4 ways AI neural networks will disrupt banking

“It should be better known as artificially inflated,” says Fabrice Brossart, CRO of AIG, upon discussing the role that Artificial Intelligence (AI) plays within the insurance giant.

---

A buzzword of many years, Brossart believes that there is not enough penetration into the various sectors of AI.

Chris Gledhill, a former Lloyds Group developer and current CEO of Secco, is of much the same opinion when it comes to AI.

“That’s the problem with emerging tech,” says Gledhill, “For whatever reason we don’t think about the unknowns-unknowns, chaos-style disruptions that may arise from the technology.”

For Gledhill, that technology, the one that he is most excited about, is neural networks, a sub-sector of the AI label. 

Artificial Neural Network (ANN) mirrors the concept of biological neural networks within the human brain. Made up of interconnected processes, ANN develops algorithms that can be used to model complex patterns the weights on the connections and, as a consequence, aid the user in decision-making.

Speaking to bobsguide Chris Gledhill reveals that “neural network AIs will have the most impact in the back office for roles that are performing menial tasks which more often than not are outsourced to offices in India. It is those roles that will be replaced by automation.

“In the short-term neural network Al will help with risk profiling, credit scoring and trading.”

Here's our four ways neural networks will disrupt banking:

1. Improved evaluation of loan applications

In order to grant a loan application, the aim of banks is to reduce the failure rate of loan applications and, in turn, maximise the returns of the loan issued.

For this reason, more and more banks are now beginning to use ANN to aid them in the decision to grant a loan application. Such a process works by analysing past failures and making decisions based upon past experiences.

ANN provides an 85%-90% accuracy in the decision-making, which would be a significant improvement over human-methods.

2. Help credit card companies find the ideal customer

To maintain a sustainable, profitable revenue, credit card businesses need to find the ideal customer. Should the customer not make adequate use of the credit card, the business’ profit would be significantly hindered, as the incremental and incidental costs would exceed the revenue, which would then result in a non-profitable business.

As a result of this, more credit card companies have started to leverage ANN to identify the best customer that would generate sufficient revenue, through the evaluation of the client’s credit-card-habits.

3. Prediction of stock market/stock market index

Through the use of historic data and prediction based on different parameters, ANN are used to handle predictions of both stock market indices and stock values.

When it comes to the prediction (via ANN) of such indices and values, the accuracy is enhanced through the use of hidden layers and more training variables.

For instance, with the case of the daily NASDAQ stock exchange rate prediction, it was discovered that a network with three hidden layers, coupled with 20-40-20 neurons in equally hidden layers, was the optimised network with an accuracy of 94.08% for validation dataset.

4. Better detection of fraud through character and image recognition

Due to ANN’s ability to take in multiple of inputs, process them to deduce hidden, as well as intricately interwoven relationships, ANN plays a significant role in image and character recognition. What this means is that more banks can take advantage of ANN to better detect fraud. For instance, banks could identify whether two written-signatures derive from the same person.

 

New tech, new ethics?

The moral issues associated with ANNs can be problematic. When it comes to learning patterns and identifying trends in data, ANNs are prone to forming their own biases which can, in turn, become prejudices when the discriminated data segments also correlates with defined segments among customers.

Gledhill makes it unequivocally clear that “if you ask the machine” based on “historical data of repaid and defaulted loans”, whether “the next person will repay or default […] the answer is probably not one you want to hear, because after all, you’re asking the machine to discriminate, often by group, and answers can be racist.”

He suggests that “you’d need some way of implementing a regulatory framework or moral code to ensure that the machine doesn’t discriminate unfairly.”

It is thus imperative to take ANN with a pinch of salt, not as a be-all and end-all solution, but rather as a tool to the human user, flagging areas that require a human touch, all the more pressing with GDPR.

 

 

• Read more under:
• BankingTech
• InsurTech