Automation of machine learning models

Reduce bottlenecks by setting up automatic retraining and deployment


Financial Crime departments within banks are developing an increasing number of new (machine learning) models. Making these models ready for production, together with the frequent retrainings is increasingly becoming a bottleneck for the whole workflow. At the same time criminals are constantly changing the methods of financial crime hence retraining is a necessity.


Automatic retraining and deployment may the key to solving these issues. RiskQuest first established a mature deployment platform by leveraging software accelerators such as Azure ML Ops and ML Flow. These accelerators allow for a faster time to market of development models and ensure full reproducibility and traceability. The main advantage of this mature platform is automation of deployment and (e.g. weekly or monthly) and automatic retraining. Data scientists can bring models live without the need for additional IT capacity and new forms of financial crime are directly incorporated within training. Moreover automation on reporting brings many advantages since it is common for machine learning teams to spend more than half of their time maintaining existing models.


In this project the team assisted in the development of a mature deployment platform. These improvements allow to reduce the bottlenecks that IT departments encounter when bringing to production machine learning models. They also drastically reduced the time data scientists spend to maintain existing models. Moreover by automating the retraining process, new forms of financial crime were spotted directly.

Related cases
Anti-money laundering
Related cases
A data-driven proactive client strategy