 All right. Well, good afternoon to everyone in Europe and good morning to everyone in the US. We're very excited to have Arun talk to us about his US project. And why don't you take it away Arun. All right, thanks Andre and thank you to everybody for the invitation to present to the seven or my apologies for thank you guys for showing up I mean I know I canceled last time so I feel like the following cancelling a seminar the probability that the speaker will show up the next time I feel a sort of slightly higher than like you know the base rate so this was sort of a reliable bet for you guys to make. So this is joint work with a post-it loss to the past who's an assistant professor at Fordham and was a PhD student here until two or three years ago. And Shri Khan Jagabatullah who's a colleague of mine in the operations management group of my department. So post-it loss I'm hoping will be calling in during this during the seminar so that he can take all the difficult questions. And, you know, has was sort of the primary driver of the experiment that will follow spend time with the company actually sort of implementing the experiment. But I'll get to that in a bit on the. I guess the motivation for me for this project was a deep interest in platforms over the last 10 years. There's a new mix of things that these platforms are doing that to me distinguishes them from being the kind of hands off marketplace that eBay was when it first emerged. And more craigslist continues to be where there is sort of minimal responsibility taken by the platform in facilitating the economic activity, and those platforms are really sort of matching markets or, you know, perhaps even sort of lighter touch than like you know sort of a coordinated matching market that is simply places to sort of aggregate supply and demand. There are more modern platforms, some of which you know I've written about a lot under the umbrella of the sharing economy but you know others sort of see more broadly as including, you know, YouTube and Amazon seller marketplace and do a wide variety of other things they you know facilitate search they facilitate discovery they facilitate matching some of them, you know, determine what the transaction prices going to be. You know the early markets would use auction mechanism some of them still do. You know modern marketplaces like Uber will often, you know, determine what the prices according to a preset algorithm. Most modern platforms facilitate some sort of you know transaction cost minimizing or facilitate the lowering of transaction costs by providing some kind of trust and risk management that, you know increases the likelihood that trade will actually occur. This could include insurance it could include like you know feedback based reputation system. It could include standardization of terms. And then, you know, some platforms like Amazon's marketplace provide logistics and delivery, or, you know, do or dash which matches restaurants with people who want to eat at one or want to sort of order food from them. There's a broad mix of things that these platforms do. You know, if it was a different audience I would sort of try and convince you that platforms are different and asked to sort of warrant independent study, but but I'm just going to sort of dive into what the, you know, what the context of this research problem is and like you know what the research was so. So, so this was motivated by an observation that many of these platforms seem to face what appears to be a somewhat unique challenge, which is that, you know, we like organizations always face the, you know, the issues of adverse reaction and model hazard when, you know, managers or owners delegate activities to agents. And there's a variant of that in platforms, except that, you know, there are the key distinguishing factor here seem to be that the nature of the relationship between the platform and the provider the entity that, you know, is on the other end that is actually the supplier in a sense, the seller the Airbnb host, the person who's renting out the car the person who's, you know, driving the car the person who the restaurant that is selling you the food. All of those I collectively refer to as providers in the rest of the talk. These are non experts. And so like you know this was sort of more salient to me in the case of Airbnb where we have millions of people who have become providers of short term accommodation who by and large don't have any formal training or prior requirements in becoming in in sort of providing this kind of short term accommodation. But they are depending on the platform, given a bunch of tactical decisions that they need to make. They are often like, you know, sort of inventory choice decisions they are often given the choice of deciding whether or not they accept a transaction. They are often, you know, asked to price what they are selling. And, you know, this could make sense if the provider has better local information. But, you know, sort of as platforms grow they tend to have potentially sort of superior, a superior ability to make some of these decisions. But, you know, just sort of sticking with the Airbnb example because it's familiar. I think a hotel chain which has a formalized revenue management system where inventory and pricing decisions are being made centrally with Airbnb. There's sort of much greater level of decentralization involved in these choices. And, you know, so one trade off here and like, you know, where should the decisions be made has to do with, you know, who has better information, and who has better capabilities. The platform has better capabilities ostensibly the, you know, provider has better information. And then a misalignment of incentives and this is not entirely new right I mean, you know, there are always misalignment of incentives between owners and managers but that exists potentially for, you know, a platform versus any individual provider. There's a pattern of availability sticking with the Airbnb example a pattern of availability and pricing that may optimize like you know the total amount of business done over like you know sort of the Airbnb platform, while not serving like you know the individual provider. And it's more than that the metric by which the performance of the platform like sort of ends up being like you know ends up the metric that ends up maturing the platform may be different from the metric that ends up maturing to the providers because the platform may not sort of hold the capital or certain costs that providers do. You know, I mean at the outset of the pandemic, within a few months of the pandemic sort of hitting both Uber and Hertz suffered a sort of very significant top line hit I mean they both saw revenue declines from their core business. Uber saw an 80% decline her so close to a 90% decline hurts, you know filed for bankruptcy soon after because you know their cost structure is very different they hold the leases on the cars that you know relying dormant Uber on the other hand partly because of course they have Uber eats but partly because of the their cost structure is fundamentally different. You know survived and thrived and so you know it both has to do with you know sort of the individual versus the collective and optimizing different objective functions potentially. And this is sort of really outside the realm of you know my core expertise but you know it's important to bring up that there are certain things that facilitate control by traditional organizations over their employees that platforms do not possess. There's like you know this corporate culture there's like you know sort of a set of instructions that an employee can be given as like you know this is how business needs to be done. That ensures a consistent brand experience now if you're an Airbnb host. Every time a guest stays at an Airbnb property Airbnb is putting their brand on the line. In a sense but they have no directive authority over the host in terms of like you know sort of setting standards in the same way that you know a Hilton franchise he might have for its employees to ensure a consistent band. And so there are new sort of provider management challenges. So every platform that I've encountered is trying to sort of balance. Both choose like you know what are we going to do and then you know sort of contingent on what we're doing. You know sort of how do we balance decentralization and control and like the final sort of stage setting point before I sort of get into the details of what we actually did and why it matters is that I've noticed that platforms seem to become less decentralized over time. And as they start out light touch they start out designed for rapid rapid scaling and sort of minimum liability and then you know as they get bigger they have more data. And therefore are actually sort of able to take advantage. They have greater expertise because like you know they've hired people they've hired data scientists. They may have a need for consistency as they grow. They may face competitors who are doing certain things that cause them to want to increase the scope of the decisions that they make. And so they have to manage this transition from being light touch to sort of being a little more heavy handed and that that that can be a challenge. You know there are a few examples of like you know successful and unsuccessful management of this kind of transition where the scope of decision authority that the platform took on changed or the platform takes on and the platform delegated changed sort of during the evolution and my first couple of examples are again of Airbnb even though my data is not from Airbnb. Which is there was a point in Airbnb's evolution where they introduced this feature of instant book. And if you think about it from the platforms for interview they have multiple choices like you know prior to this thing introduced in order for a guest to stay at a particular Airbnb, the host had to agree to accept the guest the guest had to request the host and the host had to agree. And this is unlike booking a hotel room where you don't have to ask for permission you just sort of book a hotel room. So naturally sort of a point of evolution in Airbnb becoming sort of like you know more standardized was, hey can we like you know switch to a model where people can book a property directly. And you could think of three choices one choice being now I just sort of stick with the status quo, allow hosts to have a final decision over whether or not they accept a guest. Second is switch to a system where like you know, once a guest requests a property it is automatically granted. And the third was sort of the middle ground where introduced as a feature where host can opt in to decide whether or not they want to you know allow guests to book instantly and so the reason why I'm sort of delving deep into like you know what is sort of a seemingly trivial feature on Airbnb is that it involves a somewhat complex set of factors for the platform to decide whether or not like you know they're going to make the switch how are they going to make the switch and there are multiple options available. You might think of Airbnb is having sort of gone through a similar process when they you know when you know when in my mind they realize that you know they could potentially sort of set prices in a way that was more efficient for some providers and for the platform as a whole. And again they had three choices they could either sort of let the host continue to price they could you know set the prices themselves and mandate that that was the way the platform was going to work much like Uber does or a wide variety of the platforms do or they could say well here is the price that we think you should be charging you can take it or leave it or you know there's a wide variety of other options and they ended up choosing the middle ground where like you know they offered this pricing system as a pricing tool and that was sort of a decision that they made relative to mandating the pricing or doing nothing. There are other examples of platforms that have made this kind of transition DD the ride sharing platform in China. You know used to allow drivers to accept rides or not they switch to a system where they took away the control from drivers and assigned rides to drivers lift early in its evolution did not have a price at all. You know there was a suggested donation that you know the ride was free and you could donate to the driver and they had to make the transition from a donation based system which is what they were in 2012 when I first used them to you know actually sort of setting a price and deciding like you know that that was another transition. Sidecar which was a lift competitor and an Uber competitor for many years and one of the pioneers actually the ride sharing space tried a different pricing structure where you know rather than saying it's up to the rider they tried a system in which it was up to the driver where rather than the platform setting prices each individual driver could decide on what their pricing was going to be they had like a little tool and so when you requested a ride you would get like you know multiple drivers each with their own wait time their car and the price that you would pay for the ride. It turned out that you know this was not a feature that the market wanted and you know it didn't work out well for sidecar. A different kind of transition you know was experienced by TaskRabbit which is one of the early you know back in 2012 when people talked about this new platform economy they would talk about Uber and Airbnb and TaskRabbit. TaskRabbit was then a sort of an auction marketplace where you could list whatever you wanted done and then you know providers would bid on it. They switched to a more centralized more uniform or standardized marketplace in 2014 where they had different categories. The users basically or the providers revolted I think two thirds of the providers actually left the platform following this transition and this was sort of a turning point in the evolution of the company where in my opinion they didn't really sort of live up to their early potential and eventually sold themselves to IKEA for you know about a hundred million dollars which is you know successful exit but still not compared to some of the peers that they had early on. So anyway those are many of the examples that I think sort of motivate our study which is you know when you are thinking about making this trade off between centralizing and decentralizing a decision in a platform setting and you have a current status quo where a certain set of decisions are being made by the platform and other decisions are being made by the providers and you have to make this transition. You know what are the kinds of trade offs that you face and so the partner that we worked with was a leading peer-to-peer car rental platform which shall remain unnamed. The business that this platform is in is in facilitating the rental of cars owned by providers to people who want to rent cars so it's like Hertz or Avis you go on the platform you open it up you say that this is the date that I want this is the window of time that I want the car for and then you're given a bunch of options and you're also given a bunch of locations these aren't of course car rental counters these are where the cars are actually located. I'm sure many of you have used these car rental services. The key difference of course between this platform and something like Hertz or Avis is that the cars are not actually being rented to you by the platform they're being rented to you by a provider on the other end who owns the car. A majority of these are owned by individuals who are for example renting out their second car. There are also a number of small businesses that might have a fleet of like you know five to ten cars that they are then providing on the platform. And you know there are some some sort of other exceptions for a while there used to be a service in the Bay Area called City Car Share. I think they had like 300 cars for rent. There was a period where they started to offer this car rental entirely through the get around platform. I don't know if that is still the case but you know I did notice that they stopped sort of offering these cars directly and started operating through a platform so vast majority of these are individuals providing their personal car or second car but there are also some other providers. And you know at the time that we studied the platform the status quo was that a provider would provide its availability the vehicle availability they would set the price and they had transaction approval in the way that I described the sort of Airbnb pre instant booking where you know the person who wanted to rent the car would say I want to rent the car and then the person who owned the car could say yes or no. So they could decide whether a requested transaction was actually approved and this sort of matters a little in what follows. If you're a renter you search you compare and choose and so they are providing some of that matching and discovery capability the platform. They're also providing an access technology. This is the technology that you know at least I first saw with zip car where you could sort of in a decentralized way open the car without actually going and picking up a key from somewhere. In the case of these peer to peer car rental platforms that is typically with your smartphone. Now you will unlock the car and they also provide some dispute resolution. They have a rating system for both renters and owners and they provide insurance. And that was a critical innovation for the early movers in the space partnering with an insurance company to create an insurance product that actually you know allowed me as a driver who had my own insurance but was not insured to drive your car to drive your car while sort of like you know I can't drive it under my insurance under your car. And so this was sort of a peer to peer car rental insurance product that you know relay rides which was an early mover here and get around as well. These two early companies in the United States created an insurance product that facilitated this. So those were their responsibilities at the time that we started this study. And the providers controlled availability pricing and transaction approval and of course made the choice as to whether or not to offer their car. And this is what the pricing tool looked like for a provider. They would move a slider in order to control how much the car cost per hour. And that would translate into a daily and a weekly rate in a pretty sort of systematic way the daily rate was 10 times the hourly rate and the weekly rate was six times the daily rate. So this was their one sort of tool that they had. And some providers wanted you know sort of the ability to price differently on weekdays and weekends. But that was the status quo. The you know prior to us running the experiment. And I don't remember exactly what window of time this was for but it was certainly before we sort of like you know went through any transition. The two two observations about activity on the platform. One was that vehicle utilization as measured by like you know the number of hours a vehicle was rented divided by the number of hours that it was listed as being available on the platform was fairly low. You know a vast majority of cars were not rented for a majority of the time that they were made available. Part of this could be because like you know some people would list their cars as being available 24 hours a day. But you know it still seemed like a you know given the promise of peer to peer car rental as being like you know hey we'll sort of you know monetize your idle capacity it felt like you know utilization was low and people within the company felt that they could increase utilization. And the second observation was that people didn't change prices very frequently and so if you measure the number of price changes per month that a typical provider would perform. You know it was a relatively small number of people who were making any changes at all. The platform felt that they could increase utilization they hired a team of data scientists were saying hey we can do better than this and the major competitor of the platform centralized pricing like you know sort of prior to this whole thing beginning. And so in 2015 we did a sort of an informal pilot I would report on the results of that but it sort of taught me a few things. This was voluntary people like you know we're asked whether they wanted to participate in this a handful of people participated in it half of them dropped out. When we asked them why did you drop out some people just wanted to sort of control their pricing. And we're uncomfortable with the platform setting the price the two more interesting reactions were there was a set of people who felt that the platform was pricing their car too low. Not from a revenue optimization point of view but from the point of view of hey if people think that this is an inexpensive card and they're going to treat it badly. And so I want to keep my price high because I want to sort of ensure that there's high perceived value so that someone is responsible when they use my card they think it's a $35 an hour car, not a $10 an hour car. Which was sort of interesting you never really sort of dug very deep into that and our experiment but it's sort of an interesting direction for future work and there was another set of people who dropped out because you know the reaction was hey like you know a mile away there's another Camry that's being listed for 14 bucks an hour and your pricing my thing at seven it's not fair. So anyway when we the platform was keen on rolling out a centralized pricing system they wanted to sort of they had decided that this was going to be the pricing structure that they would vary sort of hourly rate. Based on our of day and based on day of week in sort of a continuous fashion. And you know we believed that you know a middle ground might be a good thing to explore I mean based on you know the experience of other platforms and so on. And so the experiment eventually consisted of three groups about 70% of our sample was you know left exactly the way that they were they were you know they had complete control over their pricing. About 15% of our group. pricing control was taken away from them completely and so this was the first treatment group. And the second treatment group pricing was control was taken away from them but they had some ability to you know sort of raise or lower the price. Like you know sort of based on and let me actually show you the interface. So this was the status quo. So this was the status quo. The first treatment group where the pricing was completely centralized. Did someone have a question. Okay. Where pricing was completely centralized the interface look like this and so sort of a continuous price depending on time of day, and it was a seven day sort of pricing schedule. And what the user would be shown was here's sort of the sort of the integral under the curve price that would be charged if you rented this out for 24 hours. And you know depending on the window that was requested the price was set according to sort of what hours this overlapped with. So this was sort of a non trivial pricing centralization of pricing it wasn't sort of simple. And the partially decentralized was exactly the same thing except we also gave them a slider in which they could sort of raise or lower the curve by up to 30% on both sides. And so they could either sort of you know if they set this in a way that it was up by 20% then every price on the centralized schedule was 20% higher than the centralized algorithm prescribed it to be if they said it at say 30% lower than the entire curve was lowered by 30%. Okay, so these were the three treatment groups centralized partially decentralized and the control was status code T zero. So we ran this experiment for eight weeks and we studied a number of outcomes and the results I'm going to report to you are sort of very straightforward of the form. You know let's look at the treatment effects for each of the two treatment groups and compare them to the control. And so the graphs I'm going to be showing you are, you know, should be interpreted as like you know by what fraction. This says percentage but it's really fraction by what fraction was the outcome of the treatment group different from the outcome of the control group. And so the first set of results are for the group treatment one where control was completely taken away. And over that eight week experimental period for this treatment one group revenue went up by a little over 50% on average. And so the revenue that was generated by a provider on average was substantially higher than like you know for a control. The revenue increase came from a dramatic increase in the utilization. And so the utilization of the number of hours rented for. Actually, let me say that more precisely the number of hours that a provider who was in the group treatment one rented their vehicle out relative to one in the control group was like you know about. 130 to 140% about like you know a factor was higher than the control group by over a factor of two. And so, from those metrics, I mean like you know the centralization of pricing seem to be a success. You know there was a dramatic increase in revenue and a dramatic increase in like you know sort of the volume of business that this these providers were getting. On the other hand there was also a very significant negative reaction from people in this in this treatment group and this was a great interest to us because part of our interest was in like you know sort of like you know how should these transitions be managed. And so compared to the control group, the churn rate of the treatment one group went up by 30%. Can I ask, can I ask a question, it's Jack Kramer speaking please. I mean, of course you would have very strong incentives to put down yourself at 95%. I mean, you're playing a game against a whole bunch of people. Some of them, and you've been told how the others are pricing. So, I assume that lots of people must have price just below what just below one because you knew that the people who are in the centralized group was at the slider set that exactly at one. Well, it seems to be a strange way of doing things I mean because some sense you have a game, which is being played by the different renters. And one of them you're, you're announcing the strategy and you're preventing. I mean, one of them is the first mover but you've got the first mover disadvantage seems clear to me. Yeah, you'd imagine that I mean there are a couple of. So, I think the graph you just showed us for about half a 10th of a second was showing this wasn't it. Yeah, it was actually showing just above I mean this this was, you know just illustrating the next slide. Yeah. Next slide. Next. So this is no no. Yeah, so the bottom thing is just showing that people are pricing just below. Yeah, sorry. Yeah, no, I mean, you know that your intuition sort of makes a lot of sense that you anticipate that the people who have a little bit of control would just sort of set the slider a little below. And so they are pricing just below like you know the 15% who are in the treatment one group. One thing is that you know the participants did not know that what these fractions were. They simply informed that this is the group you're in. And the second thing is that, you know, the, the entire set consists of, you know, this is a small numbers experiment there are, like, you know, sort of thousands of providers. There's a great deal of heterogeneity in the product. You know, not all of these are exactly the same car. They are all in different locations. You know, it's not entirely common that you have an identical competitor. Like, you know, who is a perfect substitute for you. So these are sort of differentiated on many dimensions. Thank you. I just wanted to let you know that you have five minutes left. So, Okay, use your time well. All right, so, you know, so the, the Well, what we observed in terms of non price reactions by participants, you know, you take pricing control away from them so they react with whatever sort of levers they have. There was a substantial increase in the rate of exit from the platform. There was a substantial decrease in the extent to which they provided availability and so you reacted to, hey, the pricing is not what I like, but like, you know, so I'm going to pull some of my inventory. And there was also a substantial increase in I'm not sure why this isn't showing up on the cancellation rates of like, you know, sort of this, this group, this group had substantially higher cancellation rates than the control. Now the interesting contrast comes in comparing treatment group one with treatment group two, because in some ways these were two of the three choices that the platform could make do nothing. Like, you know, centralized pricing completely and then provide partial control. So with partial control, what we noticed was that like, you know, the revenue impact was roughly the same. People who had partial control saw the same revenue increases on average. But those revenue increases came with a substantially lower increase in the number of hours rented. So there was some sort of optimization they were doing that was allowing them to sort of generate the same revenue with fewer hours rented. What was more interesting from a managerial point of view is there was substantially lower significantly lower exit rates the churn rate of this group was substantially lower. The availability curtailing of this rate was also substantially lower. And this is not going to show you the cancellation rates for this group were also substantially lower. And so the non-price reaction of this group that had partial control was you could say as anticipated, significantly less adverse for the platform than like, you know, the centralized group. And so you saw a revenue increase, which was good for the platform, but you saw these adverse reactions that in many ways the decision at hand was like, you know, sort of what do we do next. And so, you know, just to sort of summarize this contrast, I mean, like, you know, the exit rates for the treatment one group was significantly higher, the availability sort of cutting was significantly higher and the cancellation rate was significantly higher for the people who had no control. And giving them partial control seemed to sort of lower the churn rate, lower the rate at which they were sort of cutting inventory and lower the rate at which they were canceling like, you know, requested transactions. So in my last two minutes, let me sort of try and understand sort of help you understand, you know, why do we see these differences. I mean, there are two theories that we've got sort of, you know, we've partially tested in some sense. One is the theory that, you know, the cost structure that is or like, you know, the optimum that the objective function of the providers is different from the objective function of the platform. The platform wants to maximize revenue, because they take a street commission on revenue. The provider on the other hand is bearing different kinds of costs. One cost is simply a per transaction cost, there is some sort of bring to market cost that you might think of as associated with like, you know, sort of checking out the guest. Like, you know, so deciding that you're going to sort of be engaging in this front section. And the second cost is a increased depreciation cost. I mean, if someone is driving your car, yeah, they're putting in the fuel, but there's a greater level of depreciation that you are bearing per mile driven. And so we did notice that for the treatment group one, the revenue they were generating per mile was significantly lower. We didn't have a way of directly estimating what these costs were. But one explanation for this difference in sort of like, you know, the non-price reactions of treatment group one was that the sort of like, you know, the net profit that they seem to be collecting, without being able to adjust price, sort of adjust demand, were significantly lower. They were making less money per mile and less money per transaction. And which is sort of why they were reacting in that adverse way. The second difference is more of a sort of a behavioral difference. We surveyed them after the experiment was over. And we found that people in treatment group two were significantly more satisfied with the intervention than people in treatment group one. This is despite the fact that they generated exactly the same sort of level of revenue increase on average, roughly the same revenue increase on average. And so like, you know, my colleagues in the management department have taught me about something that like, you know, this concept of a psychological contract that emerges between employees and employers, which is like, these are not contracted on explicitly, but these are expectations that I have about how I'm going to be treated or what the rules of engagement are. And when you change those without even if it's not contracted upon the employees can react adversely to that kind of change and it's possible that there was something akin to a psychological contract violation that was perceived by providers when this pricing change was made. In any case, Arun, you're out of time. So can you get to the conclusion please? Sorry to interrupt. Sure. And just to your point, I mean, most of the providers actually did not change the slider at all during those eight weeks and maybe in the Q&A session we can sort of explore why. We sort of, I'll skip over those but like, you know, the platform eventually rolled out the partial control like, you know, option. The eventual sort of marketplace impact was lower than the measured treatment effects. Like, you know, this is sort of something that we documented a lot in other studies as well. But our estimates were that like, you know, there were revenue increases of about 25% on average post rollout. And we did some sort of simple difference in difference analysis that is in the paper that like, you know, sort of documents that. And actually, like, you know, what I learned from this study was that, you know, when you're designing these mechanisms and centralizing things with the platform, managing the transition matters just as much as like, you know, designing the optimal mechanism. And there are different factors that sort of mediate this balance between centralized control and decentralization objectives, like, you know, sort of contract violations and so on. So, you know, the platforms are well advised to sort of use nuance when balancing these provided incentives with that form incentives. I'll stop there. Apologies for taking three minutes more than I was supposed to. So thank you. Thanks so much for in those very interesting. So Shane, now it's your turn for the discussion. All right, let's see. Can you hear me okay. Yes, good. Okay, great. Thanks. I learned a lot from Arun's work. It's really a pleasure to be asked to discuss at this time and thanks for getting me the paper well in advance. And, you know, I really appreciate that. And, you know, my job here is to gin up some questions. So, let me just say who did a great job of summarizing the motivation and the, you know, the basic framing and results. So that makes my job a little easier. So let me go straight to a controversy. Let me try something. If you were, if you were trained in the Chicago tradition. You know, one of the earliest principles you learn in the Chicago tradition is price controls are doomed to failure. Centralized price controls never can overcome the benefits of decentralized decision making because so many of the participants in a decentralized marketplace have far more information about their own reservation than they use and so on. Then any centralized decision maker could have and the decentralized price control, I say price control because that's where many much of this intuition comes from, like a government is bound to fail. It's bound to get the prices wrong. And that's going to induce distortions. You know, we've come a long way from that initial intuition and the paper. In fact, I think one of the most interesting things about the papers that made me think really hard about why platforms have made us reconsider that in that Chicago and tradition. And I'm going to rephrase some of the things everyone already said, but as answers to the Chicago challenge. Okay, so one of the ways the paper says you can answer this challenge is that the platform can design a menu. There's even a menu for pricing, such that it simplifies user search, and that benefits users so much that it generates adoption for the platform as a whole and therefore you get larger sales, and whatever might have been lost to any individual provider from losing pricing controls is gained overall for the platform be by having more users and a larger user base. The second thing related to that is you could have a commitment to a pricing device that then also generates a user response that's positive and then again you get the same dynamics of a larger user base and what any individual supplier might have lost is more than overwhelmed by user adoption. Now by analogy, the paper doesn't really address that with empirical facts, but it seems like it could merely from just telling us if there was an adoption decision, maybe it was too short term to address because the experiment was too short. But I would love to have seen something like whether repeat users came back more frequently or something like that. Or whether actually the drop in options maybe was a positive for some people because it makes search easier. So that that's one provocative thing I could say here. Another provocative thing another way you might respond to the Chicago challenge and Arun sort of said this explicitly so I can say it's great so I can say this is really quick, which is its specialization. Suppliers or amateurs don't have the, you know, specialized knowledge to price appropriately and the centralized platform is is a specialist it's not an amateur it's in this market. Repeatedly, and so collects information so much information that it can make a better estimate of what prices ought to be than any amateur on average can make. And even if it was imposing a distortion, it would more than make up for that by having a better estimate that's the kind of model that we have in mind. You know, the paper actually does start to make that argument and it seems to me that if that's an argument you want to say a couple things you want to talk a lot about trust, Arun brought it up but it didn't bring up empirically. That would imply probably the suppliers who have less experience with drop first, or exit first or you would get something like experience on the platform would, you know, it might be continuous and experience willingness to use the to submit themselves to centralized pricing. I go to a third explanation that way you would answer the Chicago challenge and and and actually the paper talks a lot about this and Arun alluded to it you were running out of time. I actually found this very interesting. And it's the it was the explanation that it's some sense the hassles is the way I understood your managerial explanation the hassle to a supplier here, bringing the product to market is so high that there's a upfront hassle, and that they're amateurs, and they have to learn what they need to do. And then there's a per transaction hassle, again, because they're amateurs, and then they have to continually find out what the market's doing this week or that week. And that the hassle itself induces them just to stay out of the market because it's unless the price is high enough. And this this explanation I thought was actually really rather interesting and it Again, it would look for you would look for evidence of it in things like experience, or maybe type of car or value of car. So the lower value cars just wouldn't it wouldn't be worth the hassle wouldn't be worth the fixed cost, but with a high value car or high margin car you'd be worth the trouble. So I was looking for something like that as as further evidence for that explanation. And then finally I'll finish. I got five minutes right that's what you wanted. Right, you know with thanks for your patience. I finished I love the last explanation about the ones who had some pricing flexibility seemed to have been happier. And that's, which is so interesting. And what that reminded me of and is this interesting tension you find between manufacturers and district distributors where the manufacturers have a suggested retail price and the distributors, you know, can or can't stay with this or discount off it or, you know, margin on it and get and try not to be informed about it. And as you probably know there's this tension in that relationship and has been for, you know, no under scene for decades. And I sort of reminded me when they had like the suggested price from the central planner, and then says well you can vary 20% here, 30% there, you can go either way. It's, it's as if a buyer, I mean a supplier supposed to know well I have too many dens of my car so I'm going to discount it or I consider my car to be high end so I'm going to make a higher margin or something. And I didn't quite understand exactly how to think about that. But it seemed like that was really pretty interesting and an interesting, you know, place to finish, because the suppliers and do much happier with that. So, that was it for trying to be provocative. I, you know, I thought it was a wonderful paper I really suggest anybody read it because I learned a lot by reading it. Okay, thank you Shane. It's a, I mean, all of these are sort of good competing explanations. I'll just clarify a couple of things. I don't know if, you know, if Apostolos is online he may be able to address this. I don't recall if we had looked at sort of the variation in tenure of people who actually exited. You know that's that's actually sort of I think there's a lot we could learn from that. I was just sort of quickly glancing at the current draft of the paper to see if there was anything in there and. But but but you know if we do, if we haven't that's certainly sort of something that that we look into. I mean the papers that sort of an interesting point where it's sort of been accepted for publication conditionally but you know and the conditions are weak but it also means we can sort of make whatever changes we want. Before it before it actually gets published one one one would hope. Like you know within within bounds of course. Now the. I think the real explanation, you know at least for why there's this revenue benefit is twofold. One is the, you know the amateur versus specialist. You know there is certainly a sophistication in pricing that you know sort of could. An Apostolos isn't is actually sort of online he has managed he's in the chat box and is willing to take sort of all the difficult questions about methodology that you guys might have. And he says we did look into tenure. No different same with car value that like you know there were no sort of systematic patterns of difference. You know, within the those who exited and those who didn't. This expertise but it's also, you know in some ways this misaligned objective story which is, you know the platform is optimizing revenue, and this is the case with a lot of platforms right I mean this is the whole joy of the platform model. Like you know scale without mass like you know sort of the asset like world in which you have a very, you know, a simplistic way to think about it is that the platform is maximizing revenue, while the provider is maximizing profit. So, you know that that that was what motivated looking at sort of revenue per mile as sort of a rough indicator of like you know sort of what are these variable costs that people are bearing and what are the fixed costs that they're bearing and so you know I I think that the difference between the two transaction, the two treatment groups as being a combination of, you know, making, you know, the people with pricing control had the ability to make adjustments so as to align. So to sort of take out the transactions or to sort of change the price in a way for those for whom price was too low, given heterogeneity. In a way that kept their revenue at where it would have been while sort of taking out the unprofitable transactions, or the hugely unprofitable transactions. And given that a lot of the providers didn't make any changes at all. This sort of psychological contract violation thing that like you know I'm just happier when I have control. So it's, you know, I wish we had the ability to sort of more clearly tease out these two alternative explanations the sort of the quote unquote economic explanation and the behavioral explanation but maybe that's something that I can sort of toss back to the audience as like you know something to delve deeper into in the centralization versus decentralization studies in follow yeah. And David to your question, yes, this does. It seems like yield was much lower for T one. I'm not capturing part the different incentives of the owners and the platform, you know the treatment groups were ex anti identical, you know, on a wide variety of measures that we sort of tested them on. And so, you know, they start out with the same incentives it's just that the people who have the pricing control are able to change price in a way that is aligned with whatever their objective function is and, you know, as a consequence or overall, like, you know, have to resort less to the non price reactions like you know, cutting availability or, you know, sort of canceling transactions.