Banking Solutions

Client here was looking to transition from solely relying on traditional underwriting and credit scoring approach to a machine learning model in order to accurately predict the credit worthiness of the borrower across a range of attributes and factors not part of regular underwriting. This involved

• Integration of learnings and past performance of the loans into the ML model.
• Bringing in the individual characteristics of the borrower such as personal data, public data, credit history, environmental factors etc.
• Integration of NLP solutions to bring in unstructured data from social media platforms.
• Model went through training and self learning with human assistance to help with robust underwriting decision making.

Machine learning is hugely capable of capturing non linear risk factors, which could otherwise slip through the gaps in traditional rules based risk management frameworks. Client here was using advance BI platform for portfolio segmentation and risk concentration reports. They were looking to reduce portfolio losses by augmenting LPM with AI to enable

• Early detection of risk concentration in a portfolio by segmenting it on a range of risk parameters with neural network model.
• Probabilistic graphs to provide better picture of developing risks and their sources.
• Models with data driven clustering that are extremely effective in detecting hidden corelations which may otherwise go unnoticed.

This client, a mortgage provider had deployed multiple models to manage the various stages of mortgage lifecycle. However internal audit revealed that there was a breach of compliance as there was no second line of defence to monitor models' performance and accuracy and the documentation was not clear and sufficient. Hence AI solutions were put in place 

• To monitor regulated activities, customer interactions and financial transactions.
• To identify false alarms and assist compliance analyst to focus on priority alerts.
• Apply advance analytics to continuously monitor and validate model performance and risk exposure.

This client was looking to bring in co-pilot assist into their L1 and L2 support levels to improve the service KPIs and to minimise the redundancies and latencies impacting the efficiency of the operations. The task was to

• Implement digital assistant to help the service desk personnel.
• Put an ML model in place that trains from system logs to detect anomalies and raise alerts and tickets.
• Deploy a fine tuned Action BOT for classification of ticket and prioritisation of support requests.

 A credit card provider wanted to optimise their collections strategy on non performing assets in order to minimise the cost of managing their portfolio. ReInvent Ideas' data scientists partitioned the portfolio into different risk segments and designed a tool to help the collections manager devise appropriate strategies for each risk category. 

• Tool was trained on historical data on delinquency and defaults.
• Tool was also trained on good and bad collections strategies, that way it could predict optimal strategies for customers under given scenarios.
• Provided proactive analysis of borrower's financial circumstances by combining data from origination, data from performing and non performing periods and bank data, plus data from other sources.

ReInvent Ideas was approached by one of its clients who were looking to enhance the cross sell opportunities to their customers through personalised offers on products and services. The ask was to develop 'behaviour scorecards' for existing customers which could then help in tailoring their journeys.

• Huge sets of customer attributes were extracted from company data warehouse and fed into logistic regression models for customer profiling.
• Behaviour scorecards were then applied on customer profiles to carve out individual journey plans.
• This simple scoring model was effective in fetching greater adoption and better response from the end customers.