Tuck Knowledge in Practice Podcast: Can AI Write Accurate Online Product Reviews?

[This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of the Tuck Knowledge in Practice Podcast is the audio record.]

Praveen Kopalle: It’s not because of the few thousand francs which you’d have to be spent to put a roof over the third-class carriages, or to upholster the third-class seats that some company or other has opened carriages with wooden benches. What the company is trying to do is to prevent the passengers who can pay the second-class fare from travelling third class. It hits the poor not because it wants to hurt them, but to frighten the rich.

[Podcast introduction and music]

Kirk Kardashian: Hey, this is Kirk Kardashian and you’re listening to Knowledge and Practice, a podcast from the Tuck School of Business at Dartmouth. In this podcast, we talk with tuck professors about their research and teaching and the story behind their curiosity. Today my guest is Praveen Kopalle, the Signal Companies Professor of Management at Tuck and the area chair of the marketing group. Praveen is an award-winning researcher and teacher who studies marketing statistics Pricing, new products, promotions, customer expectations, and e-commerce. Lately, Praveen has been focusing on AI in two contexts. As a writer of product reviews and as a tool firms can use to set the optimal prices for their products. In this conversation, we’ll talk about both of these areas, including how AI can solve the hallucination problem. In generating online reviews and in the context of a book he’s working on called AI Driven Pricing Analytics. Praveen Kapali, thank you very much for appearing on Knowledge and Practice. We really appreciate your time. Uh, to kick things off, I wanted to talk about your recent paper on using generative AI to create accurate online product reviews. The paper is called Generating Accurate Online Reviews, augmenting a Transformer based Approach with Structured predictions. Uh, everyone is coming up with new ways to use AI these days, and marketers are no different. Um, one of the ways they have used it is to create product reviews that don’t require human input. Tell us about the opportunities in this space and the challenges that your paper addresses.

Praveen: Great. Thank you. It’s a pleasure to be with you. And this is a paper that I’m working with our colleague here at the Tuck School, Professor Prasad Vana, who teaches Analytics with me. He’s also with the marketing faculty. And the paper, the previous paper that we published, where we actually showed that machines can write human-like reviews. And now Generative AI has picked up on that. And the challenge in all in all of this is, gen AI as well as our previous model. It’s called the transformer model. They all tend to hallucinate. And I want to make a distinction between hallucination with an uppercase H and a hallucination with a lowercase h uppercase h. Hallucination will be with ChatGPT will say hello. Kirk and Praveen wrote this amazing paper when in fact we have not. So those can be solved by simple fact checking. Lowercase h hallucinations are more nuanced. For example, in our context, we have taught machines how to write online reviews in the context of wines. We’ve trained the machine on 140,000 wines from all over the world over a ten-year period, and it has learned how to write. And an important thing in in wine reviews, as a consumer of wines, is the taste of the wine. So the lowercase h hallucination is the review would say this wine tastes very blackberry. Well, how do you know the machine hasn’t tasted it? You know, it’s not in one of the training sets. It’s a new wine, so that requires some expertise. So unless you are a real expert in who knows about that wine, you won’t be able to realize that that that that lowercase hallucination, that’s the problem that you’re solving. It’s a big challenge in Chennai. That’s where domain expertise is important, to even notice that it is hallucinating that part. So that’s the challenge that we take here to see how do we address this lowercase hallucinations in online reviews in the context of wines? And we provide a generic approach on how to solve it.

Kirk: Okay. Yeah, it sounds really cool. Um, I’m wondering about the use case here. Like, what do you think this is going to do for marketers for firms to be able to deploy AI in creating this, this marketing content?

Praveen: Absolutely. One is as a consumer, let’s say you are at a restaurant and bartender is, or the waiter or the waitress are trying to sell a particular wine. And wouldn’t it be great if you can punch in the attributes of the wine? And out comes a very quote unquote authentic review, including the taste of that wine. It’s very useful because now the wines that you normally see are not we don’t have actual reviews, so you’ll have it as a consumer, it’s very useful. Or as a retailer establishment, they can say, hey, you know, while you don’t—we can present a pretty good accurate review for that or, or wine, makers themselves can use it as a very accurate description of the wine. And if you want to extend it beyond wines, if you look at retailers like, you know, Etsy or Amazon.com, there are thousands of products without any reviews. Wouldn’t it be great if there is an algorithm that can actually provide a quote unquote accurate review for these products. When known review sites exist that will help both the sellers and the and the consumers.

Kirk: Yeah. Yeah. Um, it sounds it sounds like I mean, it really could do a lot for helping companies market their products and helping people to kind of get a better understanding of what it is they’re looking at. Um, are there are there any like, besides the hallucination problem which you address in your paper, are there any potential pitfalls to using AI in this way as it could it fool people into thinking that they’re reading actual reviews, when in fact it’s generated by an algorithm?

Praveen: Good point. This raises, you know, some ethical issues. We, to see, you know, who is actually writing the review. So in in the interest of full disclosure, if a retailer were to, use this, they should say, hey, this is generated by a machine, but it is very authentic because it is, using a lot of appropriate data, which is what our methodology is to generate this. For example, in our case, what we that’s why we call it a two-step process. To do that we actually collected there is research in in wine economics that shows that the taste of a wine is a function of three things. One is precipitation. That’s rainfall in that that location in that time, in that time period when the grapes were grown. And second is the temperature during that time in that in that zone. And third is soil. So we collected data on these thousands of locations in over a 15 year period from all over the world. What was the minim rainfall, maxim rainfall, minim temperature, maxim temperature. And based on that, we’re actually predicting the taste. So one option, as if firms were to be using I think detailing that methodology should be Transparent, as opposed to implying that this is being written by an expert. But it raises some ethical issues. But what we say is that it should be fully transparent, that here is how we are able to do it. And here is the accuracy of our thing, which is what we established in our paper that using our methodology, the accuracy goes up from 23% to 87%.

Kirk: Yeah, it’s really amazing. I know that you, as part of the paper, you created a website, right? Yes. That allows people to compare human expert written reviews to AI written reviews. And the AI reviews are kind of different, I guess. Levels of sophistication. Right, right. Um, and so tell us a bit about like that, that process and making that website and kind of what you’re hoping to put out there.

Praveen: In that website is an app that we’re planning to create where if you can punch. If you punch in the attributes of various attributes of the wine, like the brand of the wine. Year of production and alcohol content. Price. Various attributes if you punch in that and outcomes are an actually an expert generated review. If we have that in the database and otherwise you can have you will have a ChatGPT review for that. And then we also have our own, review from our model. And in the cases where we actually have the experts review, when you compare our review with respect to the expert review, and we have done this study with 500 people on Qualtrics, and there was a significant our review was found to be significantly closer to the expert review in terms of taste, as well as the other aspects of it related to the ChatGPT review. On a one on a 1 to 5 scale, we have like one point ahead of ChatGPT in terms of that, how close we are to the expert review, and in the event in cases where, you know, that’s a proof of concept to show that, you know, our prediction of the taste is, is very similar to what the experts have said. And in the cases when expert review doesn’t exist, then you know one can trust our review more so than ChatGPT because ours is based on this two step, you know, structured prediction algorithm that we devised along with the transformer review.

Kirk: Yeah, yeah, I thought it would be fun to just read one of these reviews. Um, to that compares, you know, different, different reviews. So, so I pulled one off just kind of randomly. It’s for a Deerfield Ranch 2002 Chardonnay. Um, and the predicted taste is, is apple. And so the human expert review says, quote, tastes a bit old for a shard with a fruit fading, it still lively and acids, and has enough pear and apple flavors to satisfy finishes quick, and overall it’s not what it should be. So then the enhanced I review from that you that your model created. So that’s what this, this one says quote, there’s a lot of oak on this wine, but the underlying fruit isn’t there. It’s all about apples and peaches with a creamy texture. So that’s the enhanced version. I guess the kind of the most sophisticated version. And then there’s the unenhanced, which says, “There’s lots to like in this dry, fruity wine. It has flavors of peaches and cream and is balanced with crisp acidity.” And then there’s finally there’s the ChatGPT version, which says “It’s pretty long. Deerfield ranch is 2002 Chardonnay from Sonoma Valley, despite its modest 84 rating, offers an approachable white wine experience. At a price point of $25, this chardonnay reveals flavors of ripe pear, citrus zest and a touch of vanilla, creating a balanced and refreshing palate. Produced by Deerfield Ranch in Sonoma, California, this 2002 vintage presents an affordable option for those seeking a simple yet enjoyable white wine from the region.” So I think it’s kind of interesting to see the differences.

Praveen: Exactly. If you focus on the taste, look at our enhanced version picked up on the apple taste very quickly and very well. And in fact, ChatGPT has no mention of that. And even an Unenhanced version did not pick up on that. And that’s because of our structured prediction algorithm, where we went to the trouble of collecting the precipitation and temperature data from these 10,000 or 15,000 locations all over the world over a ten year period, because there is science that shows that the taste of the wine can be predicted based on Precipitation and temperature and soil. If we had gathered data on the soil, we would have been even more accurate. And you can see that in what you just read.

Kirk: Yeah. Yeah, it’s really fascinating. Um, why did why did you pick wine to study in this paper?

Praveen: Great question. So we wanted to pick if you think about the products that we use, some are a lot some have very many search attributes. That means if you do enough search on the internet about the product, you know, it will uncover all the underlying search attributes. Many other products have experiential attributes. No amount of search will uncover that. So why the experiential attribute in the wine is a taste of the wine. And that’s where expertise comes into play. What we wanted to show is that the, the lowercase hallucination happens when there is when there is no expertise. So this shows the value of expertise that, that, that if you can predict these experiential attributes that only experts can really, you know, uncover, or someone who actually tastes the products we wanted to use, that if we can prove our concept in that particular category, then other categories will follow suit much more easily. And that’s a plus. I like wine.

Kirk: Yeah. Fair enough. Um, so it sounds like the hallucination problem really is a result of sort of a bad data set or an incomplete data set. Is that true? I would say.

Praveen: Incomplete. Not bad. An incomplete data set. Because, you know, for ChatGPT, for example, it’s very hard for ChatGPT to collect, you know, precipitation and temperature data from all over the world over a long period of time because its use is limited. So it’s a more general purpose machinery. Um, and what we are saying is that one needs to augment it. And also it’s not even just augmenting, even if you gave all of this data to ChatGPT, I’m not sure it’ll be able to predict the taste because one needs to some expert. There needs to develop a predictive algorithm in a structured sense that he used this temperature and rainfall data to make a prediction for the taste, and use that as an input in your review. So I think that’s what I mean by the role of expertise becomes important.

Speaker3: Yeah. Yeah. So it’s both the.

Praveen: Data the domain expertise or theory that that it comes from.

Speaker3: Okay.

Kirk: So tell us some of the key takeaways from this paper from your perspective.

Praveen: So the key takeaway is while things like gen AI and so on are really good take away, you know, the lower level coding type of jobs. But this actually enhances the role of experts. And they are the ones who can actually come in and solve the lowercase hallucination problem. That’s part one. And second is that there is a role of theory that is important, that there is science that, you know, X predicts Y in whatever domain. And, and one needs to pay attention to that underlying, theory that dominates. And the third is based on these two factors. One needs to go ahead and collect that specific data that are essential to act as an input in a, in a, in a structured predictive sense that can address the hallucination problem.

Speaker3: Okay.

Kirk: And do you foresee this type of model being deployed relatively soon by firms?

Praveen: No. We have the technology. We have the proof of concept. And it’s not that hard to implement. That’s why we created the website, just to show that this is this is this can be easily done. And then we created this website just with the expertise from our tech research, computing people. So we have a lot of in-house expertise to actually be able to do it on our own.

Speaker3: Yeah.

Kirk: Very cool. Well, it’s a brave new world. And yeah, an AI is just a really fascinating, and it’s just, it’s just a great technology that we’re all kind of learning more about these days. And we’ll, we’ll see kind of how it works its way into our lives in different ways. Um, so let’s shift gears now a little bit and talk about, your retail pricing analytics. Um, remind me the name of your book again.

Praveen: It’s AI driven pricing analytics.

Speaker3: Okay.

Kirk: So we’re going to talk about your course and your book, AI driven retail pricing analytics. Um, can you help orient us? What? What is retail pricing analytics? What is AI driven retail pricing analytics? And why? Is it something that you really like to study and teach?

Praveen: So the question is, you know, why do we need AI driven pricing strategies? All companies, you know, particularly in the retail, need to figure out the prices for their products and services. And then the question is, you know, how do you how do you figure that out? There is a sweet spot for pricing. Or if you draw prices on the x axis and profit on the y axis, if you give the product away for free, you’re going to lose money. If you price it at an infinite price, no one is going to buy, you lose the money. So it’s a really nice inverted U-shaped curve. And then folks are trying to find that sweet spot in the middle in terms of what the right price that sort of, you know, maximizes profitability. And if you remember, if you if you recall calculus, you know, if you take the first derivative of a function and set it equal to zero and check the second order conditions and it will give you the maxim. I think that’s where that’s the starting point for AI. It’s driven by mathematics. The human math comes into play. Ai is an immediate cousin of mathematics. And that’s why you need, you know, AI driven pricing analytics. I just gave you a very simple example, but you can add bells and whistles to it, like competition and reference prices and intertemporal demand dynamics. And it can get complicated pretty fast once you have nonlinear functions and so on. So it is inevitable that AI is, you know, is going to be really good at solving these mathematical equations. Then you and I can do it with paper and pencil. I mean, that’s essentially the motivation for this.

Speaker3: Okay, okay.

Kirk: So it’s leveraging the power of AI to help firms set the best price possible.

Praveen: Absolutely, absolutely. That’s it. You know, and it’s retail because there’s a huge amount of data in retail. So if you think about Amazon, Amazon and Amazon knows much more about me than I know myself. Because computers don’t forget I forget. So it knows exactly when I’m going to run out of a particular product. So the data storage costs have gone down exponentially and the computing power has gone up exponentially. And the combination is fantastic, right? And that’s the idea.

Kirk: Yeah. So not only do we have AI, which is like a supercomputer that can crunch huge amounts of numbers and everything, we also have big data, which is feeding the AI. All this personal data from consumers. Right. The combination of those makes a really powerful, I guess, pricing engine, so to.

Praveen: Exactly. Okay. 

Kirk: So you teach an elective course on … Pricing and retail. Pricing analytics.

Praveen: Retail pricing.

Kirk: Analytics. Right. Um, I’ve sat in on some of these courses, and I think they’re really just fascinating, even from a layman’s perspective. Um, because there’s kind of like a blend of philosophy in there, like, like the philosophy of pricing and human behavior and kind of why people choose items that a certain price, but not a different price. Um, I’m kind of curious just from a personal perspective, like what you like about this, this whole field of pricing, like, do you kind of get into, like, the human behavior side of it, too? Do you like.

Praveen: Absolutely. What I like pricing and both researching and teaching pricing is that it’s an extremely interdisciplinary, topic. It brings in principles from economics, operations, marketing, consumer behavior, psychology, finance so on. And you need, in a combination of all of these factors to really figure out that sweet spot. So you mentioned consumer behaviors. How do consumers process price information. That’s really important. And for example, if I walk into a store and find a six pack of Coke on sale for $0.99, I know it’s a good deal. If it’s 4.99, I know it’s a bad deal. Why? Because I’m carrying these reference prices in my head, and I’m comparing this observed price with a reference price and then deciding whether to buy how much to buy. So these are this is how consumers process price information. And so in that sense it’s really important. Now we don’t know what’s going on in the black box of the consumer’s mind. But if we can build a pretty cool, you know, mathematical model which is a parametric representation or a parametric mathematical representation of how consumers are thinking about prices. And we can incorporate in this, in this AI model that I’m, that I’m, that I mentioned before. So that’s why I think the consumer behavior is a very important input to pricing decisions.

Kirk: Let’s talk a bit about the book. Um, who is it for and what do you hope to convey in it?

Praveen: This book is both for business students and for managers. And it, it has, both rigor and relevance. And the rigor comes from the AI driven pricing analytics in the background. And we are actually going our plan is to create an app for each of the chapters where companies can use that app by plugging in their data and out comes the result. Uh, so it’s very we want to make it extremely user friendly and also very applicable to both managers and business students alike.

Speaker3: Oh.

Kirk: It sounds great. Um, so one of the parts of the book, I mean, I’ve kind of seen an initial, you know, some, some initial text from it, but, you have an example in the book where you talk about the difference between pricing tactics and pricing strategy, which is, I think, an interesting discussion. Um, and you use an example to illustrate this. Um, from the 1800s, during the advent of passenger train travel. Tell us about what trains have to do with pricing strategy.

Speaker3: Great.

Praveen: Great point. In fact, this is a quote from an economist, called Jules Dupuis. Um, in 1847 of early train travel. Let me read that quote to you quickly and then highlight where the pricing is. It’s not because of the few thousand francs which would have to be spent to put a roof over the third class carriages, or to upholster the third class seats that some company or other has open carriages with wooden benches. What the company is trying to do is to prevent the passengers who can pay the second class fare from traveling third class. It hits the poor not because it wants to hurt them, but to frighten the rich. And it is again for the same reason that the company is having proved almost cruel to the third class passengers and mean to the second class ones, become lavish in dealing with first class passengers. Having refused the poor what is necessary, they give the rich what is superfluous. Sounds a bit crazy, but fast forward to airline travel. You know, when I signed my life off to Delta Airlines and I’m one of the Diamond Medallion customers, and they treat me great. When you walk into an airplane, if you make a left turn, life is good when you make a right turn and when you have to. When I make these 14, 15-hour, travels to non-stop travels to Asia, and you find when you when you take the basic economy, the seat doesn’t recline at all. Main cabin reclines a little bit. Economy comfort reclines more. And then you go to first class.

Praveen: You know, it’s completely flat bed and if you think about it, everybody is going from same New York to Mbai. The origin and destiny are the same. Everyone will arrive there in one piece. Except that one. In in the in in the in the basic economy, you know, may may not be there, but they’ll still arrive in one piece with some discomfort the and the and the plane is being it’s the same pilot. So the probability of landing is the same for both. What they have done is they have created different products and, and so that someone who made the left turn can look back and say, Thank God, it’s not me. In fact, once my flight from Mbai Delta Airlines flight was delayed significantly. They put me up in first class in Emirates I was sitting next to Salman Khan, who is a big Bollywood superstar and was chatting with him. And the flight attendant comes in. Sir, would you like to take a shower? I said, who needs a shower in a plane? But I took it, so it’s so great. So. And so what they’re trying to do is, you know, it’s they are creating a, a first class and an economy so that they are providing a very different product in the first class experience versus in the economy. And by doing so, they are all this is good for everybody because people who would otherwise not if they had a single price, people who would have otherwise, you know, wanted to travel by plane would not have averted.

Praveen: Now, by having the economy, you are actually able to cater to a segment that would not have otherwise. They can play in travel, on the other hand, by providing first class amenities and so on, you are able to charge more from the other passengers who are willing to pay more for a different product. So my mantra in this course is there’s only one strategy for any company, whether it is retail or B2B or B2C. It’s dynamic pricing. That’s the strategy. And then the tactics are how do you how do you do that. So in this in the airplane example or even the train travel, they’ve essentially created a third class with wooden benches and not even a roof. And this, that, that bogie is right behind the engine. And during that time the engines are driven by coal and all that coal dust and so on. Where does it come on? To the third class passengers. But they are still going to the same destination at a much cheaper price in that sense. You know, these folks would if they if the trains only had a single price, they would not have to afford that single price. Now they’re able to Take the train, travel. And then on the other hand you have the first class seats as well. So it’s a different product altogether, even though the same train, same origin, same destination, same driver. That’s the that’s the distinction between the tactics and the strategy. Strategy is dynamic.

Praveen: Pricing. Yeah, right.

Kirk: So the overall strategy is dynamic pricing, which caters to a range of different consumer segments with an ability and a willingness to pay. Right.

Praveen: Exactly, exactly.

Kirk: I think it’s interesting just to think about where else you see this in the economy, right? I mean, you see it all over the place.

Praveen: You see it all over the place. You know, that’s exactly change in different locations. You know, the same product is priced differently in Main Street versus outlet stores. Um, and the same product, you know, same customer sometimes, you know, the product is on sale on, on, on some days and it’s not on sale. So as time changes, the price changes and we see it all over the place.

Kirk: I mean, I’m a skier. So this winter, as I was thinking about your book. I was thinking about how do different ski resorts differentiate themselves. Right. So like some ski resorts have 150 trails and high speed lifts and really fancy lodges and lift ticket is 180 bucks.

Praveen: Exactly.

Kirk: And some places, you know, have slow lifts and old lodges and the snowmaking is not as great. And their ticket is 75.

Praveen: Exactly.

Kirk: And they and they each have their customers who love them or hate them.

Praveen: Absolutely, exactly. And that’s, you know, that’s that’s dynamic pricing. I think that’s, you know, and airlines have been really good at that.

Kirk: Airlines especially. Yeah. And they have their pricing like by the second.

Praveen: Almost.

Kirk: I want to talk about another section in your book, which I think is really interesting too. You talk about price elasticity estimation, which sounds kind of, technical. Right. But I think you make it you make it a little bit more approachable. You use this example of Aravind Eye care, which is.

Praveen: Oh, yes.

Kirk: Yeah. Which is which, as you know, is a clinic in India that provides ophthalmic surgery to anyone who walks through their door, regardless of their ability to pay. Um, tell us a bit about Aravind Eye Care. What makes it special and how you use it to teach the lesson about price elasticity?

Praveen: Yeah, it’s a great it’s a great, eye care facility they have as a patient. You walk in, they have two windows. You can either go to the free window where you don’t have to pay anything. It’s the surgery is free, or you go to the paid window. And it’s not like they ask for your salary statement or income tax returns or nothing, you know, and, and, and, and all the surgeries are done by highly qualified surgeons. Same quality. And there is no differentiation. It’s not like people who pay have a, you know, more experienced surgeon versus less experienced the same surgeon. And it’s not even like it’s a different operating room for them. It’s the same operating room. Everybody is operating in the same operating room. And so the first question is, you know, what stops someone to a paying customer to mimic a non-paying customer? Of course, one is, you know, the association with the mission. Their mission is to eradicate needless blindness. So there’s a lot of social cause. Uh, so 111 has the conscience, you know, but you can easily agree that it’s not 100% now. So and then the question then becomes for the paid customers, even though, you know, it’s the same operating room, same surgeon, they’re for example, their aftercare is going to be different. It’s going to be—it’s again the airline example economy versus first class. They go to an air-conditioned room. Their meals are more customized. Um, and the lenses that they are given are more, higher end lenses. So they’re creating a different product.

Praveen: Then the question is, so their idea is let me optimize my price to maximize my gross profit contribution from the paid customers so that this is a not-for-profit organization. Then I can use that money to care for as many non-paying customers as possible. The only way they can expand because they don’t take any donations or anything like that, they’re self-sufficient. So the only way they can really cater to that mission, in terms of having as many free surgeries as possible, is by maximizing this contribution from the paid customer. Well, how do we do that? Then they need to understand what is the price elasticity of these paid customers. That means if they increase price by one 1%, to what extent the demand will drop or they decrease price by 1% by how many percentage points demand will increase. If they can get a good understanding of that, then they can price their Product to the paid customers in the in the most profitable way, and then use that money not to give it to investors, but to cater to the mission of the organization. So it’s a great case study of how what I talk about in my pricing analytics class, as a pure capitalist of a profit maximizing pricing is all those principles, applies to a not-for-profit organization as well. And the heart of that is price elasticity estimation to be able to understand how sensitive people are to prices, because that’s an important input to get to that sweet spot that I talked to you before.

Kirk: Yeah, it’s really interesting to think about that elasticity. Um, basically you’re trying to predict how people are going to respond to changes in prices, right?

Praveen: Yes, exactly.

Kirk: What data do you use to make those predictions? And I’m sure AI plays a big role here. If they if they have all.

Praveen: Data. Yeah. And the question is, you know, what kind of data. So we collected data on all the paid customers nationally for Ikea for a particular type of new lens that they are, that they have come up with and used a proxy for all the similar types of lenses. We collected about 6000 7000 observations nationally, where prices are different and we were able to figure out then we can use machine learning algorithms, particularly supervised machine learning algorithm. Uh, in an estimate, a non-linear model, after controlling for many, many factors of, you know, the state it comes from, the type of product and so on. And see, you know, if there are any, you know, seasonal effects and where you can say how your log of those unit sales is impacted by the log of the price. And if you run that machine learning model, what you find is the coefficient of the log of price directly gives you that price elasticity. And from there, you can jump to optimal price pretty quickly, by something called as a Lerner index of market power.

Kirk: Um, so I think you used this example in your book because you have personal experience with Aravind Eye Care.

Praveen: Absolutely right.

Kirk: So you go over to India at least once a year, right?

Praveen: Exactly. And I spend time there, and Doctor Aravind actually came to Tuck as well. He was a participant in our, global leadership program to. And that’s exactly where I met him. And their organization was nice enough to invite me over and give me all the data that that we need to write this case, which I wrote with, with, with a few tuck students, actually.

Kirk: Oh, wow. I didn’t know that. That’s really great. Um, and you go over to India to be, like, a visiting professor, too, right?

Praveen: As a visiting scholar at the Indian School of Business. And the distinction is as a visiting scholar, it’s purely for research and not for teaching. I think that’s, you know, that’s a that’s a designation. So I go there, for to conduct research with India based companies or, you know and work with my co-authors at the Indian School of Business in Hyderabad, India.

Kirk: Well, Praveen, thank you for your time. It’s been great to speak with you.

Praveen: My pleasure. Thank you. Kirk, it’s been it’s always a pleasure working with you and talk to you about an answer, the insightful questions that you have for me.

Kirk: I’d like to thank my guest, Praveen Kapali. You have been listening to Knowledge and Practice, a podcast from the Tuck School of Business at Dartmouth. Please like and subscribe to the show and if you enjoyed it, then please write a review as it helps people find the show. This show was recorded by me, Kirk Kardashian. It was produced and sound designed by Tom Whalley. See you next time.