The goal of human resources analytics is to provide an organization with insights for effectively managing employees so that business goals can be reached quickly and efficiently. The challenge of human resources analytics is to identify what data should be captured and how to use the data to model and predict capabilities so the organization gets an optimal return on investment (ROI) on its human capital.

In our HR analytics lab, we employ techniques to prioritize and target applicants who are most qualified for a specific position, forecast workforce requirements and determine how to best fill open position, identify the factors that lead to greater employee satisfaction and productivity, and discover the underlying reasons for employee attrition and identify high-value employees at risk of leaving. We help clients and others to get a better insight in the trends, and how they can use the trends to stay or become leaders in their business.

The availability of police workforce ‘at the right time at the right place’ is crucial for the feeling of safety by citizens and for arresting criminals. Smart planning of police workforce based on accurate forecasts of crime incidents can make the difference. In this project, we develop new methods for accurate forecasting of crime risks and planning methods for optimizing police workforce, in close collaboration with Politie Amsterdam/Amstelland.

In life-threatening emergency situations, the ability of ambulance service providers to arrive at the emergency scene within a few minutes to provide medical aid may make the difference between survival or death. To realize such extremely short response times at affordable cost, efficient and proactive planning of ambulance vehicles is crucial. In this project, we develop new and efficient planning methods for proactive planning of ambulance services. Our results show that strong reduction of the response times can be obtained by smart, proactive planning.

The fashion industry is largely confronted with products having seasonal demand. This brings forth a risk that the supply is not totally cleared at the end of the season. Hence, near the end of the season, the retailer is forced to markdown the overstocked supply. Since there is always a gap between the demand that is expected and the demand that is actually realized, it is necessary to find the optimal markdown path during the season. The central question is when and how much to markdown in order to optimize the expected total profit given the available supply.

In this project, we focus on dynamic pricing policies that determine when to apply a markdown (based on the observed inventory levels) and how much to markdown (based on price elasticity which is estimated from sales data).

The model for deriving optimal dynamic pricing policies builds upon models from econometrics, statistics, operations research, and revenue management. The sales data provides input to price elasticity models from econometrics. By using Cox regression from statistics, one can obtain a model for the lead time of products. The two previous models are combined into a revenue management model with demand unconstraining. This is then used in a Markov decision model from operations research to determine optimal policies. The model has proven its success in a real retail environment.

Surveys are an important tool to collect and analyze quantitative information about members in a population. When individuals that are included in the sample do not disclose full information, there is non-response in the survey. The non-response affects the sample inclusion probabilities, which in its turn can lead to biased estimators. This means that the obtained responses of the sample can no longer be representative of the larger population defying the purpose of the survey.

Classical research has focused on the reduction of non-response through better survey design, studying contact strategies and the modes of collection, and by providing response incentives. The development of techniques to deal with missing data also helps to deal with non-response. In spite of all the methods, non-response is ever increasing and warrants new techniques to deal with non-response in surveys.

In this project, we focus on optimal allocation of survey resources when non-response can be expected. The main objective is to efficiently assign different data collection strategies to population units given a certain budget or given certain constraints on the accuracy of estimators. As such the focus is not on reducing non-response but on enhancing response given the available means. The use of external information sources (e.g., demographic data) is helpful in the development of new survey strategies. These strategies are based on predictive models in which intermediary results of surveys are used for adaptation of the model.

A well-organized network of depots and hubs is the basis for services provided by express delivery companies. The efficiency and effectiveness in any distribution network is largely determined by the operation of the nodes in such a network. Therefore, depots' performance strongly influences services delivered by TNT Express to its customers.

The objective of this project was to evaluate and benchmark performance of the depots and hubs in a large-scale delivery network.

Data Envelopment Analysis (DEA) was found to be the most suitable approach for the evaluation of depots' performance. A DEA model was developed and used for the benchmarking. As a result, out of 44 depots, 31 depots were identified as efficient and 13 depots as inefficient. Furthermore, the statistical analyses indicated four factors that have a significant influence on the efficiency score. Moreover, the analyses showed that the efficiency level does not depend on the depots' size neither on the geographical region. The application of DEA to the distribution network of the express delivery company gave new insight into the evaluation of express services and filled an existing knowledge gap in the academic literature on this topic.

Our current society gives rise to a huge volume of data that is created at every second. A big part of this flow comes from social media platforms. These platforms provide a very volatile flow of information. However, leveraging the information which is encapsulated in this fast stream of messages proposes a serious challenge. Finding the right messages in this stream of information is like seeking a needle in a hay stack.

In our social media lab, we employ techniques from various disciplines to find the right information in the bulk of data. Linguistics, machine learning and OR techniques are just a few examples of approaches we use to leverage the information. Our lab is set up in such a way that we are both able to perform real-time analysis as post-hoc more fine-grained analyzes of specific topics of interest.

The Dutch government introduced a new health care system in 2006. An important element of the regulated competition is a system of risk adjustment. Risk adjustment refers to the practice of paying insurers prospectively a subsidy per person that is related to the expected health care expenditures of that individual. The system of risk adjustment levels health care expenditure differentials between insurers that are due to differences in their population mix. This is important, since insurers are obliged to accept all enrolees for the same flat rate premium. In the absence of risk adjustment, insurers with a less healthy population will have a competitive disadvantage, as they must charge higher nominal, community-rated, premium.

In this project the objective was to analyse how the composition of an insurer's portfolio affects the risk adjustment compensation; specifically the project studied the question whether an insurer with less healthy policies in the portfolio would get fully compensated for the unfavorable risk it is exposed to.

The research issues were addressed by a simulation study. A simulation model was developed representing a portfolio of one million policies. This model was estimated and validated by insurance data. The development of the portfolio was simulated over a five year time horizon. Various scenarios were implemented that represent people's mobility behavior concerning changing insurance company based on the prices of the insurance premiums. We developed our own software in Matlab for executing these extensive micro-simulations.

Based on the results of the project, the insurance company gained insight into the expected development of the composition of its portfolio at various scenarios of premium increase and decrease changes. Furthermore, the simulations resulted in estimated profits/losses associated with these scenarios.

Reliability theory in engineering, duration analysis in economics or sociology or survival analysis in biology or medicine all deal with the same problems: what is the fraction of a population of individuals or products which will survive past a certain time? Of the survivors, which part will die or fail within a certain time? Here the definition of lifetime and death are ambiguous and can be interpreted in different ways, depending on the application. Many standard techniques can be applied to answer the questions just mentioned. The problem becomes interesting from a mathematical point of view if it is not a standard problem; for instance if several factors affect the survival time, there is dependency between the lifetimes of the individuals/products, if individuals/products are selected instead of randomly chosen from the population, or there is a lot of missing information.

So far, we have applied survival analysis in medicine (for instance, what is the probability a woman will have breast cancer within 10 years?), psychology (do monozygotic twins ask for social help more often and at a younger age than dizygotic twins do?), biology (what is the expected time two particular cells interact?), and historical demography (what is the life expectancy of a woman from the seventeenth century or for a landholder in the thirteenth century?). The application can be simply changed to answer questions like: what is the probability an individual will have a serious car accident within 1 year? Are males more often involved in car accidents than women? What is the expected time an individual will drive a car without submitting claims?

Within capital intensive industries, companies are constantly balancing high availability of assets against minimising costs. For example, due to the high costs of airplanes, every minute a plane is on the ground money is lost. In order to reach high asset utilisation, stocking reliable spare parts (modular components) and fast replacement procedures are important prerequisites. Managing spare part inventories, often spread across multiple warehouses, is crucial in this process.

Logistic Service Providers (LSP), who are responsible for spare part availability (e.g. KLM who provides parts to other airline companies), usually operate under criteria captured in complex and elaborate service level agreements. Time based service levels are such an example (e.g. "deliver 95% of the parts asked for within 72 hours"). Next to this, the fact that customers of these LSPs are usually geographically spread, each having multiple warehouses with their own specification of stock levels, also adds to complexity.

The aim of this project was to develop an inventory optimisation model that can be used for planning target inventory levels for each item at each warehouse that deals with both SLA and costs constraints.

Traditional inventory optimization techniques use the so called instantaneous fill rate as dominant performance measure. For organizations operating under strict time based service levels, instantaneous fill rate techniques are less adequate. In such a situation the focus must be on time based fill rates, explaining the fraction of demand that can be fulfilled on time.

In a case study performed at KLM, we developed a simulation model that used time based measures. This model showed that, when operating under strict time constraints, the use of time based measures perform significantly better than models that use traditional performance measures. The use of the appropriate (time based) measures does not only save money, but also adds to supply flexibility.

Summarizing, this study enhanced insights in the relation between time based service levels and inventory optimization. By implementing time windows in inventory optimization, companies can directly reduce inventory investments, whereas the inappropriate use of instantaneous fill rates as performance measure will lead to overinvestment in inventories.