To effective IRB data management (and beyond)
Risk predictions for 2018 - #7: Getting back to traditional risk and increasing profitability
Three AI questions lenders must answer in 2018
13 February 2018
Advances in AI and machine learning will impact organisations beyond simple technical capability. They will face new types of challenges in terms of skills gaps, different approaches to implementation and new methods for monitoring models.
In this blog, we explore how lenders can prepare for AI implementation, to ensure they invest wisely in new technologies and stay ahead of their peers.
In order to remain competitive, financial services organisations need to make smarter, faster decisions. The best way to achieve this is by embracing AI-driven technologies, however, there are some key questions that lenders need to consider before embarking on this journey.
Is your data ready for AI?
The best technologies in the world can only make use of the data we provide. It’s crucial that organisations have good quality data and appropriate metadata, in a consistent format, to begin with. Even if organisations already use a data warehouse, the information still needs to be transferred to an appropriate analytics platform to deliver insights via mathematical modelling.
From data quality to data storage, the process of transforming data into insight needs best practice data management. What’s more, with new GDPR laws, these practices are more important than ever before. AI predictive models will assess whatever data is presented to them, therefore getting data management right at the beginning is essential to the compliant use of AI.
Are your people prepared for AI?
AI technology is growing in sophistication, but the level of knowledge in the industry isn’t necessarily keeping pace. In a recent Infosys research study, 53% of organisations surveyed cited developing knowledge and skills as the key to preparing for AI deployment and use. Alongside this, many organisations recognised the value of outside specialist help – whether to assist with planning (46%) or for knowledge gathering (40%).
That’s why it’s crucial that businesses ensure their people have enough knowledge of the analytics behind the AI models and technology systems. Many organisations find it useful to bring in specialist outside help to help with knowledge transfer during set-up and deployment.
How do you solve the black box problem?
The final area that is still causing hesitancy in the adoption of deep neural networks in credit scoring is the ‘black box’ problem. Without more certainty in interpreting, predicting and understanding decisions made by deep neural network models, their applicability to the world of lending was always going to be limited – especially in credit risk, where full transparency is a must for the highly regulated industry.
There are a number of known methodologies to constrain the AI but in most circumstances, they are reductive such that the outcome is no better than with traditional models. The key is to ensure a methodology that avoids the earlier issue of modest gains, which is the thinking that we applied to solve the ‘black box’ issue once and for all, while also achieving attractive uplifts in performance.
The technology and expertise now exist to create intuitive deep neural networks, which ensure models behave in an intuitive and understandable way and allow businesses to impose constraints on specific fields within the model so that the output adheres to certain business rules. For example, you might expect credit risk to increase as salary levels fall, or decrease as disposable income and affordability improves, and the model needs to reflect these expectations.
By guaranteeing those behaviours, the methodology ensures the models obey common-sense relationships and can be understood by the business and signed off through normal governance routes.
This development opens up the real possibility of having automatically-generated models, constantly learning and updating from new data sets, with no black boxes to worry about. This allows for ’always optimal’ model performance as models can be redeveloped at will, without the usual cost and effort of redevelopment.
A new era for AI
The successful use of AI presents organisations with many challenges. Yet with best practice data management processes, including compliance with GDPR, the right skills and external support, and crucially, the right technology, organisations will be well positioned to take advantage of the new era of AI.
For organisations to harness the benefits they need to act quickly – AI is currently a hot topic across most industries and plenty of organisations are starting to make moves in this direction. And with no more black boxes to worry about, the business case for AI is significantly strengthened across the business thanks to the improved techniques now available – and the more consistent, transparent and accurate results it delivers.