Conclusions and Recommendation to Philips: Prediction of Suppliers’

sustainabil-ity score using different parameters as dependent variables. There are 3 different dependent variables in the models. The maximum improvement potential, the learning rate and the ini-tial score are used as dependent variables. The maximum improvement potenini-tial and learning rate are common parameters for all the learning behaviors. However, the initial score is used

as a dependent variable only for lagged fast learner suppliers as their learning process starts at sequence 2 which is considered as the initial score. The same independent variable set is used for all models. However, the important variables are different for each topic and each parameter. XGBoost regressor and random forest regressor are used to predict the parameters by using supplier characteristics. By looking at RMSE values, random forest regressor are cho-sen as the selected model as it outperforms XGBoost at most of the sustainability field. Also, satisfactory performance was obtained by using random forest regressor. However, the perfor-mance for predicting the learning rate of slow learner suppliers in business ethics topic might be improved as the performance can be considered as poor. The reason for poor performance in that topic could be the small sample size in the training data-set. Nevertheless, the models can be used for predicting the learning curves of suppliers for different sustainability topics.

Predicted learning curves can be used as a basis for understanding the suppliers’ willingness to participate in SSP. In addition to that, depending on the predicted improvement potential and learning rate, experts may provide more target-specific actions by looking at the learning curve of suppliers. Also, the learning curves of new suppliers in the program can be predicted with the models. It might help experts and planners to understand at what degree suppliers need guidance.

The models are embedded as a tool in Microsoft Excel and linked with Python for future usage.

The layout of the tool can be seen in Appendix M.

5 Structuring the sustainability improvement process by incorporat-ing suppliers’ learnincorporat-ing behaviors

The third RQ is about structuring the sustainability improvement process by incorporating suppliers’ learning behaviors into the decision-making phase. Models that are constructed for RQ3 are investigated, and an application as a company case can be found with the results.

5.1 Theoretical Model

Companies that are willing to improve their supplier sustainability development program can use the models that are presented in this section. Companies that have already a sustainability improvement program where suppliers are given improvement action plans related to sustain-ability topics, a strategy or a decision-making tool might be beneficial to provide tailor-made improvement suggestions for the suppliers while considering their resources and capabilities.

For constructing such a model, measuring the suppliers’ capabilities and resources, the buyer involvement as providing guidance or financial support for the suppliers in the program must be implemented. Therefore, in this section, a framework, that can be applied to companies who have a supplier sustainability program and measure their suppliers’ sustainability score, is explained.

5.1 Theoretical Model 5.1.1 Conceptual Framework

Supplier sustainability improvement process can be improved by considering the results of RQ1 and RQ2. As RQ1 reveals that suppliers have different learning behaviors, the learning behavior information can be included in the decision-making process. For example, depend-ing on the learndepend-ing behavior of suppliers, each supplier might have different learndepend-ing capacity which can be used while providing action plans and implementation. In addition to that, findings in RQ2 can be also incorporated in the process as the learning curves of suppliers might imply how much resources or capacity a supplier is willing to allocate to the program.

Consequently, for the companies that do not keep information related to suppliers’ capacity and resources, learning curves can be a solution for estimating the suppliers’ learning capaci-ties beforehand. Consequently, an optimization model that considers the learning behavior of suppliers, capacities and resources of suppliers and investments of the buyer as buyer involve-ment and sustainability improveinvolve-ment results from action plans must be built to structure the sustainability improvement process.

5.1.2 Methodology

For RQ3, as the main purpose is constructing an optimization model for structuring the sus-tainability improvement process, mathematical modeling techniques such as linear or integer programming can be used. To include the learning behaviors and learning curves of suppliers, outputs of RQ1 and RQ2 is used.

5.1.3 Model

Theoretical model constructed for solving RQ3 is explained in this section. The optimization model considers that buyer involvement can be in terms of financial supports to suppliers or providing guidance or training for suppliers. In addition to that, according to the learning be-haviors and learning curves of suppliers, each supplier has a learning capacity that should be considered as suppliers’ learning capacities for the sustainability improvement process. Fur-thermore, buyer involvement might enhance that learning capacity, as the buyer involvement is considered as a supply chain collaboration which increases the willingness of suppliers for participating in the sustainability program. Suppliers have resource constraints in terms of learning capacity which is estimated from its learning curve. On the other hand, buyers have resource constraints in terms of time and budget, which are incorporated for an upper bound for providing guidance or training and financial investments to suppliers. The model aims to assign actions to suppliers within their learning capacity limits while providing buyers which suppliers to collaborate with, at what degree and how. The degree of buyer involvement can be understood by how much time to allocate a supplier for training or finances must be provided to suppliers as investments to enhance their improvement potentials or learning capacities.

Sets, parameters, decision variables and the mathematical model presented as follows:

5.1 Theoretical Model Sets:

I :Set of actions for suppliers J :Set of suppliers

K :Set of sustainability topics

Parameters:

lbjk : Learning behavior of supplier j in topic k, ∀j ∈ J, ∀k ∈ K a_i: Improvement potential for action i, ∀i ∈ I

ic_jk : Improvement capacity in topic k for supplier j, ∀j ∈ J, ∀k ∈ K f os : Fixed cost of conducting on site visit

vos : Daily cost of conducting on site visit

days :Buyer capacity for conducting on site visits in terms of days

actionsij :Binary parameter represents if action i for supplier j not already implemented, ∀j ∈ J, ∀i ∈ I budget : Buyer capacity for investments and conducting on-site visits in terms of MU

osincj,k :Parameter represents capacity increase from on-site visits for supplier j in topic k invincj,k:Parameter represents capacity increase from investments to supplier j in topic k M : Sufficiently large constant

1, if action i in topic k assigned for supplier j 0, otherwise

1, if supplier j visited on site 0, otherwise

, ∀ j ∈ J

z_jk = Duration of on-site visit for supplier j in topic k ηjk = Investment amount for supplier j in topic k

Objective Function:

5.1 Theoretical Model

Objective function maximizes the total absolute improvement potential of all suppliers. Basi-cally, if an action i is assigned to supplier j, xijk = 1 . There is also a parameter called initial capacity which is a function of the learning behavior of a supplier and sequence number of a supplier. Improvement potential comes from actions is determined as it can be estimated from historical data or it might be already defined at the company. The model is constructed with different sustainability topics as it can be the case for most of the companies. However, it is also possible that set K can be excluded to focus only a topic or the overall score itself.

Constraint (1) guarantees that the learning capacity of the suppliers be less than the total of improvement potential coming from all the assigned actions. Here, the learning capacity is defined as follows: the initial learning capacity estimated from learning curves without any buyer involvement, learning capacity increase results from financial investments to suppliers for the related sustainability topic, and learning capacity increase results from on-site visits which are considered as providing guidance and training to suppliers. The learning capacity increase from a day of on-site visits and a monetary unit of investment can be estimated by using the historical data. Constraint (2) ensures that an action cannot be assigned to a supplier if the action is already implemented in that supplier. Constraint (3) refers to the buyer cannot allocate its resources to a supplier for on-site visits, if the supplier is not selected for an on-site visit. Constraint (4) ensures that the total duration of on-site visits for all the suppliers can-not be more than the suppliers’ capacity for conducting on-site visits. Constraint (5) is used for ensuring that the budget limit is taken into account. As there are costs that result from financial investments and conducting on-site visits to suppliers, the cost must be less than the total budget allocated by the buyer. The model can be considered as a Mixed Integer Linear Programming (MILP) which can be solved by using a simple or decent solver depending on the problem size.