Skip to content

Client case

Get useful answers from your data

Problem

In cooperation with an international knowledge institute from Delft, we carried out a project to see whether it is possible to make predictions regarding budget overruns on projects, by means of data analysis techniques (machine learning). Can we predict at an early stage which project will run on budget and which project will be exceeded by 20% or more? 

Maarten has led this project. He has carefully composed the project team and safeguarded the necessary project preconditions. He made sure that international knowledge was available on time. In a short period of time, he achieved great results with this team and learned a lot about what does and does not work around machine learning.

Approach

The project had an exploratory approach. With a small team, composed on the basis of Belbin profiles, we worked together for six weeks on drawing up a data model. For this model we used the seven steps of machine learning. A step-by-step plan that Google also uses successfully.

Results

After six weeks we had a working algorithm and presented our results to the management. In addition to deeper substantive knowledge on new techniques, such as machine learning, the lessons learned from this project apply to many more projects. For examples: (innovative) projects require trust between team members, select team members who complement each other, make agreements in advance about semantics (language) and make sure you have data in order (structured and well maintained) before you start. Some of our key indicators which help predicting budget overrun are the ‘core team size’,  the department which is responsible for the project, the average hours per day a person is spending on a project and the seniority of the team.

Decisive factors during this project

As we see it, the crucial moment was before the start of the project. Agreeing together that we should stop talking and analyzing, but just free up the time, budget and space to start this kind of exploratory project, without guarantees on the outcome.