ProtectRobo

Psychographics and alternative data

With ProtectRobo Lenders can improve their risk management by getting a continuous monitoring of their customer base. Think about default predictions or churn. It’s on an individual level so based on these early warnings a lender can take preventive actions with specific treatments which will reduce the risk in the customer base. The predictions are based on the Ai models on the AR platform and can feed directly into the banks system through an API.

Methodology

The AI driven risk prediction models are ready to use but can be tailored towards the specific Lender. This depends on the quality and quantity of the available customer behavioural data.
The models are developed following a strict procedure and will make optimal use of the available customer data at the lender. Also, Psychographics could be added to the models and so create a much richer customer profiling.

Methodology – Approach

Methodology – Quality of model

Metrics for binary classification

• Accuracy



• Precision



• Recall



• F1



• Matthews

Visual representations

• Area under the ROC



• Confusion matrix



• Life chart

Case study

The challenge

A tier 1 global bank with a large portfolio of credit cards in Latin America has the ambition to support customers better that are or could become vulnerable.

They selected AdviceRobo because the combination of predicting vulnerability and enriching this with Psychographic profiling could help them not only with predicting the risk but also developing treatments for the specific customers.

The objective was to beat earlier developed prediction models (numbers cannot be disclosed)

The approach

Around xx customers were invited to fill out the online questionnaire. They were rewarded by receiving a top up for their mobile telephone when completing the questionnaire.

The bank delieverd the performance data of these customers and AdviceRobo combined these with the Psychograhic scores and profiles.

The results

The customer experience showed an excellent result:

• 92% of customers started after clickthrough
• 83% completed the questionnaire

Linear modelling was applied since the number of vulnerable customers in the sample set was too low for Machine Learning technology.

The results of the AdviceRobo’s Psychograpic scores (PCS) show good results and better than the existing generic vulnerable client score. The model results whowed that precision doubled as well as the recall.

Also, a model ws developed in which PCS and the generic model was combined. This improved the results a further 20% on top for precision as well as recall.