Client management and specifically on-boarding in banking and wealth management is still a tedious and complex process. It typically requires collecting, cleaning and processing of client data, applying a number of more or less automated rules and finally some human intervention to make a final decision to flag or unflag clients.
In a earlier articles https://www.linkedin.com/pulse/machine-learning-wealth-management-part-1-patrick-rotzetter/ and https://smartlake.ch/machine-learning-in-wealth-management-a-revised-view/, we went through some of the potential of machine learning in the on-boarding process, now let us go a step further and list down potential projects to enhance the client management process and specifically the on-boarding process to start with.
All the problems linked to client management belong to so called classification problems, where we want to find a way to classify information in certain buckets ( like yes or no). Typical classification problems that can be supported by well known machine learning classification models to improve human or semi-automated accuracy, increase automation and ultimately continuously improve the accuracy of the classification problems. These algorithms are well known, do not need a lot of computing power and large amount of data like neural networks. Prototypes can be built quickly even using Machine Learning as a Service for example.
Typical projects to implement intelligence in the on-boarding process would include:
- Prediction of prospect conversion
- Segmentation of clients
- Risk assessment with regards to anti money laundering
- Sector and industry risk assessment
- Client suitability
- Product matching classification
- Document and signature matching
Once the client is on boarded, we can also augment following processes with additional intelligence based on classification methods:
- Product target segmentation
- Continous risk monitoring
- Churn prediction
So what is next ? A typical process can look like this:
- Identify the problem or challenge you want to address: for example I want to better predict customer conversation rate or I want to better segment my customers for risk assessment purpose
- Define which metrics you will be using to check if the model will be better than today’s situation: for example accuracy of conversion prediction, false positive rate, risk classification error
- Collect the required data, clean the data ( missing data for example)
- Visualize the data to get a sense of what is available and to show easy to identify patterns or correlations
- Start testing classification models ( logistic regression, random forest, decision tress and others)
- Find the one where the metric match the defined objectives
- Prepare findings
Do not hesitate to comment to discuss further.