Decision support modeling for milk valorization

The research presented in this thesis concerns decision problems in practice that require structured, precise, scientific studies to provide strong, reliable answers. An opportunity to contribute to both practice and science emerged in 2008 when two large, Dutch dairy companies merged, creating FrieslandCampina (FC), which was the fourth largest dairy company in the world at that time. In 2009, a new Milk Valorization & Allocation (MVA) department was created at the corporate level to optimally utilize raw milk (the main raw material) in all business units. The main goal of this research was the development and application of decision support models to help MVA attain its mission of “getting more out of milk.” The dairy processing industry is a specific and challenging research field. This is related to the fact that the raw milk is transformed into thousands of end products via highly interrelated production processes. These processes are affected by uncertainties related to supply, processing capacities, and demand. Attaining high profitability requires a central, integral planning process that facilitates the optimal allocation of raw milk to a large range of products. Optimal allocation of raw milk is achieved when it is successfully allocated to the most profitable end products and all important constraints are taken into account. This process is defined as <em>milk valorization</em>. Contribution to the improvement of milk valorization in the dairy industry was the main objective of this thesis. We approached the problem from a Logistics Management perspective. We focused on decisions supporting the optimal flow of raw materials to end products, from farmers to consumer markets. With the use of Operations Research techniques, we developed quantitative models and frameworks to improve the mid-term milk valorization process. As the first step towards the improvement of milk valorization we developed a mid-term Dairy Valorization Model (DVM). The model creates optimal plans for the allocation of milk, and the production of end products and byproducts. It captures the dynamics of dairy production and incorporates all relevant elements and constraints. The following elements were indicated as important and included in the DVM: recipes based on raw milk composition (dry matter, fat, and protein content); seasonality of raw milk composition and supply; a complete dairy product portfolio; by-product utilization; network of supply regions and production locations; by-product and raw milk transportation; and changes in sale prices. Including all relevant elements assures DVM comprehensiveness. This important aspect achieves truly integral valorization of milk. Furthermore, the developed DVM also fosters understanding of complex, underlying production processes. Moreover, by means of additional analysis we have also shown that the seasonality of raw milk components (dry matter, fat and protein) plays an important role in the valorization process. It considerably affects decisions regarding milk allocation to end products (up to 50% difference in production volumes of clustered end products) and company profit (up to 4% difference in monthly profit). Given the complexity of the dairy system, the development of a high class valorization model required a gradual approach. The developed DVM focuses on the valorization of milk-based end products (main milk products). The production of those products, however, results in large volumes of byproducts.In the second step of this research we investigated the effect of whey valorization (byproduct of cheese) on the valorization of main milk products, as well as the added value of integral valorization (simultaneous valorization of both main and byproducts). We developed a new Integral Dairy Valorization model (IDVM) to allow for an integral milk valorization. We also developed a three-step evaluation approach to compare results of stepwise valorization (in which whey valorization only follows after main milk products valorization) and integral valorization. The results show that the explicit valorization of whey flows leads to significant economic gains for FC. Profit obtained from post processing of whey byproducts amounts to circa 20% of the total profit. Furthermore, the effect of integrating both valorization processes is currently small (on average 0.0089% increase in monthly profit). There is, however, a potential in the integration of two processes. In case demand for, and sale prices of, whey-based products, sale prices of milk powders or processing capacity for whey increases, the gain from the integration can be considerably larger (up to 1200% stronger effect in comparison to the current situation). We have also shown that currently whey products are not profitable enough to drive the production of milk products that are the source of the whey by-product. In the next step we focused on the accuracy of solutions obtained with the DVM. Because the DVM is a deterministic model, uncertainties present in input are not incorporated, and as such the stability of valorization plans is affected. Stability of plans is often referred as to the ‘robustness’ of plans: <em>the degree to which the optimal solution might change if realization of certain input parameters turn out to be different than the forecasted values</em>. The robustness is important, because the valorization plans that are initially indicated as optimal can easily become sub-optimal or costly. Therefore, the overall goal of the third study was to develop a framework for robustness evaluation of valorization plans obtained with deterministic models. We developed a five-step framework comprised of the following: (1) definition of Key Performance Indicators (KPIs), (2) selection of relevant input parameters, (3) definition of scenarios, (4) evaluation of robustness, and (5) extraction of conclusions. The output from Step 4 of the framework is multidimensional, and thus to arrive at the final robustness degree, a number of decisions must be made <em>a priori</em>: acceptable KPIs limits (robustness bounds); evaluation time (month or year); evaluation depth (parameter or element); and the grouping approach of KPIs. The results show that depending on the selection of these aspects different conclusions regarding robustness of valorization plans are obtained, (average robustness degree varied from 48% to 92%), and thus the final conclusions regarding the robustness degree of plans is affected. The overall robustness degree of valorization plans (at FC) obtained with the DVM was 90% and was indicated by FC as sufficiently high to attain successful milk valorization. The calculated robustness degrees also identified the parameters with the greatest effect on robustness (composition and supply of milk). The effectiveness of valorization models is mainly linked to the optimality, feasibility and robustness of obtained plans. However, even if these three aspects are satisfied, the success of the valorization process is still very much dependent on the performance level of actors and units that are involved in the process. Given the fact that processing units (factories) are the most important units in the supply chain of a processing company, because they can easily affect the value of each ton of raw milk used in the production process, the last study investigated the performance of processing units. We developed two Data Envelopment Analysis models for performance measurement and improvement, and applied it to the case study of TNT Express. The models allowed us to identify: inefficient units (30%); parts of efficiency levels (of inefficient units) that result from either management practices (85%) or a favorable external environment (15%); potential reductions in consumed input resources that allow for the same output levels (17% less labor and subcontractors could be used); and role models that can be treated as master units in efficient use of certain inputs and thus should play leading roles in setting benchmarks. We concluded that in order to successfully valorize raw materials, companies should: develop their own valorization model, possess a comprehensible overview of the complete production system; and have access to necessary input data. Furthermore, there is a potential in integrating main product and by-product valorization processes. The added value, however, depends on the information on market and production capacities of by-products and related to them main products. To ensure that possible future integration of both valorizations processes occurs correctly, companies should investigate future market developments and the possibility of increasing production capacity. Moreover, we have also shown that robustness of solutions obtained with deterministic valorization models can be sufficiently high to obtain reliable plans. This means that it is not always necessary to implement complex modeling techniques (such as stochastic programming). To ensure accurate solutions, companies should also focus on improving forecast accuracies of parameters affecting the robustness. The robustness degree should also be regularly assessed with the developed framework. Finally, managers should also focus on performance levels of processing units. A DEA model should be developed to identify inefficient factories and provide new insights to improve performance. In order to properly valorize milk or other food resources to its maximum an integral point of view should be chosen. Operations Research techniques should be used because the complexity of many processing industries makes applying practical rules of thumb insufficient and often inadequate. The models and frameworks developed in this thesis provide new perspective on and new insights into the complex problem of milk valorizations. We have shown that analyses of results obtained with the developed methods can answer many managerial questions, and thus support the decision making process within a company. This improves overall raw material valorization, creates more value for companies, and leads to more sustainable dairy chains.