**Mass Customization: ****Overview**

As my primary line of research during my Ph.D. studies, I have explored new challenges in *Mass Customization* where manufacturers offer a significantly large number of *end-products*. Moving from *limited variety* to mass customization has made it impractical to perform any analysis at the level of end-products. As a result, traditional approaches in demand forecasting, parts-capacity planning, production planning, pricing decisions, and so forth are no longer applicable in the context of mass customization. So far I have studied the problems of demand forecasting and parts-capacity planning that has resulted in the following papers (with Reza Ahmadi and Sriram Dasu):

[1] Mass Customization and “Forecasting Options’ Penetration Rates Problem.” ** Operations Research. **Accepted.

[2] The Weighted Non-Negative Least-Squares Problem with Implicitly Characterized Points.

**. Accepted.**

*Operations Research*[3] Mass Customization and the “Parts Capacity-Planning Problem.”

**. 1st major revision was submitted.**

*Operations Research***(Job Market Paper)**

**Our Research**

There is a trend that firms are moving towards *Mass Customization* by allowing customers to configure products by selecting among options (features), e.g. automotive industry, consumer electronics, computers, furniture, and aircraft. My research is motivated by the problems faced by a global auto manufacturer (GAM) that offers 100-500 options for a car. A *configuration (end-product)* is comprised of a collection of options that satisfy *engineering (technological) constrains*, e.g., some options are mutually incompatible while, in other instances, selection of an option may require selection of another option. There are usually 10^25-10^40 different configurations for a car model.

Since it is impossible to forecast the demand at the level of configurations, firms forecast options’ demand. In [1], we note that when firms forecast options’ demand, they fail to consider all *engineering constraints* that define relationships between options. As a result, the forecasts are frequently *inconsistent* with the engineering constraints. This has resulted in significant excess inventories, shortages, and customer dissatisfaction in GAM. In [1], we study the problem of finding the best consistent forecast. This is a new and important problem. We formulate this problem as finding a point in the convex cone of the configurations that has the minimum Euclidean distance to a given forecast. We present an approach that sequentially constructs the convex cone and stops when it finds the best consistent forecast. We analyze the theoretical properties of our approach and provide insights on its convergence rate. We also show the effectiveness of our approach on a set of real instances that we have received from GAM.

In [2], we present a generalization of our model and methodology in [1]. We introduce a new variation of the *Non-Negative Least-Squares* (NNLS) problem that is defined as finding the Euclidean distance to a convex cone generated by a set of discrete points. Existing works in the literature assume that the set of discrete points are *explicitly known*. In the new variation, the discrete points are *implicitly known* and there are an *exponentially large* number of them—e.g., the feasible solutions of an integer program. This problem can have applications in manufacturing, machine learning, clustering, pattern recognition, and high-dimensional statistics. In [2], we design an effective solution approach for this new problem, present a lower bound, and establish the convergence rate of the lower and upper bounds.

Firms use demand forecast to determine parts’ capacities—i.e., the number of units of each part that is required over the planning horizon to satisfy demand. GAM signs contracts with suppliers based on obtained parts’ capacities. The total parts’ cost in GAM is around $90 billion per year and hence errors in parts’ capacities can result in significant wastage. The challenge is that parts’ requirement cannot be directly determined based on options’ forecast since a large number of parts’ requirements (up to 60%) is based on the combinations of options selected. The problem of determining parts’ requirement in the context of mass customization is new. We develop an effective approach for solving large industrial instances of this problem and compare our approach to that of the current practice.

I am currently studying the problem of *end-of-horizon portfolio management*. Once GAM determines parts’ capacities, they sign contracts with their suppliers according to which they guarantee to utilize pre-specified capacities of parts over the planning horizon. If GAM violates the reserved capacities (use less or more than what is specified in the contract), they have to pay penalties. Due to the uncertainty of demand, some of the parts’ capacities are depleted at the end of the planning horizon while there are unused capacities for some other parts. Hence, at the end of the planning horizon, it is important to determine which configurations to produce and in what quantities to effectively (in the most profitable manner) utilize the unused capacities of parts. We aim to develop a model and methodology for addressing this problem.