Abstract:
We consider the non-cooperative choice of arrival times by individual users, or customers, to a service system that opens at a given time, and where users queue up and are served in order of arrival. Each user wishes to obtain service as early as possible, while minimizing the expected wait in the queue. We first analyze this problem within a simplified fluid model where we identify the unique Nash equilibrium arrival profile and show that the price of anarchy of the system equals 2, under the assumption of linear waiting and time to service costs. We then address the non-asymptotic stochastic system, assuming a finite number of homogeneous users and exponential service times. In this setting as well we show that there exists a unique Nash equilibrium, which is symmetric across users, and characterize the equilibrium arrival-time distribution of each user in terms of a corresponding set of differential equations. We further establish convergence of the Nash equilibrium solution to that of the associated fluid model as the number of users increases to infinity. We also show that the price of anarchy in our system exceeds 2, but approaches this value for a large population size.

Abstract:
All the major economies in the world are now dominated by services. Developed economies have also become true “information economies”. Some of these changes can be justifiably described as “service industrialization” and they are visible from the sector and firm level down to processes and tasks. There are very significant opportunities for research at all levels. I will present an overview of our current research in areas including economic evolution, technology and competition, collaborative services, service experience and process design.

Abstract:
Scientific pricing incorporates process and data-driven intelligence into a company’s selling process. It fundamentally uses the law of supply and demand; however, its applications vary considerably based on many factors, including product characteristics, data availability, business requirements, managerial objectives, and competitive environments. High-tech products are characterized by rapid price erosions, super high price sensitivity, short and unpredictable lifecycles, strong tendency of cannibalization and extreme competition. This presentation provides a brief overview of scientific pricing and discusses key challenges in its application for the high-tech products.

Abstract:
We consider a flexible policy for a single location, single item inventory under rather general assumptions. Under the (Q,R,S) policy, orders are placed in multiples of Q units, at review epochs that are spaced R time units apart, and such that after ordering the inventory position is as close to but no larger than S units of product. By choosing Q = 0 (or 1 for discrete demand), the policy reduces to a base stock policy, and by choosing R = 0, continuous review results. Hence the general (Q,R,S) policy includes periodic review base stock and (r,nQ) policies and continuous review one-for-one and order point order quantity policies as special cases. The model therefore allows fair comparisons of costs between all these policies. We provide a unified derivation of the inventory related costs for the (Q,R,S) policy and completely characterize its (joint) convexity properties. We also study review and ordering costs incurred under the (Q,R,S) policy, and compare optimality conditions and costs for the periodic review base stock policy and continuous review order point-order quantity policies. Finally, we provide some real life examples where it is natural to use (Q,R,S) policies.

Abstract:
We study how the risk of disruption impacts the performance of a supply chain with a goal of identifying policy recommendations for designers. A classic example is that of a fire in a supplier plant shutting down production at Ericsson for 30 days and resulting in $400 million of lost revenue. There are two questions we focus on in a context like this. The first relates to the fact that the risk of disruption is very hard estimate. There is now way that Ericsson could have estimated the probability of this fire. How should it account for estimation error when designing its supply network? The second question relates to the idea of “integrating” supply chains using policies such as common parts, single suppliers, centralized inventories. In each case, the supply chain action improves performance when faced with recurrent risk (for example demand fluctuations) but makes the supply chain more fragile when faced with disruptive risk. We identify factors that influence the fragility of a supply chain. Sunil Chopra (joint with Achal Bassamboo, Mark Daskin, Michael Lim)

Abstract:
We consider an approach for estimating substitute and lost retail demand when only sales transaction data are available and not all products are available in all periods (e.g., due to stock-outs or availability controls imposed by the seller). Our method combines a multinomial logit (MNL) demand model with a Poisson model of arrivals over multiple periods. The problem we consider is how to jointly estimate the parameters of this combined model using only sales transaction data. Our key idea is to view the problem in terms of primary demand (or first-choice) demand -- that is, the product choices that customers would have made if all products were available in all periods -- and to treat the observed sales as incomplete observations of primary demand. We then apply the expectation-maximization (EM) method to this incomplete, primary demand model and show that it leads to a simple, highly efficient iterative procedure for estimating the model which provably converges to a stationary point of the log-likelihood function. We illustrate the estimation procedure on several industry data sets and discuss extensions of the approach to non-parametric models of preference. Garrett van Ryzin (Joint work with Gustavo Vulcano, NYU and Richard Ratliff, Sabre Holdings)

Abstract:
Organizational knowledge management has become a critical competitive capability for professional services (engineers, doctors, lawyers, etc.). Professional service firms are grappling with how best to manage knowledge flows within the firm; and the flow of knowledge in and out of the firm. Understanding knowledge flows and leveraging it translates to faster (and cheaper) client solutions. Specifically, I will be talking about two on-going projects. The first involves the dynamics of the transfer of knowledge within business units of the firm. If one business unit of the firm is working on a client problem, for example, how well is it able to leverage not only all of its prior work, but by other business units in the firm. Another related project investigates if professional service firms are able to leverage the experience of their sub-contractors. They use but do not own the work product - can firms leverage this type of knowledge? The focus of this talk will be on the issues related to leveraging such “distributed knowledge” within the firm via empirical estimation of learning curve models. We will report preliminary results based on our experience with an Architectural/Engineering design firm. Ram Ganeshan (joint work with Tonya Boone and Robert L. Hicks)

Abstract:
In this session I will describe some of the projects that I’ve been working on while at ISB. They include the following: • Sustainability in Healthcare: Towards a Theory of Environmental Capabilities. This project draws on research in operations strategy and resource based view of the firm to develop a model that describes how healthcare organizations are developing capabilities that support sustainable operations. • Investigating the Alignment of Sustainability Perspectives in the Retail Fashion Supply Chain (with R. Batra). This project empirically examines the congruence in attitudes about sustainability among consumers, retailers and designers. • Exploratory Study of Sustainable Luxury (with R. Batra). This project uses qualitative methods to define sustainable luxury, identify the tradeoffs and the implications for service design and delivery.

Abstract:
After a decade of rapid globalization, there is increased interest to bring offshored production closer to home. A key driver behind global sourcing has been its ability to increase margins by using cheap labor and materials. However, those benefits do not come for free: offshoring suffers from higher transaction costs, long and complicated logistics that are sensitive to increased volatilities in demand, supply, currency exchange rates and oil prices, and political and environmental criticism. In addition, responding to customers' new-product requests, shorter delivery times, and swift corrections to improve designs and quality has magnified the need for responsive and agile supply chains. A complete reversal to local sourcing, however, is unlikely and ill-advised. Indeed, the concepts of global and local sourcing are not mutually exclusive. Rather, the combined use of multiple supply sources, each of which is different and possesses unique advantages, might be better than any single sourcing strategy. The main contribution of this paper is to provide the first exact and analytically-tractable analysis of a dual sourcing policy that is easy to implement. This policy allows us to design an ordering policy that allocates the order volume to both sources so as to optimally trade-off cost and responsiveness. We present a tight approximation for the optimal volume fraction ordered from the slow but cheap source, which we refer to as the strategic base or offshoring allocation, and its corresponding total landed cost. The strategic allocation is characterized by a simple formula that provides structural insight on the impact of financial, operational and demand parameters, and a starting point for data-driven decision making.

Abstract:
Concierge Medicine is the most visible implementation of a new trend of introducing a fee-based premier option in a service offering. This is usually done for the purposes of segmenting customers based on their willingness or ability to wait for the service. Using a realistic yet representative model setup, we perform a detailed analysis to (i) determine the conditions (fees, cost structure, etc.) under which the concierge option is profitable for the service provider, (ii) quantify benefits accrued by the premier customers; and (iii) evaluate the resulting impact on the other customers. We show that, under a wide range of system parameters, introducing a concierge option benefits everyone and also compute the magnitude of these benefits. These benefits are larger when the variance in the customer waiting costs is high and the system utilization is high. We complement these results with data on the adoption of MDVIP (the most popular concierge medical service in the US) service and show that the service has been adopted in areas where the median income is significantly larger and the population is older. Both income and age are good proxies for the waiting cost that a customer attributes to waiting for receiving medical service.

Abstract:
With the recent explosion of choices available to us in every walk of our life, capturing the choice behavior exhibited by individuals has become increasingly important to many businesses. At the core, capturing choice behavior boils down to being able to predict the probability of choosing a particular alternative from an offer set, given historical choice data about an individual or a group of "similar" individuals. For such predictions, one uses what is called a choice model, which models each choice occasion as follows: given an offer set, a preference list over alternatives is sampled according to a certain distribution, and the individual chooses the most preferred alternative according to the sampled preference list. Most existing literature, which dates back to at least the 1920s, considers parametric approaches to choice modeling. The goal of this thesis is to deviate from the existing approaches to propose a nonparametric approach to modeling choice. Apart from the usual advantages, the primary strength of a nonparametric model is its ability to scale with the data -- certainly crucial to applications of our interest where choice behavior is highly dynamic. Given this, our main contribution is to operationalize the nonparametric approach and demonstrate its success in several important applications. Specifically, we consider two broad setups: (1) solving decision problems using choice models, and (2) learning the choice models. In both setups, data available corresponds to marginal information about the underlying distribution over rankings. So the problems essentially boil down to designing the `right' criterion to pick a model from one of the (several) distributions that are consistent with the available marginal information. First, we consider a central decision problem in operations management (OM): find an assortment of products that maximizes the revenues subject to a capacity constraint on the size of the assortment. Solving this problem requires two components: (a) predicting revenues for assortments and (b) searching over all subsets of a certain size for the optimal assortment. In order to predict revenues for an assortment, of all models consistent with the data, we use the choice model that results in the `worst-case' revenue. We derive theoretical guarantees for the predictions, and show that the accuracy of predictions is good for the cases when the choice data comes from several different parametric models. Finally, by applying our approach to real-world sales transaction data from a major US automaker, we demonstrate an improvement in accuracy of around 20% over state-of- the-art parametric approaches. Once we have revenue predictions, we consider the problem of finding the optimal assortment. It has been shown that this problem is provably hard for most of the important families of parametric of choice models, except the multinomial logit (MNL) model. In addition, most of the approximation schemes proposed in the literature are tailored to a specific parametric structure. We deviate from this and propose a general algorithm to find the optimal assortment assuming access to only a subroutine that gives revenue predictions; this means that the algorithm can be applied with any choice model. We prove that when the underlying choice model is the MNL model, our algorithm can find the optimal assortment efficiently. Next, we consider the problem of learning the underlying distribution from the given marginal information. For that, of all the models consistent with the data, we propose to select the sparsest or simplest model, where we measure sparsity as the support size of the distribution. Finding the sparsest distribution is hard in general, so we restrict our search to what we call the `signature family' to obtain an algorithm that is computationally efficient compared to the brute-force approach. We show that the price one pays for restricting the search to the signature family is minimal by establishing that for a large class of models, there exists a "sparse enough'' model in the signature family that fits the given marginal information well. We demonstrate the efficacy of learning sparse models on the well-known American Psychological Association (APA) dataset by showing that our sparse approximation manages to capture useful structural properties of the underlying model. Finally, our results suggest that signature condition can be considered an alternative to the recently popularized Restricted Null Space condition for efficient recovery of sparse models.

Abstract:
While product modularity is often advocated as a design strategy in the operations management literature, little is known about how customers respond to modular products. In this research we undertake several experiments to explore consumer response to modularly upgradeable products in settings featuring technological change. We consider both the initial product choice (between a modularly upgradeable product and an integral one) and the subsequent upgrade decision (replacement of a module vs. full product replacement). We uncover the following paradox: while modularity might seem to be most-advantageous for a short life-cycle product (because a modular design would avert having to fully replace the product after only a short time), such a product faces two strikes: first, consumers tend to excessively discount the cost savings associated with the modular upgrade, and second, we observe a preference reversal between the initial purchase and the point of upgrade (at the point of initial purchase, people foresee making a full product replacement in the future, yet, when faced with the actual upgrade decision, they are more likely to revert to a modular upgrade). On the other hand, consumers insufficiently discount cost savings when the time-to-upgrade is long. Finally, we discuss and test several pricing and product design strategies that the firm can use to respond to these cognitive biases.

Abstract:
We examine transshipment incentives in a decentralized supply chain where a monopolist distributes a product through independent retailers. A key insight is that the transshipment price determines whether the firms benefit from, or are hurt by, transshipment. In particular, we show that the manufacturer prefers to set the transshipment price as high as possible, while retailers prefer a lower transshipment price. Given the important role of the transshipment price in determining the benefits that each firm gets from transshipment, it is useful to consider transshipment in the case where retailers are under joint ownership (a “chain store”) and the transshipment price does not play a role. This comparison yields two surprising results. First, if decentralized retailers control the transshipment price, they will choose a relatively low transshipment price as a way to mitigate the manufacturer's ability to extract profits by increasing wholesale prices; therefore, the manufacturer may prefer dealing with the chain store which does not have a transshipment price rather than with decentralized retailers. Similarly, the decentralized retailers can use a low transshipment price to achieve higher total profits than a chain store.

Abstract:
We examine resource allocation issues in a setting with uncertain task resolution times. Using software maintenance requests as a context, we model task resolution times in corrective software maintenance process using multiple hazard distributions. we present a hazard model to capture the decrease in the marginal likelihood of successful resolution of a task as the number of effort-cycles expended on the task increases. We estimate the model parameters using real-life data from a systems software product. We demonstrate that the model fits real-life data very closely. Using the model, we numerically analyze implications for capacity planning and service execution in the context studied. Specifically, we demonstrate that imposing temporal cut-off policies in task resolution substantially reduces waiting times for incoming MRs in the system while minimally impacting the rate of successful MR resolution. Further, we show that these approaches can be used to improve productivity and resource utilization, and reduce the occurrence of “firefighting” behavior commonly encountered in this environment. We demarcate the diverse tradeoffs that managers need to evaluate during resource allocation in order to enhance the overall performance of the software maintenance operations. Finally, we discuss other managerial settings in which the modeling approach can be applied.

Abstract:
A major constraint in scaling up large scale treatment programs in resource-limited settings is the unavailability of appropriate diagnostic devices that can inform clinical decisions in a timely manner. Most existing devices, due to their technical complexity, are placed in a few central laboratories serving hundreds of remote health facilities. Such centralized diagnostic networks are characterized by long delays in providing results and consequent poor patient retention. Several point-of-care (POC) devices that aim to obviate the need for such complex diagnostic networks are under development. Occasional attempts at evaluating POC devices have focused on technical dimensions such as accuracy. However, this approach does not incorporate the key value proposition of POC devices{improved access through timely provision of test results. In this paper, we develop a mathematical model that explicitly incorporates the tradeoff between accuracy and access to evaluate the network-level effectiveness of POC devices. Using this framework, we argue that the key operational decision at the policy maker's disposal is placement of the devices: Which facilities should receive the device under resource constraint? We compare the optimal placement solution with rules of thumb that are followed in practice and/or are suggested in practitioner literature. Our analysis suggests that these heuristics can result in significant loss of effectiveness in general. However, their relative performance depends on device characteristics and network characteristics. We characterize conditions under which these rules of thumb are optimal. We apply our methodology to infant HIV diagnosis and calibrate our model using representative data from a sub-Saharan country.

Abstract:
Manufacturers typically sell consumer products through retailers and the presence of intermediaries has interesting ramifications for their product variety and pricing decisions. Retailers may want higher variety to help reduce price competition but the costs of variety are borne by the manufacturer. The increased variety may increase demand and profits for the manufacturer too but this depends on market-specific factors as well as costs. We explore these interactions through a model wherein a manufacturer sells multiple product variants at a wholesale price to two retailers who in turn compete for consumers. Consumers choose between the retailers based on the price and variety offered by each retailer. Several insights emerge from the analysis. We find that some retailer differentiation benefits the retailers (not the manufacturer) but too much differentiation hurts both the retailers and the manufacturer. If the market is fully covered, then the channel is coordinated even with a simple wholesale pricing contract. If the retailers incur costs to sell the product, the manufacturer loses out more than the retailers and in fact absorbs some or all of the retailer costs. Finally, asymmetry between retailers has some interesting consequences.

Abstract:
In this presentation, we deal with the issue of how the continued growth of peer-to-peer (P2P) used goods markets for durable products interacts with channel returns policies and affects pricing and product introduction strategies. We do so through two separate papers - while a P2P used goods market is there in both papers, they differ in terms of returns policies that are considered. In the first paper we focus on extra-channel returns policies, i.e., returns policies offered to the end customers by the retailer, while in the second one we concentrate on intra-channel returns policies, i.e., those offered by the manufacturer to the retailer. Our analysis in the first paper establishes that frequent product upgrades and rising retail prices in many durable product sectors are indeed due to the emergence of the P2P used goods market and how this market interacts with the extra-channel returns policy in altering the relative powers of the channel partners. We also provide empirical support for our theoretical result regarding product upgrades using data from the college textbook industry. In the second paper, we show that a stronger P2P used goods market generally increases the likelihood of an intra-channel returns policy to be the equilibrium strategy. This insight contradicts the burgeoning managerial trend to replace returns contracts with price-only ones for products having rapidly growing used goods markets. Furthermore, we show that when the customers are forward-looking, viability of a manufacturer returns policy is, in fact, negatively impacted by such behavior.

Abstract:
In this paper we examine the learning rates of individuals working under normal workload conditions vs. those working under time-pressure conditions. We find that those working under time pressure learn much less than those working under normal time. We also find evidence for significant forgetting by individuals. We suggest that this is one mechanism by which time compression diseconomies affect capability accumulation processes in firms. We also discuss implications for the literature on learning curves: specifically to our understanding of why learning curves may be heterogeneous across organizations and to our understanding of why knowledge may depreciate even in organizations that are continuously operating.

Abstract:
We analyze the choice of delivery system for a firm that sells make-to-order physical goods and/or services. The market consists of two segments that differ in their preference for product quality as well as their disutility from waiting time. The firm chooses between -- two dedicated delivery systems one for each market segment, and a flexible (pooled) delivery system that serves both segments. While pooled system allows firm to reduce operational cost of delivering products, dedicated systems allow firm to increase revenues by price discriminating customers more effectively. We characterize the optimal choice as a function of market scale, segment ratios, and performance deterioration that may result from mix variability. Our research highlights the effect of market cannibalization on the operations strategy decision of delivery system design.

Abstract:
Consider the estimation problem for the multiple linear regression (MLR) setup, under balanced loss function (BLF), where both goodness of fit and precision of estimation are modeled using either squared error loss (SEL) or linear exponential (LINEX) loss functions. We derive the minimum risk estimates for two different variants of BLF and prove for both the cases the existence of the ubiquitous SEL and LINEX estimates at the boundary conditions. Conclusions draw from the exhaustive simulation runs prove the general nature of our proposed theorems.

Abstract:
Revenue management practices often include overbooking capacity to account for customers who make reservations but do not show up. In this paper, we consider the network revenue management problem with no-shows and overbooking, where the show-up probabilities are specific to each product. We propose a randomized linear program to jointly make the capacity control and overbooking decisions with product-specific no-shows. We establish that our formulation gives an upper bound on the optimal expected total profit and that this upper bound is tighter than a deterministic linear programming bound that appears in the existing literature. We describe how the randomized linear program can be used to obtain a bid price control policy. Numerical experiments indicate that our approach is fast, able to scale to industrial-size problems, and can provide significant improvements over standard benchmark methods. This is joint work with Kalyan Talluri(UPF) and Huseyin Topaloglu(Cornell).

Abstract:
We consider the problem of determining the optimal assortment of products to offer in a given product category when customers have heterogeneous tastes. Each customer is characterized by a type, which is a list of products he or she is willing to buy in decreasing order of preference. We assume that customers arrive sequentially over one period and choose from the set of products remaining in inventory at the time of their visit. This is called consumer-driven, dynamic, stock-out based substitution. It is conventional to assume that the type of a customer is known only upon arrival and is independent of the type of other customers. This is called the random proportions demand model. This problem has been shown to be very hard to solve and no efficient method to obtain the optimal solution is known to our knowledge. However, if the number of customers of each type is a fixed proportion of demand there exists an efficient algorithm for solving for the optimal assortment. In this paper, we show that the fixed proportions model gives an upper bound to the optimal expected profit for the random proportions model. This bound may be used to compare the performance of the many heuristics that have been suggested by various authors to solve the assortment planning problem with consumer driven, dynamic, stock-out based substitution and random proportions of customers of each type. We also provide a bound for the componentwise absolute difference in expected sales between the fixed proportions and the random proportions models, which is asymptotically tight as the inventory vector is made large, while keeping the number of products fixed. This result provides us with a lower bound to the optimal expected profit. When the optimal assortment under fixed proportions is large, the lower bound and upper bound are very close to each other.

Abstract:
This paper examines how the unit sales (or demand) of a manufacturer are affected by the variance in its prices (due to promotions or various incentive schemes) and by the fixed ordering cost of its industrial customers in the maintenance, repair and operations (MRO) industrial context. Unlike existing models in the literature, we consider the case when both the order size and the order interval are “endogenously” determined by rational customers. In our model, the customer consumption rate of these MRO products – light bulbs, cleaning solution, lubricant, cutting tools – is constant and it is independent of the purchasing price. We develop closed-form expressions for the variance of unit sales, and we examine the impact on unit sales when the firm offers price promotions for multiple products and when there are multiple customer segments with different consumption rates.

Abstract:
The appointment scheduling problem arises in health care services, where two conflicting costs need to be minimized: the cost of under-utilization of high-cost installations (such as MRI scanners and operation rooms), and the cost of inconvenience to patients and the medical staff if a particular medical procedure starts late. Traditionally, the problem has been studied as a stochastic optimization problem. We present a novel robust optimization model for the problem. For each job, we are given its minimum and maximum possible execution times. The objective is to find an appointment schedule for which the cost in the worst case scenario is minimized. We present a simple heuristic, called the global balancing heuristic, which gives an optimal schedule when the underage costs of the jobs are non-decreasing. We also give a closed form optimal solution for the problem for this case. The advantages of our model over the traditional stochastic optimization models is that it provides more insight into the structure of the optimal solution, and it can be used even when historical data about the processing times of the jobs is not available.

Abstract:
We consider the dynamic risk management problem for a commodity processor facing uncertain commodity prices. In each period over a multi-period horizon, the firm procures an input commodity and processes it to produce an output commodity. The processed commodity is sold using forward contracts while the input itself can be traded at the end of the horizon. The firm can also trade financial derivative instruments to manage the commodity price risk. We propose a dynamic risk measure DCVaR, based on the conditional value at risk (CVaR), to model the firm's risk aversion in a time-consistent manner over the planning horizon and obtain the optimal procurement, processing and financial trading policies. We show that the optimal procurement and processing policies are characterized by price dependent inventory thresholds and conditional on optimal financial hedging decisions, the operational policies can be calculated without knowing the details of the financial hedging itself. However, these optimal thresholds are hard to compute and we develop efficient heuristics to obtain the operational and financial decisions in each period. Using numerical studies, we show that a) these heuristics are near optimal and b) optimizing a time-consistent risk measure provides a better mean-risk tradeoff for the total profits and reduces the probability of extreme losses in intermediate periods, compared to optimizing the CVaR of total profits. Please click here for more details.

Abstract:
Demand often exceeds capacity at the congested airports. Airline frequency competition is partially responsible for the growing demand for airport resources. We propose a game-theoretic model for airline frequency competition under slot constraints. The model is solved to obtain a Nash equilibrium using a successive optimization approach, wherein individual optimizations are performed using a dynamic programming-based technique. The model predictions are validated against actual frequency data, with the results indicating a close fit to reality. We use the model to evaluate different strategic slot allocation schemes from the perspectives of the airlines and the passengers. The most significant result of this research shows that a small reduction in the total number of allocated slots translates into a substantial reduction in flight and passenger delays, and also a considerable improvement in airlines’ profits.

Abstract:
It is common for procurement managers to frequently organise auctions among qualified suppliers to stay abreast of current supply-market pricing. The process of qualifying suppliers is expensive to the buyer; as a result, a buyer often maintains a pool of qualified suppliers, the supply base. In traditional auction models, the buyer uses the auction to ¯nd and award the business to the lowest-cost supplier in the supply base. In practice, however, sole awards can alienate losing suppliers and cause them to defect from the supply base. Therefore, to maintain the supply base | and thereby control supplier qualification costs | buyers often employ split awards. Hence, there is a trade-o® between the e®ective purchasing cost on the one hand, and the qualification cost paid to maintain the supply base on the other. We model and investigate this trade-o® and characterize (1) the optimal split award that minimizes long-run costs (purchasing and qualification) and (2) the optimal supply base size that the buyer should maintain. We also determine that higher per-supplier qualification cost leads to a smaller supply base but does not necessarily increase the extent of multi-sourcing. Finally, we show that when variability in supplier cost increases, the buyer maintains a larger supply base but does not necessarily decrease the extent of multi-sourcing.

Abstract:
We study the economics of sharing demand information between a dual sourcing firm and its retailer. Our analysis demonstrates that employing Advance Purchase Discount (APD) scheme in the supply chain synchronizes the timeline of the dual sourcing firm’s decisions with that of its retailer. This enables accurate, timely, and self-enforcing information sharing, which reduces the demand–supply mismatches, and improves the profitability of each of the agents in the supply chain. We provide prescriptions on the appropriate design of the contract that enables this Pareto-improving information sharing. Next,we extend this analysis to incorporate realistic constraints on the dual sourcing firm’s limited knowledge of its retailer’s administrative cost and information quality. We characterize “certainty-equivalent” values of the unknown retailer parameters, which facilitate analogous prescriptions for the design of APD contracts in these realistic settings. It is interesting that the unobservability of retailer parameters leads to an asymmetric and “degree-of-unobservability dependent” departure from the full-observability design of the APD contract. If the uncertainty in the unobserved parameter is small, then the optimal discount is higher compared to the case of full observability; conversely, when the uncertainty is large, the optimal discount is lower. It is significant that, despite the departure in the design of the APD contract, this analysis reiterates that the benefit of the APD persists even under this practical constraint.

Abstract:
The Simulation Optimization Problem (SOP) is an optimization problem where the objective and constraint function(s) are observable only using a Monte Carlo simulation. Likewise the Stochastic Root Finding Problem (SRFP) is the Monte Carlo analogue of the problem of solving a nonlinear system of equations. Owing to their flexibility, both of these problems have recently found enormous expediency in a wide variety of application settings where the functions involved cannot be specified in closed-form, but are instead expressed conveniently and implicitly using a simulation. Sample-path methods, i.e., methods that generate an approximate deterministic problem using a “large enough” sample size and solve it to “adequate” tolerance, are currently amongst the attractive methods for solving SRFPs and SOPs. In this talk, we first answer the question of how to choose sample sizes and error tolerances within sample-path methods for finding some solution to an SRFP (or a local minimum in SOPs). We will characterize a class of error-tolerance and sample- size sequences that are superior to others in a certain precisely defined sense. Second, and time permitting, we will visit the question of finding all solutions to an SRFP (or a global extremum in SOPs). Specifically, we will demonstrate that a simple relationship between the number of random restarts and sample size should be in effect for maximum efficiency. We will provide motivating applications and a numerical example to illustrate key results.

Abstract:
Cooperative (co-op) advertising is an important instrument for aligning manufacturer and retailers decisions in supply chains. Co-op advertising programs are common in practice, and as much as 27 billion dollars were spent in such programs in 2007. We analyze a case of retail market duopoly where one or both of competing retailers are supported by the manufacturer in their advertising costs. We model the problem as a Stackelberg differential game in which the manufacturer announces his shares of advertising costs or his participation rates, and the retailers in response play a Nash differential game in choosing their optimal advertising efforts over time. We obtain the feedback equilibrium solution providing the optimal advertising policies of the retailers and manufacturer's participation rates. We identify the key drivers that determine the optimal participation rates and, in particular, obtain the conditions when the manufacturer supports one or both of the retailers. Finally, we analyze the extent to which cooperative advertising coordinates the channel.

Abstract:
This paper documents that excess inventory announcements, an indication of demand-supply mismatch, are associated with an economically and statistically significant negative corporate performance. Over a two-day period (the day of the announcement and the day before the announcement) the mean the stock market reaction ranges from -6.79% to -6.93% depending on the benchmark used to estimate the market reaction. When excess inventory is at the announcing firm’s customers, the market reaction is more negative than when the excess inventory is at the announcing firm. The stock market reaction is less negative for excess inventory announcements made by larger firms but more negative for firms with higher growth prospects and with higher debt-equity ratios. Excess inventory situations leads to higher stock price volatility and lower operating profitability.

Abstract:
This paper studies a model of community-based health care delivery for chronic diseases in a non-profit setting. In this setting, patients periodically access the health care delivery system, which influences their disease progression and consequently their health outcomes. The provider's goal is to maximize community-wide health outcomes, subject to capacity constraint. We investigate how the provider can improve her objective purely by making better operational decisions pertaining to capacity allocation across different patient classes. We model the provider's problem as a finite horizon stochastic dynamic program, where the provider decides which class of patients to schedule at the beginning of each period. Therapy is provided to scheduled patients, which improves their health states temporarily. Patients that are not seen follow their natural disease progression. First, we derive an analytical characterization of the optimal policy for a stylized version of this model. Second, we use this characterization to design a heuristic for more general settings. Third, we calibrate our operational and disease progression models using data from Mobile C.A.R.E. Foundation, a community-based provider of pediatric asthma care in Chicago. For small problem instances, we find that our heuristic performs very close to the optimal policy. For larger problem instances, we find that our heuristic can improve the health gains of the community by up to 15% over the current policy. We also find that the potential benefit is higher when capacity is tighter. The benefit is driven by a more effective capacity allocation: our heuristic allocates less capacity to patients with better health and more capacity to those with worse health states.

Abstract:
In many services, referred to as “discretionary” services, the value or quality derived by a customer from a service depends upon the time the service provider (SP) devotes to the customer and the valuation differs across customers. Customers procure the service based on the expected value to be delivered, the timeliness of service and the fee charged by the SP. In this work, we explore the impact of two widely used fee structures on service performance including SP's profitability, demand, utilization, congestion level, etc. One is a fixed fee wherein the SP commits to a value to be delivered and charges a fixed price. The other is a time-based payment scheme, whereby the SP charges the customer a rate per unit of time and the value (or quality) delivered depends upon the customers' choice of service time. Our results show that the fixed fee scheme yields higher revenue when customers' valuation of service time is homogenous. But when they are heterogeneous, the skewness in the valuation distribution plays a critical role in determining which scheme provides higher revenues. Moreover, the scheme that maximizes revenue may also provide better performance along several other performance measures. Interestingly, a higher (lower) utilization is always accompanied by a lower (higher) wait time. The time-based scheme seems to be better along several service performance dimensions in many scenarios. It always generates higher revenue per customer while at the same time providing higher average value to the customer. Further, even though the average customer pays more, demand may actually be higher under some mild conditions. This is joint work with Chunyang Tong (Shanghai University of Finance and Economics)

Abstract:
Firms in innovation-intensive industries such as pharmaceuticals have to deal with patent expiration and consequent loss of monopoly position for key blockbuster products. One strategy to beat "patent death" is to introduce a vertical line extension / upgrade with an add-on patent, a strategy referred to as "Evergreening". However, the process of evergreening is fraught with uncertainty since this add-on patent is subject to tight guidelines and may also fail other approval processes (such as FDA approval in the case of pharmaceuticals). Thus, an incumbent firm has to make an upfront production capacity commitment without clarity on whether the upgrade will reach the market. This uncertainty also affects the capacity commitment of a generic entrant who introduces a clone of the incumbent's existing product ex-post patent expiration but whose market demand depends on the success or failure of the incumbent's upgrade. We analyze a two stage competitive game between the incumbent and the generic manufacturer. In the first stage, before new product uncertainty is resolved, the incumbent commits to new product capacity while the generic manufacturer commits to generic product capacity. In the second stage, new product uncertainty is resolved and both firms make pricing decisions. We find that the minimum upgrade success probability required for an incumbent to commit to evergreening is higher when generic competition is anticipated as opposed to a monopolistic setting. However, upgrade success probability is typically a decreasing function of the level of product improvement in the upgrade. In the context of this risk-return trade-off faced by the incumbent, we characterize the feasible levels of product improvement for the incumbent and the equilibrium capacity commitments of both the incumbent and the entrant. We find that the nature of these results depends on the shape of the risk-return profile faced by the incumbent. This implies that the strategy of both the incumbent and the entrant could be different based on incumbent firm type and the product category under consideration This is joint work with Ram Bala and Sumit Kunnumkal (ISB)

Abstract:
We study a producer’s dynamic pricing policy when marketing a durable good, particularly an item that provides consumer utility via two mechanisms; specifically, consumers experience an intrinsic consumption utility and an externality (denoted ‘fashion utility’) that depends upon the conspicuousness of the product and the 'type' of people who consume it. Past literature has emphasized the signaling feature of visible goods, ignoring intrinsic product quality; consequently, these studies indicate that an individual’s preference for a product is enhanced by merely charging a higher price, thereby making the product more exclusive. In our analytical model, we consider the joint impact of consumption utility and fashion utility, thereby reversing the direction of causality long emphasized in prior studies. We show that products with high intrinsic quality command higher prices due to greater input costs; with a higher retail price, such products become exclusive and, hence, more fashionable when consumption is visible. Our model derives a series of empirically testable propositions that arise from our dynamic model. This is joint work with Richard Schaefer (University of Texas at Austin)

Abstract:
Limited availability of resources makes operations in start-ups quite distinct from well-established firms. Financial and operational decisions in these firms are a challenge due to the limited availability of cash and the risk of bankruptcy if expenditures made do not result in expected cash flows. Traditional models have focused on either maximizing revenues or minimizing the risk of bankruptcy. However, startups may be interested in making judicious tradeoffs between these two extremes. Unfortunately this is not easy to do given existing approaches. In this talk we present a heuristic to identify the efficient risk reward frontier for operational decisions. Efficient frontiers are popular in finance literature, for example for portfolio optimization, but have not been as widely used in operations because identifying it in stochastic operational models is a challenge. We use the startup application as an illustration of our heuristic methodology for creating efficient frontiers using a stochastic dynamic programming model. From an implementation standpoint this implies carrying information for both risk and reward for each state in the problem, which typical dynamic programs are not designed to do, since their functional equations optimize for only one value, either minimizing risk or maximizing reward. Our methodology overcomes this impediment, as we will discuss in the talk.

Abstract:
We visit the following basic problem: For a ‘generic’ model of consumer choice (namely, distributions over preference lists) and a limited amount of data on how consumers actually make decisions (such as marginal preference information), how may one predict revenues from offering a particular assortment of choices? We present a new non-parametric framework to answer such questions and design a number of tractable algorithms -- from a data and computational standpoint -- for the same.

Abstract:
Revenue management models in the literature, and in many implementations, make some important assumptions such as Poisson arrivals, independence, and multinomial logit customer purchase behavior. If one analyzes sales data, it quickly becomes clear that many of the assumptions are clearly violated. The eventual revenue impact varies; simulation experiments seem to indicate for instance that frequently reoptimizing the controls makes the simplified model very robust to many of the violations. The performance of the models however is very sensitive to the estimation of the price-sensitivity parameter. This task is especially challenging in RM applications since cannot observe customer no-purchases, and the number of samples is typically very small. We examine the standard finite-period, one-arrival-per-period dynamic program and its incomplete data likelihood function. We discuss a number of problems with the likelihood function. We augment the study with simulation experiments comparing the ML estimates vs. true parameters for small samples. We also propose a new risk-ratio procedure that under some assumptions leads to an exact unbiased estimator. We show conditions under which this estimator can be calculated by solving a convex or quasi-convex program. We describe simulation experiments where the method in many cases recovers the true parameters to the second decimal place, without observing no-purchases.

Abstract:
There is a natural order to most events in life: Everything from learning to read to DNA sequences in molecular biology follows some predetermined, structured methodology which has been refined to yield improved results. Likewise, it would seem that firms could benefit by adopting and implementing technologies in some logical way so as to increase their overall performance. In this study of 555 hospitals, we investigate the order in which medical technologies are transformed into information technologies through a process of converting them from stand-alone technologies to interoperable, integrated information systems and whether certain configurations of sequences of integration yield additional value. We find that sequence does matter and that hospitals which integrated foundational technologies first – which in this case are known to be more complex – tend to perform better. Theoretical and practical implications of this finding and others are discussed.

Abstract:
Distributed product development, in which a firm's new product development operations are carried out in more than one geographic location, is becoming increasingly prevalent in a number of industries. While the monetary rationale for distributed development has received significant attention, some of the deeper implications of such development are just beginning to be understood. Motivated by our field work at software and electronic firms, we ask the following question: How does the decision to distribute product development work across geographic locations impact the firm's product offering and market coverage decisions? We propose a stylised model to understand the linkages between the various drivers of distributed development such as capacity constraints and cost differences and its market implications such as customer response to remotely developed products. Analysis of the model helps unearth a subtle interaction between a firm's distributed development practice and its product line design decision. In particular, under some conditions, distributed development whether pursued for cost reduction or capacity enhancement can make it more profitable for the firm to limit market coverage and offer an exclusive high-end offering. Further, distributed development also conditionally makes it more profitable for the firm to offer a product line that expands market coverage. Unexpectedly, product line optimality occurs only at intermediate values of cost advantage, capacity limits and customer aversion. These interactions provide a supply-side explanation for the versioning of development-intensive goods. Further, they underscore the need for managers to integrate their product line design decision with their distributed development decision.

Abstract:
Services processes are often broken into multiple parts and when outsourcing services, a fundamental question is which part(s) of the service process to outsource.  We consider a two-level service process, where the first level serves as a gatekeeper for experts in the second level.  When a customer request enters this system, it is first assessed by a gatekeeper.  Depending on the request's complexity, the gatekeeper can refer it to an expert (direct referral) or attempt to perform the service. If the gatekeeper successfully performs the service, then the customer request leaves the system.  Otherwise, the request is referred to an expert, who always completes the service.  This two-level process is common in practice.  The customer request could be a call to a technical support call center or a visit to a medical clinic; in both these cases the customer is present.  The request could also be a credit or loan application, in which case the customer is not physically present at the time of service. A firm may choose to outsource all or parts of this two-level process to a vendor.  In this talk we examine the optimal approach to outsourcing and describe the parameters of optimal contracts between the firm and the vendor.  We also investigate how vendor labor cost advantages and other parameters influence the firm’s choices.  We draw upon results from a series of related papers involving joint work with Pinker, Hasija and Lee that provide building blocks for the analysis of this problem.

Abstract:
In an experimental newsvendor setting where 310 subjects make 50 repeated newsvendor decisions with the same known ex-ante parameters, we investigate three biases: Level bias – the average tendency of ordering away from the normative order quantity; adjustment bias – the tendency to adjust period-to-period order quantities; and observation bias – the tendency to let the degree of demand information observed influence order quantities. We study these biases in terms of decisions (quantities) and performance (expected mismatch cost) and find evidence to support the presence of all three as well as significant interaction between them. We find that the portion of mismatch cost due to adjustment bias exceeds the portion of mismatch cost due to level bias in three out of four conditions; highlighting the importance of considering adjustment bias in addition to the more commonly studied level bias. Observation bias is studied through censored demands, a situation which arguably represents the majority of newsvendor settings. When demands are uncensored, subjects tend to order below the normative quantity when facing high margin and above the normative quantity when facing low margin, but in neither case beyond mean demand (a.k.a. the pull-to-center effect). Censoring in general leads to lower quantities, magnifying the downward adjustment when facing high margin but partially counterbalancing the upwards adjustment when facing low margin, violating the pull-to-center effect in both cases.

Abstract:
This paper extends the theory of supply chain incentive contracts from the static newsvendor framework of the existing literature to the simplest dynamic setting. A manufacturer distributes a product through retailers who compete on both price and fill-rates. We show that inventory durability is the key factor in determining the underlying nature of incentive distortions and their contractual resolutions. When the product is highly perishable, retailers are biased towards excessive price competition and inadequate inventories. Vertical price floors or inventory buybacks (subsidies for unsold inventory) can coordinate incentives in both pricing and inventory decisions. When the product is less perishable, the distortion is reversed and vertical price ceilings or inventory penalties can coordinate incentives.

Abstract:
This talk is based on a research paper, `Incorporating price and inventory endogeneity in sales forecasting` by Saravanan Kesavan, Vishal Gaur and Ananth Raman, as well as experience gained in co-developing an MBA elective course with the same title. Investors assess future value of firms based on the historical information provided by them. However, historical time series information is the aggregated outcome of a complex operational process. Therefore, can one use operations management theory to develop estimation models that yield better forecast of future performance than equity analysts? In this paper, we develop this idea using the context of inventory management in public-listed US retailers. We construct a model which forecasts future sales by constructing joint forecasts of annual cost of goods sold, inventory, and gross margin for retailers using historical data. More importantly, the residuals from our model in one year are predictors of bias in analysts’ sales forecasts for the subsequent year. Thus, our model predicts when analysts will forecast poorly, and provides more accurate forecasts in such situations using inventory and gross margin data. In numerical tests, sales forecasts from our model are more accurate than forecasts made at the same time by equity analysts as well as standard time-series models. Thus, we show that analysts do not fully utilize the information contained in historical cost of goods sold, inventory, and gross margin of firms.