Customers are becoming more informed and expect transparent as well as consistent and reliable services; often these services are accessible on a 24/7 basis. AI can be leveraged to obtain a holistic view of the customer and provide timely support.

Industry example:

• Text-to-speech technology is being used by a large Australasian bank to support natural interactions in customer support channels.
• Disaster response models are being applied by multiple Australasian banks to monitor multiple data sources from official emergency sources and alert systems. Using a customer engagement engine, banks are able to connect with impacted customers.
  

The Global regulatory Outlook 2020 has indicated that 33% of banks have allocated or will allocate more than 5% of their annual budgets to compliance. This is a major opportunity to leverage AI technologies, however according to Deloitte’s State of AI in the Enterprise, 3rd edition, only 4% of the respondents in financial services used AI in legal and compliance.

Industry example:

A large Australasian bank used AI for outlier detection to validate a model calculating investment projections. This enabled the identification of areas for potential improvement that traditional model validation techniques would struggle to pinpoint.

AI can help automate and standardise process flows and create financial sustainability by reducing operating costs. Furthermore, automating repetitive tasks helps employees focus on high-value activities. 

Industry example:

A large Australasian bank uses speech analytics on check-in calls related to loan repayment deferrals. The calls are passed through AI models trained to detect vulnerability, flagging interactions that needed specific manual review and helping identify the customers that most needed additional support. 

The examples above have all used validated AI tools to recognise patterns, create rules or improve communications channels. However, the use of AI for prediction and forecasting poses additional risk, especially when these predictions are based on personal data. This is because models are built from a chosen subset of available data and are not only prone to bias but assume that patterns from the past will continue into the future.

For example, credit scoring models have famously had issues with predictability. An example is the models incorrectly accepting or rejecting mortgage approvals. This occurs not only because they are prone to bias but also because low socioeconomic and minority groups tend to have less available data in their credit history.

It’s fair to say that AI is successfully being used in several areas of the financial services industry but there are some areas where AI will not be as impactful, or be more risky to implement. Banking and financial services organisations should therefore be smart about choosing appropriate use cases and technologies to generate value.

A good use case should outline what ‘measurable’ success looks like both in the short and long-term.  Use cases that are small and focused allow for a ‘fail-fast’ scenario, giving the team time and flexibility to improve and iterate.

Once use cases have been generated, they need to be assessed and prioritised based on the level of business impact and technological feasibility. These use cases can then be categorised as one of the following:

No Brainer -  high feasibility and business impact and drive the highest business value.

Enhancer - low business impact, but high feasibility. These tend to drive efficiency and cost-reduction and can be a good place to start as they can be easily implemented with minimal disruption.

Transformer - we call these ‘breakthrough’ ideas; they drive high business value but require more effort and may face more resistance.

Marginal - these drive low business impact and have low feasibility and should therefore be deprioritised.

A common misconception is that AI use cases require bleeding-edge technologies to generate value. The financial services sector has many opportunities to use proven AI technologies that have been in use for several years. For example, classification type problems are commonly seen in banking. Classification algorithms learn from a categorised training dataset to then classify new data into these categories. The following transformative case study is an example of using AI to classify high-risk calls for further review.

A leading Australian bank wanted to ensure it was meeting its compliance obligations in relation to secured lending phone calls with its customers. The bank faced a challenge with checking compliance across thousands of calls with their customers and needed to confirm these conversations were consistent, appropriate and compliant with regulatory requirements around the customer onboarding process. Compliance had many facets, such as providing required disclosures, customer authentication checks as well as questions around income and expenses.

The bank was also seeking to understand their customers’ sentiments, including vulnerability and confusion, and ensure their customers were receiving the necessary support.

By applying AI techniques to automate compliance on the bank’s secured lending calls, they were able to achieve high impact outcomes. This included:

• Full review coverage of 100% of customer interactions, which would not be operationally feasible using a traditional review approach
• Directing review teams to high-risk call segments and interactions and focusing human review efforts where it matters most
• 85% reduction in test points for manual review
• Insight into the quality of customer interactions, enabling coaching of call centre and customer care agents
• Better customer experience by providing insight to multiple teams across the organisation

A further consideration was the responsible and ethical implementation of the technology and guarding against unfair bias. By analysing the quantitative fairness of key models and training data as relates to protected features (e.g. gender of call centre agent), it was ensured that no key fairness risks were being left unchecked in the implementation. The initial application of AI techniques to this area of the bank was considered a Transformer use case, but given footwork performed in a High Impact area of the bank, roll-outs of the technology to other areas would be considered No Brainer.

With so many opportunities to apply AI in the banking sector, it’s important to prioritise potential applications by the business value they would generate. It is also important to consider the risks, to ensure that customer experience isn’t disrupted by unforeseen events. 

Implementation of AI technology requires a full roll-out cycle and experts at every stage, there needs to be clear communication on how the new technology will be used and why. It is useful to break the high-level plan for full roll-out down into smaller phases that deliver benefits early - starting with lower complexity and building from there.

Finally, be clear on how the new technology will be used and why you are doing the work in the first place. By having the customer and service delivery at the core, it will help create a foundation for implementation of new technology and help build a use case for why it should be implemented.