A data science project starts with a use-case workshop. It is important to include as many stakeholders and domain experts as possible to obtain a holistic view of the current situation, available data sources, limitations, the project goals and success criteria.

The objective of this phase is the precise definition of the project goals based on measurable criteria, the setup of the project team, the identification of data sources and the creation of a project plan.
Thanks to our many years of experience, it is already possible at this stage to avoid the first stumbling blocks caused, for example, by an incorrectly selected goal or data source/granularity.

As a next step, extracts of the selected data sources are loaded into a laboratory environment separate from production systems. Here, the data is checked for completeness and quality and data cleansing is performed, if necessary. In addition, data from different sources is joined and transformed from its raw form into a format suitable for machine learning.

After the data is available in a usable format, it is important to create a first working version of the Machine Learning Pipeline to evaluate the results against the defined success criteria.

The evaluation of the results helps to iteratively refine and verify the improvement of the machine learning model and avoids expensive go-live of insufficient results.

When a successful machine learning model was created in the laboratory environment, it is time to integrate the developed solution into your operational processes.

This includes the deployment of the model to continuously generate new predictions and the creation of an automated data pipeline to load and prepare new data. This phase is completed by establishing process monitoring and performing final performance tests in live operation.