| | Credit Risk Management by Experian

Successful operationalization of Machine Learning and Data Science projects

Machine learning, artificial intelligence and advanced analytics change financial processes along the entire customer journey in credit management – from application to dunning and collection.

The first step towards a successful Advanced Analytics project is understanding the requirements of the respective use case in depth and developing a good model from this. However, the operationalization of the model is just as important for the success of the project. This includes conducting ongoing evaluations of the model quality and integrating it into operational business processes. This is the only way to permanently ensure the model’s efficiency as well as the effectiveness of the recommended actions derived from the model.

Causes of inadequate operationalization

In practice, however, it can be observed that more than half of all data science projects are not fully deployed. There are three main reasons for this:
1. The models are either not or only insufficiently integrated into the existing process flows of an organization, e.g. because this is not possible due to different processing speeds.
2. The data science team developing the model is neither familiar with nor responsible for operationalizing the model.
3. The deployment process is not managed holistically – for example, there is no corresponding communication structure between the data science team, the IT infrastructure team and the stakeholders.

Factors for successful operationalization

But how can a successful operationalization of machine learning models be ensured? In our projects, four factors in particular have emerged that will help enable complete and successful deployment.

1. An in-depth understanding of the processes of all stakeholders involved is a fundamental prerequisite for successfully developing the solution. It must be ensured that the model requirements fit the existing business processes. To guarantee this – and ultimately the success of the project – all stakeholders should be involved right from the start of the project. After all, the introduction and operationalization of a machine learning project in an organization should be regarded as a holistic corporate task and not as an isolated special project within the analytics team that only affects part of the company.

2. In order to make the best possible use of existing expertise in the areas of analytics and IT infrastructure for the project, the collaboration between the Data Science and Data Engineering departments should be continuously managed and optimized. An iterative approach with the shortest possible project intervals has proven to be the best solution. This allows necessary improvements to be quickly identified and implemented. In the event of failures, this is done at an early stage so that findings can be incorporated directly into the next project interval. In addition, this approach minimizes the economic risk and continually provides possibilities to make modifications.

3. Being able to explain these machine learning models increases their acceptance within the company and beyond. Depending on the model, it may be more or less difficult to explain the underlying processes and methods. Currently, many projects fail due to lack of explainability.

4. Sustainable application management is necessary to ensure the long-term operation of the model and to take factors such as continuous monitoring and 24/7 operation into account.

What is your experience with the operationalization of machine learning models? Where do you stand with your Data & Analytics Roadmap? I am looking forward to receiving your questions and comments by email.

Related Posts