effective entrepreneurship, you can broadly categorize that into human capital, financial capital, and social capital. So these are three main important things for any entrepreneurship, for any entrepreneur to thrive, right? And here in this particular context, in India, we see that all these women entrepreneurs had access to these forms of capital, still their survival rate was very high. And we wanted to understand why are they, you know, failing? And interestingly, what we found was, you know, we looked at areas with high crime and areas with low crime and women entrepreneurs were located in high crime areas. Although they have access to all these forms of capital, they either choose to set up their business within their household, or they try to set up their business in such a way that they, you know, go back home early. I mean, like they don't operate for a longer period of time. And as a result, their profits and their revenues are far lesser as compared to somebody who set up their business in a low crime area. So we find that presence of police station actually mitigates this effect. You know, the sense of security and the sense of psychological safety, you know, that increases when there are more number of police stations in a particular area. And basically these women are more willing to set up their business outside the household. They're more willing to, you know, run their business for a longer period of time and that actually helps them increase their survival. But this was way before I joined MIT, but right now my focus is largely on Latin America.

- One follow-up question. Are there organizations or researchers that are working to get to the places where there is a better sense of security or emotional security, safety, more protections for people so that they're feeling like they can be open longer hours or they can offer more products in bigger public spaces and things like that?

- Yes. Good question, Benjy. So my researchers and I, we are looking at conducting a field experiment to look at how their psychological safety changes depending on how secure their environment is. And then depending on that, we wanna make policy recommendations to the government.

- So what directions now are these solutions and research going? What are next steps?

- Yes, right now we are launching some field experiments with partner universities, partner suppliers, and we are launching this tool observing how the users, the final users, the micro retailers interact with these tools and what's the effect of this interaction. In parallel, we'll measure the outcome and the results so that we can link the usage of the tool, of the availability of the tool to their outcome, meaning their survivability, their profitability, their product management, okay? And just an interesting point, when we launched these tools, we're providing them with ChatGPT 4, everybody is aware, which is a very powerful tool that is able to pass some standardized tests. And in standardized tests it does as well as researcher from Wharton School says as well as a first-year PhD student. So suddenly we are put in the hands of micro-retailers free PhD student advisor. And what is the effect? How will they interact? So actually, even though we tune the tool, expecting them, and say, hey, ask inventory questions, well, they can ask any questions. So also we are finding out what are the types of interactions they have with this tool and what are the types of questions they ask? And we see that many of the retailers, they find it very useful. They use it a few times per week to ask questions that I don't know which other source they would be able to ask, when they want to ask questions about employee management, about marketing, about, well, any other topic in product management. In parallel, we must be aware that this tool is not intelligent. It's just a statistical tool to predict the next word. And I think everybody here has at least once been fooled by ChatGPT, right? Everybody remembers when they asked the question, oh, that's amazing answer And they realized that the book...

- Everything was wrong. It doesn't exist.

- Doesn't exist, right. So when we are playing with numbers and we are playing with business decisions, that's a high stake too. So that's something that we must build in the applications so that it's safe. So that it doesn't fool or doesn't lead to bad decisions our final users. So there is a large potential that in parallel we are exploring which, what are the other topics and decisions and other subjects that this tool may be able to advise micro-retailers, but also keeping in mind that there is some danger and that it must be reliable so that it doesn't actually end up hurting the micro- retailers.

- So, Josué, when you, vision of the future where these challenges are addressed effectively and solved and then what does LIFT Lab do after that? Is that possible or is it just a continuing effort on and on and on and on?

- I believe, Benjy, that if we are put out of business, that will be the greatest news for everybody. Because we are working on eliminating poverty, as you know, working with the bottom billion. Lifting the life of the bottom billion, this is more our mission. If we don't have to do it anymore, I will be happy to start playing chess from now. But the reality is that we see this is still like a big challenge. Things are promising yet we are just, as I said, at the tip of the iceberg. We are just discovering new technologies. We believe that now, actually spent many months in discussions to outline very comprehensive, what I believe a very comprehensive deployment of different projects that includes, you know, some of the things that Fabio already mentioned on inventory replenishment, the pricing, in demand management, in financial. Maybe we can have also assistance to provide accounting instead of you learning training and dissemination. So we are looking at different ones, including also all the projects in parking as well, and delivery operations as well as others indicator. So we are going very ambitiously into all these projects, yet we know that the technology is gonna keep evolving. So we are just discovering very little and we know that this prompt is gonna get only better and better. We are going to probably learn more what type of questions we should address with ChatGPT and which others we should not, and try to look for formal optimizations. But this always reminds me of the case of Wikipedia. You probably will remember because I believe we both are around same age, Benjy. But in Wikipedia when it was the first time, we all were complaining like, oh, they get it so wrong. This is very, you know, suspicious. It's not like the encyclopedia from the...

- The print volumes we had.

- It was nothing like that. And then in a matter of some years, suddenly, I remember around 2010, you know, the difference between one encyclopedia and Wikipedia was less than 5%. So in some the gaps we're closing, right, and just having this crowdsourcing approach has actually helped a lot to improve a lot of the tools and techniques. And I believe that this is going to happen also with ChatGPT and OpenAI and other tools related to the large language models. So we are going to continue innovating in this research, trying to stay very much alert on what are the new trends, what are the new challenges, and hoping that all the applications that we are building in our lab will land into practice and really make change to the nanostores. So that's our dream.

- Well, that's great. And I think the last question that I would have really, and I'll start with you, Fabio, is beyond this doctoral research that you're doing, what are things outside of the specifics that we are talking about now that you think will be affecting these nanostores or just low-income communities in general with regard to logistics and supply? Your own opinion.

- By the way, just a footnote, this is the toughest question for a PhD student, by the way.

- Oh, "what else?"

- Yes, because they spend five years, six years working on the dissertation and they say "what else?" It's like, well, wait a minute. Is it life beyond my PhD?

- Not what else that you're gonna have to solve, I just mean there things that are affecting the world.

- Yes, that's a so broad question. I can think of 10 different challenges that we could look at. Our colleague, Camilo, is tracking parking. Our colleague is Camila, is studying how delivery frequency and providing credit can affect these stores. There is other research on business models because there is a problem to be solved and there are so many new tools. So many new businesses can spring out, can be created, that actually tackle these problems in many different directions. There are so many different directions that we could look at. We haven't mentioned so much about the supplier decisions on how the suppliers better distribute to these businesses, how they communicate with the business. There is lots of e-commerce tools that the suppliers are using to reach these businesses using both applications also using chatbots to actually communicate with this business. Because right now we talked about chatbots for this business to help to make decisions, there are also new companies that are building chatbots so that these businesses can order from their suppliers, or also so that the customers, the final consumers, can order from these businesses. 'Cause not all the consumers, all the consumers feel comfortable using WhatsApp. They feel comfortable WhatsApp-ing their businesses. What if you make tools and their people building tools so that the consumers can use WhatsApp to make a conversational commerce to contact the different businesses?

- Yeah. Adding to that, it is interesting because if you look at the traditional way how suppliers reach out to these micro-retailers, it's through their sales agent. So it's a push strategy, right? The sales agent would say, look, this product is good, this is new in the market, we are offering promotions, why don't you take it? So the fact that these micro-retailers are already cash starved, they may not have enough cash to invest in all different types of suppliers or all different types of products. Now the fact that the supplier, you know, themselves are giving a chatbot for these micro-retailers to place orders, it would be interesting for us to see how now there's more control with the micro-retailers to decide the product assortment, to decide on the order quantity, and how it would eventually, you know, help them manage their businesses better. So I think that's a good study that we are currently pursuing with a wholesaler. To see, like, how it impacts between a sales agent telling the micro retailer what to buy, then a micro-retailer deciding what to buy.

- Yeah, and lastly, Josué, you talked about sort of the vision beyond this, but what is, I guess, what is that next step for LIFT Lab?

- We are gonna expand the projects, continue some of the studies that started with Fabio, because I believe it's well deserved that the idea of the large language models came from the dissertation that Fabio proposed. And we are gonna expand to the different domains, as I mentioned before. That includes the demand management, the pricing, the accounting, financial and training, and many others. We are going to expand also our outreach in Latin America, and we would like to involve other countries into this effort more. Like we started, like, as you probably know, and I would like also to acknowledge at this point great partners we've had during these past years that really helped us build a lot of the understanding we have. And to mention some of the countries, as I said, we work in Mexico but also in Argentina, in Ecuador, in Colombia, in Peru, in Bolivia, in Uruguay, Brazil. But all these countries help us. And please, I invite everybody, by the way, to go to our website liftlab.mit.edu so that they can see the names of the universities. But it's true that in the past two years, we really partner very closely with Monterey Tech and Tech Millennia, which really was a key partnership for us to escalate this in the numbers. As I said, more than 3000 students were involved and we have almost 15,000 nanostores involved in our study. So our goal is now, that effort that we have, you know, started this past two years, expand it to other countries, at least two or three countries more, and learn more, disseminate more, and grow more in the initiative to bring what we have learned, and also understand more the challenges to keep innovating. As you know, we are in the business of doing research, which at MIT is the process of innovation and shaping the world. Solving the problems of humanity.

- Well, Josué, Sreedevi, Fabio, thank you so much for joining us and really talking about this not just inspirational, but aspirational and forward-thinking types of research and the work that you guys are doing. I appreciate you joining me.

- Thank you so much Benjy.

- You can find out more as Josué had just told you liftlab.mit.edu. This is Benjy Kantor on Supply Chain Frontiers. Our guest today have been Dr. Josué Velázquez Martinez, Dr. Sreedevi Rajagopalan, and Fabio Castro of the MIT, Low Income Firms Transformation or LIFT Lab. Thank you all so much for taking the time to join us, it's been a really interesting conversation. Thanks for listening to this episode of MIT Supply Chain Frontiers, presented by the MIT Center for Transportation and Logistics. To check out other episodes of MIT Supply Chain Frontiers, visit ctl.mi.edu/podcasts. And for more on the center's, research, outreach and education initiatives, make sure to visit us at ctl.mit.edu. Until next time.