Artificial Intelligence is pervasive across the front, middle and back office functions of banks. Are there evolving nuances that can help this new-age phenomenon graduate to becoming more bankable?
Over 1.2 million banking and lending jobs are likely to be replaced by artificial intelligence (AI) by 2030, says an estimate by the American Banker. What is even more interesting is that this revolution is also scheduled to create many new technology-related jobs, hitherto non-existent. While the estimated number in itself is reflective of the changing profile of banking roles, what is more pertinent is the permeation of AI in new age banking.
In a very simplistic definition, Al is the technology that helps perform tasks such as speech recognition, visual perception and cognitive analytics, to arrive at inferences and decisions based on defined algorithms without the need for direct human involvement. When these algorithms get to learn from experience and get continuously upgraded, redefined and modified based on new data and patterns with self-automated optimisation techniques, and without human effort, we are talking about machine learning (ML).
From providing robo-advice and next product recommendations, to more complex algorithmic trading and natural language processed chat-buts and fraud detection, the impact of Al-enabled banking services has been consistently making inroads across the spectrum of the end-to-end banking domain. Let us explore a few use-cases that are relatively more common-place today:
Front Office — Enhanced Customer Experience
Natural language processing (NLP) is the single most direct impact that the customer has experienced, thanks to the advancement of Al. Having the computer to be activated, as in chat-bots or voice-bots, based on human voice or written instruction into intuitive responses, has transformed the customer experience at the first port of call, being 24x7 and literally at the customer's 'fingertips'.
Chat-bots and voice-bots, also known as 'digital assistants', help customers by answering common questions, responding to standard requests and dealing with repetitive tasks to deliver on-demand services faster, driven by Al. Banks have been able to build distinct identities into these virtual assistants, with names such as "Eno' of Capital One, Erica' of Bank of America, 'Lakshm i' of City Union Bank and eLuvol of RBS becoming household names with their respective customers.
Al-enabled analytics helps in predicting customer expectations, curating the next best offer, and improved advice on investments and products. Sentiment analysis and conversation pattern analytics driven by monitoring customer comments helps marketing to evaluate the response to campaigns and to focus the sales effort even more sharply.
Middle Office - Better risk perception and credit decisioning
Consider this: a 10% improvement in default rates and reducing 'false positives' can make a phenomenal impact to the overall bottom-line of the bank. Rating accuracy for credit decisioning has historically been built on transaction data and past credit records. What is interesting is how Al is now enhancing rating quality, by augmenting this analysis with semi-structured data sources such as usage of mobile phones, social media and consumption insights.
Fraud deduction and timely flagging of a suspicious transactions can save millions, which otherwise would not have been possible without reviewing hundreds of thousands of past transactions and correlation with behavioural variables. Al and ML now allow a bank to correlate with behavioural variables. Al and ML now allow a bank to correlate data and identify suspicious transactions that warrant higher focus and attention, so that resources are properly deployed on high risk and high value transactions.
Back office— Driving operating efficiency
While robotic process automation (RPA) does not fall strictly within the Al category, the benefits of Al remain in leveraging simplistic applications such as optical character recognition (OCR) and improving overall accuracy in documentation and areas of transcription.
More than 80% of back-office costs are driven by 20% of intensive manual activities, that have now come under review with the introduction of Al. Predictive early warning systems also help save millions in opportunity costs These are not just limited to customer transactions alone: Al for IT Operations (AlOps) also helps drive predictive maintenance for data centres and helps minimise significant operational risk.
Challenges in execution
The biggest challenge of the AI era is auditability. With thousands (if not millions) of permutations processed before the cause and effect is drawn, it is hard to reconstruct a decision in terms of the data sequence, which may be relatively opaque. This is further augmented by the limited Al talent available. This makes "explainability" a key industry challenge. The most effective approach to validating AI-enabled decision remains in correlating the outcome with an alternative approach- either manual or another tool.
In addition to compliance with data privacy and related norms, the volume of data also demands a high degree of processing power for building the efficacy of a successful Al model and an integrated approach. The issue is particularly acute where the operating model of the bank Is sliced; where disparate systems do not lend themselves to a consolidated view of the "single source of truth", leading to challenges with their dependability. When a bank is looking to drive an Al agenda, the implications are threefold:
- Building a partner eco-system: Unlike the core technology
engines, most banks depend and will increasingly look to third-party players with specialised Al-enabled applications to drive the larger agenda. Banks also need to collaborate with other banks and regulated bodies to help access related data. This has implications not only on the need for onboarding of new players, but also in building an eco-system dealing with multiple partners.
The open API framework announced by Hongkong Monetary Authority (I-IKMA), for example, has seen increased adoption by industry players. Another related industry development federated learning - a concept promoted by technology players to allow collaborative machine learning across organisations.
- Continuous improvement: Learning curves are becoming shorter and sharper, demanding development life cycles to become more agile. For example, institutions leverage Al to validate back-testing models and review market positions on a continuous basis, with additional data inputs and refinement to models. Having an agile team to drive the design/development and technology teams to drive iterative product enhancement to produce quick results becomes imperative to develop a continuous improvement life cycle for Al. Reskilling of teams to build internal AI skills is also a critical success factor to sustain the availability of talent, and augment a continuous improvement process.
- Governance framework: Arguably the most defining aspect to maximising value, banks need to invest management time and leadership bandwidth in building, managing and driving a strong governance framework across data, processes and applications. While initial approaches are peppered with caution and "discovery" of the benefits, more mature organisations look to have an evolved data culture, Al application life-cycle management and a dedicated organisation headed by a chief digital officer. More importantly, it is also critical to be In compliance with emerging regulations around Al - the publication in 2019 of European Commission Ethics Guidelines for Trustworthy Al being an interesting example. Note also the establishment in 2018 of The Council on the Responsible Use of Artificial Intelligence by Harvard Kennedy School's Belfer Center for Science and International Affairs and Bank of America.
So what does this all mean to the industry, people and the customer in specific? On a more positive note, it may mean that people across the industry would get to graduate from monotonous, repetitive and mundane activities to higher-value and more rewarding roles. And what could that mean to the customer? Besides a more inclusive and larger customer base, with Al-enabled micro-credit to the unbanked sector, this would also mean personalised seymerit-focused micro-products and services from a larger partner portfolio.
However, as has been the case with every innovation, the final proof of the pudding is in the eating: the implications of lower operating costs and improved efficiencies should translate into lower fees and reduced charges for the customer. That would be the true litmus test, for at the end of the day, the customer is king!
To read more such insights from our leaders, subscribe to Cedar FinTech Monthly View