There are tasks such as the retraining of the models and incorporating new data points from feedback in excess of regular monitoring and maintenance (CI/CD) activities. Strategies and concepts on how to deal with these items is helpful and partly the responsibility of the development team.
1. Retraining
+ Strategy & Concept: During its use, product or service accuracy and performance levels might drop. Retraining might be useful. Adding additional data points might also be cause for a retraining. Setting-up who is responsible or enabled to do so needs to be established.
2. Feedback Loop
+ Strategy & Concept: Developing a strategy and a concept to integrate the information from the user feedback into the AI/ML system was already part of the ‘Output’ module. However, in the productive environment, this needs to be implemented and even automated on a specified basis, making sure that the amount of collected data points is enough to increase the
model performance.
(potential) Outcome: new project
Those strategic decisions can be supported by the HCD/UX as well as data science and ML engineers together with the business domain expert and the IT department. The final implementation of these strategies and concepts might be done by the IT department. This module might result in a completely new project, when over time, the product or service shows that a new problem has occurred, or developing a new solution would be beneficial for the user and the business. Also, new data points might make it necessary to go back to the ‘Input’ module, or a new AI/ML method might support going back to the ‘Modeling’ module.
Those strategic decisions can be supported by the HCD/UX as well as data science and ML engineers together with the business domain expert and the IT department. The final implementation of these strategies and concepts might be done by the IT department. This module might result in a completely new project, when over time, the product or service shows that a new problem has occurred, or developing a new solution would be beneficial for the user and the business. Also, new data points might make it necessary to go back to the ‘Input’ module, or a new AI/ML method might support going back to the ‘Modeling’ module.
