Recently, Associate Professor Wang Xiaolin from Business School of Sichuan University, as the first author, published his paper "Warranty Reserve Management: Demand Learning and Funds Pooling" in the Journal of Manufacturing&Service Operations Management. This work is done by Dr. Xiao-Lin Wang from Business School, Sichuan University, jointly with Dr. Yuanguang Zhong and Dr. Wei Xie from South China University of Technology, Dr. Lishuai Li from Delft University of Technology (now with City University of Hong Kong), and Dr. Zhi-Sheng Ye from National University of Singapore.

As one of the most important after-sales services, product warranty, which provides protection against premature failures for customers, has been adopted by most manufacturers to signal the quality and reliability of their products. In general, a warranty contract defines a specific coverage period, in which free rectifications of or refund compensations for failed units will be offered by the manufacturer. Though warranty services can address customers' concerns, managing a warranty program is costly from the manufacturer's perspective, which can account for as much as 15% of net sales. Essentially, warranty expenses are random in nature, because they are closely related to two coupled stochastic processes—product sales and failure processes. For the manufacturer, servicing future warranty claims can incur liquidity risk, such as a shortage of cash, when unanticipated claims occur. A common solution to these issues adopted by manufacturers is to create an independent warranty reserve fund to cover contingent liabilities that arise from warranty obligations. However, it is nontrivial to determine a proper amount of reserves to hold, because reserving either excess or insufficient cash would incur losses. How can warranty reserve levels be optimized and promptly adjusted is a focal issue, which is the main focus of this research.

In practice, an ever-increasing warranty reserve level is a major challenge for many manufacturers, especially the big ones. For example, Apple had a $3.834 billion balance in its warranty reserve fund at the end of September 2017 and the figures for Ford and General Motors were $5.031 billion and $8.479 billion, respectively. Many manufacturers, especially automakers, tend to keep much more money in their reserve funds. According to a report byWarranty Week, U.S.-based manufacturers put aside 1.8% of product sales (on average) for future warranty claims, which is 17.2 times the amount they actually pay out in warranty claims per month. One of the main reasons is that many manufacturers are unable to accurately predict future warranty expenses, causing an over stock in warranty reserves. On the other hand, it presents a bad image to customers if a manufacturer's actual warranty expense is beyond its expectation, which reduces its earnings and upsets its shareholders. Therefore, warranty reserve forecasting and management is an urgent yet challenging problem for manufacturers.

This research is motivated by the warranty reserving issues faced by a world-leading electronics manufacturer. The firm would like to determine optimal reserve levels throughout the warranty life cycles of its products so as to reduce the corresponding reserve losses. In this work,they first develop a reserve demand forecasting model to explicitly capture the underlying mechanism of warranty claim generation. In particular, to synthesize the uncertainties induced by product sales and failure processes,they derive two important statistics, the mean and (approximate) variance of reserve demand within an arbitrary time interval, which are two key metrics for warranty reserve planning. This enables us to dynamically forecast the reserve demand and then plan the associated reserves within an arbitrary time interval (e.g., a fiscal quarter) so as to facilitate short-term and faster liquidity. Then, given the mean and variance of reserve demands,they derive optimal reserve levels periodically for a product throughout its warranty life cycle via a distributionally robust approach. In addition,they compare the total reserve losses generated by the worst-case scenario with those by a benchmark—the simulated “true-distribution” scenario. The result shows that the worst-case reserving policy performs fairly well.

The authors further investigate two efficient approaches—demand learning and funds pooling—to explore potential reserve-loss-reduction opportunities. The demand learning model dynamically updates the reserving plan with field warranty data through an exponential-smoothing-type mechanism.They prove that, when the sales period is long enough, the optimal learning parameter asymptotically converges to a constant in probability in a fairly fast rate. This not only guarantees a stable output of the algorithm, but also simplifies the algorithm implementation. In addition, simulation experiments show that by incorporating the objective function in the learning process, the algorithm can significantly reduce total reserve losses under general sales and failure patterns, for which the proposed reserve demand forecasting model fails to capture the dynamics of warranty costs.They believe that this simple yet effective “predict-then-optimize” approach would be also useful for other operations management applications.

Moreover,they analyze the benefits of funds pooling for a manufacturer who manages warranties for multiple products (e.g., Apple handles warranty claims for Mac, iPhone, and iPad).They find that the pooling benefits change over different stages of the warranty life cycle, due to the dynamics of warranty expenses. More importantly, because reserve loss reduction is the primary objective of funds pooling,they prove an important result that the relative pooling benefit in terms of reserve losses exhibits a nonincreasing pattern (note that this result analytically holds only when there are two products). This implies that it is helpful to pool warranty reserves for multiple products as early as possible. Another compelling insight is that the relative pooling benefit in terms of reserve losses decreases as the relative range of demand standard deviations enlarges, meaning that it is more attractive to pool warranty reserves for products whose AWC uncertainties are of similar magnitudes. This insight is quite general in the sense that it is applicable to traditional risk pooling problems in operations management.

Finally, their case study using real data from a world-leading electronics manufacturer shows that combining demand learning and funds pooling can lead to a significant reduction in total reserve losses

in terms of percentage (27.64% - 62.76%), demonstrating the effectiveness of the proposed reserve planning methodologies. Overall, this research offers guidelines on how manufacturers should adaptively forecast and dynamically plan warranty reserves over the warranty life cycle.

Wang, X. L., Zhong, Y., Li, L., Xie, W., & Ye, Z. S. (2022). Warranty Reserve Management: Demand Learning and Funds Pooling. Manufacturing & Service Operations Management, 24(4), 2221–2239. http://dx.doi.org/10.1287/msom.2022.1086.