How banks can scale AI adoption successfully

COVID-19 accelerated digital transformation for banks as both front and back-end services were forced to go online so that operations, run by employees from their homes, could continue. I think this has been a phenomenal achievement, particularly considering how some functions like mortgage refinancing with large and manual back-office processes are now being digitalized at speed.

On the downside, however, moving to online platforms increases exposure to security breaches and fraud attacks. Many banks want security solutions that use the predictive capabilities of artificial intelligence (AI) and machine learning so they can embrace digital while protecting their customers and the organization from risk. Having witnessed the possibilities of AI first-hand, I feel that an AI-first approach can help financial service organizations get ahead of disruption. But such an approach must be based on sound knowledge of the current state, clear view of the future state, properly devised roadmap, and a robust plan for continuous improvement.

Risks and challenges in AI

As a new technology, AI comes with some challenges. One of the leading arguments for AI is that it simplifies decision-making. Paradoxically, one of its biggest challenges is the introduction of bias within decision-making. This occurs because machine learning algorithms learn patterns and make recommendations based on training data. If the training data is too uniform, it can lead to skewed results whereby the algorithm treats any data point that is different from its training data as an ‘anomaly’. For example, a machine learning algorithm used in credit decisioning may (wrongly) reject a loan application from a prospect whose profile does not match that of the bank’s existing customers.

One way of addressing this is to expand the training datasets in terms of size as well as diversity and remove any outliers beyond the parameters. While leveraging such customer data, financial institutions must also be careful to adhere to guidelines and respect customer privacy.

Another challenge for AI models is the need to continuously improve or recalibrate modelling techniques in response to changing data patterns like customer behavior and preferences. AI platforms can be built to ingest data only for permissible purposes, augment decision-making and leverage minimal human intervention for monitoring and improvement.

Lastly, the success of AI models depends on the availability of high-quality engineering talent to train the models and ensure that these continue to align with business goals.

The road to AI adoption

Here is what I recommend for financial service organizations looking to facilitate AI adoption at scale:

●      Adopt an ‘AI first’ approach, especially for front-end operations. Agents can leverage deep learning insights, chatbots and smart assistants to eliminate friction from customer experience, personalize engagement, and offer proactive recommendations. 

●      Focus on data engineering. The quality of data determines the quality of decisions. Thus, a vast data lake is needed to consolidate good quality data. Metadata, too, should be captured to give the data meaning, while indexing capability is needed for rapid search. The most impactful data attributes should be identified and fed into the model, continuously.

●      Develop AI libraries. These will house common features that can be shared across use cases. For example, banks can develop computer vision algorithms to ‘read’ identity documents and store these in a library that can be accessed by any application requiring KYC compliance.

●      Switch to a cloud-based high compute environment. A 2020 research report shows that the finance cloud market, valued at US $22.17 billion in 2019, will grow at a CAGR of nearly 24% to reach close to US $81 billion by 2025. Financial institutions are increasingly working with hyperscalers such as AWS and Microsoft Azure to access highly scalable computing resources needed to support big data analytics, machine learning operations, and more.

●      Nurture the right talent. Talent such as data engineers and data scientists are critical for the success of AI programs. Financial institutions should set up pan-enterprise systems and environments for continuous learning, knowledge sharing and collaboration on new modeling techniques.

●      Fine-tune the operating model. Where AI is primarily used for making decisions, there should be a way for humans to review these decisions, when needed. All details, decisions, lineage of the data attributes, and the model used must be tracked to ensure that AI is auditable for regulatory compliance.

 How AI is reshaping banking

Banks and fintechs are using AI and machine learning in many sophisticated ways from customer service to fraud protection and rethinking the underwriting process through advanced decision making to facilitate responsible lending.

Regularly winning the World’s Best Digital Bank, DBS Bank is leading and driving innovation in this space. Using data-driven algorithmic credit underwriting models, they are able to approve small loans to individuals through their DBS PayLah! Mobile app. Relying on automation, DBS are able to fulfil the large market of underbanked individuals in countries like India and Indonesia; preventing themselves from being disrupted by other fintech in this area.

Democratizing opportunities for their customers to build wealth, DBS also uses AI to create a personalized experience and cross-market products to its customers. In what was previously only available to high-net-worth customers in private banking, DBS uses hyper-personalization through machine learning to recommend banking products to customers to meet their financial need. 

Chase is leading the way by investing heavily in a broad number of emerging technologies. The bank invests $12 billion per year on technology and employs 50,000 technologists. By leveraging AI and machine learning not only to help prevent fraud, increase credit lines, and tailor their marketing, but they can also anticipate the future technology needs of their consumers.

In the middle-office, AI is helping banks to prevent fraud, improve processes for anti-money laundering (AML) and perform know-your-customer (KYC) regulatory checks playing a key role in improving the security of online finance. In this space, fintech companies like DarkTrace and ShapeShift in the US are using AI technology to detect suspicious activity and threats targeting banks before they can cause damage for the world’s largest financial institutions.

 Banking on AI for resilience

 AI-led challenger models may soon replace existing operating models. But first, banks need a robust framework to deploy, validate, test, and refine a model before it is put to work on production data. This framework should include:

●      A culture, characterized by rich knowledge that can be transferred throughout the organization to develop a strong talent pool

●      A data ecosystem marked by strong data quality, stewardship, and governance

●      A platform ecosystem that enables a domain-based approach to AI-driven decision-making and provides high compute at scale

This approach with the right framework and technology partner can enable banks to successfully leverage AI and stay ahead of disruption.