The management of ameliorating food inventories with age-differentiated products entails a trade-off between immediate revenues and further maturation. We derive interpretable decision rules for purchasing, fulfilment, and issuance decisions under purchase price and decay uncertainty. We learn the rules from the optimal policy for an aggregated problem. A linear program facilitates scaling back. For a port wine industry case, our derived management strategies yield a substantial profit increase.

We observe that detecting inventory inaccuracies can be viewed as a problem of identifying anomalies in a (low-rank) Poisson matrix. We propose a conceptually simple approach whose cost approaches that of an optimal algorithm at a min-max optimal rate. Using data from a consumer goods retailer, we show that our approach provides up to a 10× cost reduction over incumbent approaches to anomaly detection.

A firm’s cash flow from operations is a function of the contemporaneous and lagged values of its operational variables---sales, operating cost, inventory, payables, etc. Estimating this function is important for forecasting and managing cash flows. However, cash flow forecasting is a challenging problem. In this paper, we propose a generalizable and data-driven model of a firm's operations to disentangle this endogeneity and estimate causal impacts among variables.

We study the data-driven newsvendor problem in which the decision-maker must trade-off underage and overage costs and only observes historical demand. Our metric of interest is the worst-case relative expected regret, compared to an oracle knowing the demand distribution. We provide an exact analysis of Sample Average Approximation across all data sizes. We also derive a minimax optimal algorithm and its performance. Our work reveals that tens of samples are sufficient to perform efficiently.

This paper studies the classic price-based network revenue management (NRM) problem with demand learning. The retailer dynamically decides prices of n products over a finite selling season (of length T) subject to m resource constraints, with the purpose of maximizing the cumulative revenue. In this paper, we focus on nonparametric demand model with some mild technical assumptions which are satisfied by most of the commonly used demand functions.

Various operations management problems can be formulated as stochastic optimization under random truncation. Leveraging a convex reformulation, we propose a mirror gradient method that achieves global convergence for the nonconvex objective with optimal complexity. The proposed method only operates in the original space using estimators of the nonconvex objective and consistently outperforms several state-of-the-art control policies in passenger and air-cargo network revenue management.

We consider a joint customization and assortment optimization problem. A firm faces customers of different types, each making a choice according to a different MNL model. The firm picks an assortment of products to carry subject to a constraint. Then, a customer of a certain type arrives into the system and the firm customizes the assortment that it carries by, possibly, dropping products from the assortment. We study the value of customization, the complexity of the problem and design novel algorithms.

In this paper, we explore dynamic ad allocation with limited slots upon each customer arrival for e-commerce platforms when customers follow a choice model to click the ads. Motivated by the recent advocacy for the algorithmic fairness, we adjust the value from advertising by a general fairness metric evaluated with the click-throughs of different ads and customer types. We propose a two-stage stochastic program and design a debt-weighted offer-set algorithm to solve the online problem.

Motivated by the middle-mile delivery operations of an e-retailer, we consider the problem of designing a sparse graph that supports a large matching after random node deletion. We study three families of sparse graph designs (namely, Clusters, Rings, and Erdos Renyi graphs) and show that their performances are close to the complete graph. We test our theory using real data and conclude that adding a little flexibility to the routing network can significantly reduce transportation costs.

We provide the first improvements to the celebrated correlated rounding procedure of Jasin and Sinha (2015), which has become a fundamental problem in multi-item e-commerce order fulfillment.

We derive rounding schemes with guarantees of $1+\ln(n)$ and $d$, where $d$ is the maximum number of fulfillment centers containing an item.

The first of these improves their guarantee of ~n/4 by an entire order of magnitude in terms of the dependence on $n$.

We also show our guarantees to be tight.

Newly emerging smart replenishment systems at the point-of-consumption track product usage via smart, connected devices and use this data to automate order processes. Based on a large industry dataset from the professional coffee industry, we develop models for demand forecasting, inventory control, and replenishment under inventory inaccuracies.

Project networks involve several participants; clients, contractors, and subcontractors; each working on multiple projects concurrently. By tracking a network of 2.6 million public projects over a five-year span, we show that when a project suffers a localized disruption, other projects in the network get delayed because participants reallocate resources to the disrupted project. This creates a domino-effect externality that ripples through the network, causing delays across unrelated projects.

We consider the problem of a 3D printer supplying a critical part installed in multiple machines embedded with sensors and interconnected via IoT. We show that it is optimal to print-to-stock predictively in advance of demand, triggered by a system-lifetime-status dependent threshold. We further quantify the impact of IoT on system cost and inventory by separately assessing the impact of advance demand information from embedded sensors and that of IoT's real-time information fusion.

We study a dynamic pricing problem over multiple periods when consumers develop a habit or satiation from their past consumption. We derive an inter-temporal demand function to capture these two effects. We establish that the profit maximization problem under our demand function is jointly concave and then characterize the trends in the optimal prices over the multi-period horizon. Finally, we provide several extensions including bounds on prices and optimal profit and non-stationary state dependence.

Ride-hailing firms set prices dynamically to match supply and demand. But rapid price changes incentivize riders to wait for low prices. When prices drop, patient customers request en masse, causing a drop in supply and a price increase. We show how dynamic pricing inherently creates such oscillations in supply and prices, that these oscillations in supply levels are inherently inefficient, and that a service model that allows riders to wait in a formal queue overcomes this inefficiency.

We study the pricing problem of an assortment of multiple, substitutable products. We propose two new methods that can support retailers with maximizing their revenues. The first method is a customer choice model based on the Markov Chain Choice model in combination with reservation prices. The second method relies on a linear approximation for the finite inventory, finite time horizon multi-product dynamic pricing problem.

Dynamic resource allocation problems arise under a variety of settings and have been studied across disciplines such as Operations Research and Computer Science. This work introduces a unifying model for a very large class of dynamic optimization problems. We show that this class encompasses a variety of disparate and classical dynamic optimization problems and we characterize the performance of the fluid certainty equivalent control heuristic for this class of problems.

We consider revenue management problems with heterogeneous resources, each with unit capacity. An arriving customer makes a booking request for a particular interval of days. The goal is to find a policy that determines an assortment to offer each customer to maximize total expected revenue. We show that we can efficiently perform rollout on any static policy. We develop two static policies derived from value function approximations, and give performance guarantees for both policies.

The seller wants to sell an item to n buyers such that the buyers' valuation distribution is from a given class (i.i.d., mixtures of i.i.d., affiliated and exchangeable, exchangeable, and all distributions) and the seller minimizes worst-case regret. The first three classes admit the same minimax regret, decreasing in n, while the last two have the same minimax regret equal to that of the case n = 1. Across all settings, the optimal mechanisms are all second-price auctions with random reserve.

We study a new selling mechanism in online display advertising markets that combines both guarantee contracts (GC) and real-time bidding (RTB) and allow advertisers to strategically choose between these two channels. Despite the complexity due to the advertisers’ strategic behaviour and the auctions, we are able to obtain the closed-form solution and show that combining GC and RTB can generate more revenue for the publisher, compared to the case when only RTB or GC is used.

We consider a platform's problem of collecting data from privacy sensitive users to estimate an underlying parameter of interest. We formulate this question as a Bayesian-optimal mechanism design problem, in which an individual can share her (verifiable) data in exchange for a monetary reward or services, but at the same time has a (private) heterogeneous privacy cost which we quantify using differential privacy.

We study a discrete pricing setting where each customer is offered a contextual price. Often only historical sales records rather than customer valuations are available, where the data is influenced by the previous pricing policy. This introduces difficulties in estimating revenue. We approach this problem using ideas from learning with corrupted labels to formulate loss functions that directly optimize revenue, rather than going through an intermediate demand estimation stage.

Problem definition: This paper investigates the joint estimation of the consumer arrival rate and choice model parameters when ``no-purchasers” (customers who considered the product but did not purchase) are not observable. Estimating demand even with the simplest discrete-choice model such as the MNL becomes challenging as we do not know the fraction that have chosen the outside option (i.e., not purchased).

Mining for heterogeneous response to treatment is a crucial step in data-driven operations. We propose a partitioning algorithm to estimate heterogeneous odds ratio, a popular measure when response to treatment is binary. We integrate an adversarial imputation step to account for partially observed treatments (e.g., if full information is only available for purchased items). We validate our methodology on synthetic data and case studies from political science, medicine, and revenue management.

Despite the potential of a battery energy storage system (BESS) to electrical grids, most standalone use of BESS is not economical due to its high upfront cost and batteries' limited lifespan. Energy stacking, a strategy providing multiple services simultaneously, has been of great interest to improve profitability. However, some key questions remain unanswered. We show that there exists cost-saving synergy, which enables stacking to double the profit of the best standalone service.

Solar power has risen as a sustainable & inexpensive option, but its generation is variable during the day and non-existent at night. Thanks to recent technological advances, a combination of solar+storage holds the promise of cheaper, greener, and more reliable off-grid power. Our work sheds light on this question by developing a model of strategic capacity investment in renewable generation and storage to match demand with supply in off-grid use-cases, while relying on fossil fuel as backup.

We examine whether and how renewable energy adoption affects energy efficiency (EE) improvement. Using site-level data from an industrial conglomerate, we find that using renewables to meet 10% more of a site's energy demand led to an additional 2.0% improvement in EE. This effect is heterogeneous in sourcing strategy where outside purchases led to gains, but on-site generation had no effect. Analysis of the mechanism suggests greater managerial focus on EE due to the costs of outside purchases.

We study a general model of reusable resource allocation problems. Customers of different types arrive according to a stationary stochastic process. The firm's goal is to maximize multiple kinds of rewards generated by customers. We develop an online policy for deciding on actions to take without knowing the distribution of customer types and show that when the usage duration is small compared with the length of the planning horizon, our policy achieves a near optimal performance.

We study the multisecretary problem with a capacity to hire up to B out of T candidates with their values drawn i.i.d from a distribution F on [0,1]. We investigate the achievable regret performance where our benchmark is the offline optimal policy. We establish the insufficiency of the common certainty equivalent heuristic for distributions with many types and gaps. We devise a new algorithmic principle called "Conservatism wrt Gaps" and use this to derive near-optimal regret scaling.

Prophet inequalities are a useful tool for designing online allocation procedures and comparing their performance to the optimal offline allocation. In this paper we derive the best-known guarantee for $k$-unit prophet inequalities for all $k>1$. We also provide a tight resolution to the related Magician's problem. Finally, we improve the guarantee from 0.2 to 0.319 for online knapsack, and from 0.321 to 0.3557 for unit-density online knapsack.

We examine whether mere adoption of fulfillment services offered by

platforms distorts competition by using data from a leading online retailing marketplace to empirically evaluate the effect on upstream

supply echelons. We find that evidence for regulatory views as the surplus welfare is absorbed by the platform. Smaller merchants with lower margin, are forced to increase price to remain profitable with platform fulfillment, leading to a price disadvantage compared to the bigger suppliers.

Capacity-constrained asset providers (APs) often compete over prices when they trade their transport capacities with logistics service providers (LSPs) via spot markets. To circumvent demand uncertainty, an AP and an LSP can negotiate a contract to secure sales and capacity, respectively. We propose a two-stage game theoretical model to study the trade-off of balancing the contract and spot market by characterizing the contracting and pricing strategies under competition and demand uncertainty.

We study the impact of promotion display for online retail platforms where customers search. Utilizing a dataset set which contains detailed behavior information, we estimate a search and purchase model. Accurate estimation also enables us to evaluate different promotion display schemes and design policies that can improve the revenue. Through counterfactual analysis, we demonstrate that our policies can improve the revenue for some product categories by 2-4%.

We study the problem when a firm sets prices for products based on past transaction data. We do not impose a model on the distribution of the customers' valuations and only assumes purchase choices satisfy incentive-compatible constraints. The valuation of each past customer can be encoded as a polyhedral set, and our approach maximizes the worst-case revenue. We show optimal prices in this setting can be approximated by solving a compact mixed-integer linear program.

We study a bundle pricing policy, Component Pricing with Bundle Size Discount (CPBSD). It sells bundles at the sum of component prices minus a discount depending on the bundle size. It subsumes many mechanisms including Component Pricing and Bundle Size Pricing. We show that CPBSD attains the optimal profit asymptotically among all pricing policies under a weak condition. We formulate MILP for the optimal CPBSD. Comprehensive numerical experiments demonstrate the good performance of CPBSD.

We consider a firm that sells some primary and ancillary products (services) to heterogeneous customers. The challenge is to determine the prices for all the products and services simultaneously, to optimize profits to the firm. We consider random utility model for customers' choice problem, and show that the choice model can be reformulated into a perturbed utility model (PUM) over the convex hull of the feasible solutions. Furthermore, we demonstrate how we can obtain a good approximation.