This thesis examines whether explainable machine learning can support bid/no-bid decision-making for large Dutch construction projects. The study is conducted as a case study at Count & Cooper, a Dutch project-management and tender advisory consultancy active in complex construction tenders. The bid/no-bid decision is a critical early decision point, because contractors must decide whether to invest substantial time and resources in preparing a tender while important project information is still uncertain. To address this problem, the thesis first analyses the Dutch tendering context and identifies decision factors from literature, desk research, and expert validation. These factors are then operationalised into a structured dataset based on 101 historical tender opportunities. Several supervised learning models are compared, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, and XGBoost. Because of the limited dataset size and data-quality constraints, model performance is evaluated using repeated 5-fold cross-validation. In addition to predictive accuracy, the thesis emphasises interpretability by analysing model coefficients and SHAP-based feature importance. The results show that bid/no-bid outcomes can be predicted to a moderate extent using characteristics that are available early in the tendering process. Random Forest achieved the strongest overall predictive performance, with Logistic Regression and linear SVM performing similarly well. Across models, tender duration, contract duration, project value, project type, and contract form emerged as influential drivers. At the same time, the analysis shows that important internal decision factors, such as strategic fit, current workload, and client capability, were not consistently documented in the available historical records and therefore remain outside the current modelling scope. The thesis concludes that explainable machine learning can provide useful and transparent screening support for bid/no-bid decisions, but that the current results should be treated as a proof-of-concept rather than a deployable decision tool. The main practical recommendation is therefore to improve the structured capture of tender characteristics, internal decision context, and decision rationale, so that future models can become more robust, realistic, and decision-relevant.
... Read more

Megaprojects frequently face cost overruns and schedule delays, a pattern known as the iron law of megaprojects, which persists partly because traditional determinis tic or probabilistic planning methods are inadequate for managing deep uncertainty. This form of uncertainty arises when the probability, timing, or impact of key events cannot be reliably estimated, often leading to unrealistic schedules and ineffective risk responses. This study investigates the question: How can Exploratory Modelling and Analysis and Dynamic Adaptive Policy Pathways be applied to improve schedule robustness in infrastructure construction projects? The research focuses on the Schiphol bridge reconstruction, which is part of the Veenix A9 BaHo project and is conducted in collaboration with Count & Cooper. A Discrete Event Simulation (DES) model is built in SimPy and structured using a task dependency graph de rived from the project’s original schedule via NetworkX. The model is sampled 10,000 times under baseline conditions using Latin Hypercube Sampling. In scenario discovery, Patient Rule Induction Method (PRIM) is used in combination with a scaling function identifying six high-impact scenarios, which serve both as inputs for robust policy search and as Adaptation Tipping Points (ATP) for the Dynamic Adaptive Policy Pathways (DAPP) schedule. Robust mitigation strategies are derived using a Multi-Objective Evolutionary Algorithm under the Multi-Objective Robust Decision Making framework, with a second PRIM experiment selecting four final robust policies. These policies correspond to at least one of the high-impact scenarios and form the backbone of a conditional DAPP schedule. The DAPP schedule is evaluated against a static baseline using 5,000 DES simulations with identical uncertainty sampling. In 20 comparative runs, it reduced project duration by an average of 67 days and cost by approximately e97.5 million on the entire project schedule. All three robust policies include the measures new design, overtime labour, and electric machinery, suggesting that a focused subset of actions can improve resilience even when the future is highly uncertain. Unlike prior Decision Making under Deep Uncertainty (DMDU) applications, which often focus on long-term or high-level strategic planning, this study embeds adaptive logic within a highly granular, task-level construction schedule based on real project data. This approach raises methodological challenges in how adaptation tipping points are defined, triggered, and monitored within network-based simulation. The findings demonstrate not only the feasibility of combining Exploratory Modelling and Analysis (EMA) and DAPP in operational construction settings but also the need for further research into real-time scenario recognition and policy switching mechanisms under uncertainty. ... Read more

Organisational learning has been essential to the success of project-based organisations, especially in the construction sector, where complex infrastructure projects require effective knowledge sharing, retention, and application. However, many organisations have been struggling to fully utilising their accumulated knowledge of the organisational memory, leading to inefficiencies and improvement opportunities have been missed. In this research has been investigated how social learning is activating organisational memory and is being applied within project-based organisations.

To explore this, the research has been structured into three parts. In part 1 the theoretical groundwork has been laid through a literature review on social learning, organisational learning, organisational memory in project-based organisations, and memory activation dynamics. In part 2 practical insights from semi-structured interviews with ten tender experts at Count & Cooper have been gathered. The thematic analysis of these interviews has been revealed how organisational memory, organisational learning, and social learning are interconnected within the dynamic environment of project-based organisations. In part 3 a problem-based learning (PBL) workshop simulating a portion of the tender process has been involved, providing practical observations on how social learning practices are influencing organisational memory use in decision-making.

The findings have been revealing a complex, dynamic flow within organisational memory, showing how it is being accessed and is being activated in non-linear ways through social learning practices. This research has been challenging the view of organisational memory as a static repository, showing it is continually being reshaped within project-based organisations (PBOs). Pattern recognition, emotional triggers, and competing memories have been shaping how organisational memory is being activated and being used. Through social learning, members are engaging with various forms of conscious, automatic, objectified, and collective knowledge, often uncovering overlooked or competing knowledge of the organisational memory. Social learning has been facilitating knowledge retention and application while guiding members through these diverse memory flows, aiding them in navigating and learning from organisational challenges.

The research has been highlighting a shift toward more structured social learning practices to ensure systematic sharing of insights across the organisation. While unstructured practices, such as informal mentorship, have been remaining essential for collaboration, time pressures in PBOs have often been limiting the revisiting of prior knowledge, causing insights to become siloed and leading to organisational forgetting. Yet, this process has also been promoting growth by discarding outdated knowledge, allowing room for innovation. Consequently, organisational memory has been emerging as both a valuable resource and a potential liability, depending on how it is managed and have being used.

In conclusion, this research has been providing insights into how social learning practices are generating and activating organisational memory, allowing organisations to capitalise on it as a resource. Through social learning, members have been using pattern recognition, emotional triggers, and the resolution of competing memories to increase awareness of organisational memory’s facets, ensuring relevant knowledge is being accessed, being applied, and being used for learning. This approach has been supporting decision-making, has been reducing corporate amnesia and has been fostering continuous learning across the organisation, preventing the reinvention of the wheel by drawing on different parts of organisational memory. By fostering environments where social learning is being encouraged where knowledge is being exchanged, validated, and applied by which organisations have been reducing knowledge fragmentation and forgetting, enhancing accessibility, and preventing silos. The findings are suggesting that project-based organisations would be benefitting from a hybrid social learning model combining structured and unstructured approaches.
... Read more