Project Insight and Forecasting

Challenge

Each year ProRail spends more than two billion euros to maintain and improve the railroad network and rail stations in the Netherlands. These are lengthy, complex projects that require various experts throughout the organization. 

ProRail wants to increase the efficiency of the projects carried out, and to monitor them better while they are running in order to make timely adjustments to projects running out of step. The organization also wanted to have a better forecast for the expert resources required to carry out these projects. Both for the individual projects, and for the total amount of FTE required for specific roles in the coming years. 

The overview will allow ProRail to plan ahead better: by either hiring more/different people or by changing the planning on projects, ProRail is able to make sure its resources are used most efficiently.

Our solution

We created an end-to-end solution, starting with ingesting raw data and ending with several dashboards. Our solution forecasts the hours needed on planned projects, and the total FTE needed throughout the organization for various roles. 

Project data is ingested and transformed on ProRail’s Azure platform, where a databricks solution is used to train a machine learning model on historical data, and make predictions for current and future projects. Our solution includes a clustering algorithm that clusters similar projects together, to ensure forecasting results remain explainable to stakeholders. In addition, many business rules have been implemented in close collaboration with the relevant stakeholders.

The results of the model are delivered to the QlikSense server where it is loaded into various interactive QlikSense dashboards where end-users can view forecasts on the required level of detail. Additionally, forecasts are included in quarterly reports to fit into the existing decision-making processes and make them more data-driven.

A particular challenge in this project was cleaning and transforming the data, while still informing end-users of the input of our model: what information is the forecast based on? In order to tackle this problem we developed several monitoring and data quality dashboard pages, which allow end-users to examine and understand the changes in the data which leads to a better user acceptance of the forecasts.

A related tool developed by Lynxx enables project managers to make a better initial assessment of the required resources for a project. Project parameters such as duration, budget and category can be entered into the tool, which then shows related historical projects. The expert can then select and deselect relevant historical projects, in order to challenge their expert judgment on how many man-hours are needed for each project.

Results

Our forecasts are presented in dashboards to end-users, but are also sent out in quarterly reports to the management team and individual regional teams. The forecasts are then used as input to the planning sessions for the coming quarters / years.

This allows ProRail to make better data-driven decisions on several different levels. On the one hand ProRail gains insight into how many FTE per role will be required in the coming years, aiding its recruitment and planning strategy. On the other hand, regional managers use the forecasts generated for their region, both on an individual project-basis, and aggregated for their region. This allows managers to closely monitor ongoing projects.

An additional benefit from the project is that the richness of project information is now easily accessible by project managers and stakeholders. This makes it easy to monitor a change of scope, budget or duration of a project,

We are continuously improving our forecasts by including additional data sources, business rules, and expanding the scope of the projects. We are also working with stakeholders to further integrate the forecasts and other insights into decision making processes.