What you need to know before implementing artificial intelligence & machine learning

As financial institutions look to uncover new growth opportunities, many are evaluating artificial intelligence (AI) and machine learning (ML) solutions to help implement appealing service offerings that previously were challenging to bring to the market.

Those who live in rural and remote areas can be an untapped market segment that is often bypassed by financial institutions. Other potential customers – teachers, waiters, laborers, musicians, artists and many others—may eschew traditional banking services for several reasons, such as a lack of credit history or a borderline credit score. The arcane credit scoring system is not adequate for many customer segments.

The result is that there’s untapped opportunity among the unserved at the bottom of the financial pyramid and financial institutions may be leaving many dollars on the table by not serving these segments. Fear of penalties around existing banking products and lack of products suitable for these lifestyles are also factors.

Millennials are another customer segment growing in economic power that must be considered. They tend to favour banking services that cross social media platforms, email, voice, mobile apps – even hashtag-based banking services on their twitter handle – and want to choose what financial information they receive and how to receive it, within their preferred channel.  Studies also show that 44% of millennials don’t feel their bank understands them, yet 60% want their bank to be a partner/friend.

Typical core banking engines can’t easily handle today’s larger and complex data.

An accurate prediction of profitable financial products and services requires vast quantities of data, yet typical core banking engines cannot easily handle today’s increasingly larger and more complex data sets and still pivot quickly to roll out new services.

Many have come to realise that the ability of banking and financial technology, or fintech, to offer multi-channel offerings which can adapt to changing consumer preferences and environments depends upon the adoption of technology to extract insight from the vast quantity of today’s data.

AI and ML are the technologies that have the promise to make this happen. The good news is that the banking and financial services sector has been at the forefront of embracing these new technologies for multiple other use cases.

To prepare for the mainstream adoption of artificial intelligence and machine learning in financial services and to extract the full capabilities, here are some important aspects to keep in mind.

Plan for the impact of AI/ML in fintech

In the banking, financial services and insurance (BFSI) arena, vast quantities of data from a variety of sources must be analysed for digital banking and back-office operations, insurance underwriting, claims processing, wealth management and more. Process automation, combined with AI and ML customer service chatbots using natural language processing (NLP) and other AI/ML tools, has the potential to advance business intelligence and contribute to growth for the institution and the economy.

Since these AI/ML tools are what will comprise the future of fintech, let’s go over some important technologies and processes to know.

Robotic process automation

Back-office operations at banks are ideally suited to be handled by robotic process automation (RPA). Intelligent robots or bots can learn and execute rule-based business processes, potentially slashing time spent for tasks formerly performed by humans significantly.

For example, banks have started using RPAs for know your customer (KYC) checks which involves 150-1000+ FTEs for a single bank and are already seeing huge cost benefits.

Another area is trade finance, which includes processing huge amount of text documents from multiple parties and has high compliance requirements as well. Banks are looking at text analytics combined with RPA for automating such processes.

The more data machine language RPA tools see, the more they learn. Since RPA is rule-driven, it’s assumed that the data used by the bot is accurate, but this isn’t always the case. While a human may immediately pick up on obvious discrepancies in a form filled out by hand, a bot may not.  Therefore, data must be properly primed and validated before a bot consumes it. This involves text pre-processing steps such as lemmatisation, stemming, contextualisation, and designing conversation flows, in addition to regulation data cleaning steps.

According to Abhiram Modak, chief principal consultant at Persistent, ML solutions should be trained in the “three Cs of data visualisation: Correctness, context and category”.

With the right data, RPA tools work extremely well. According to a Deloitte report, many banks improved compliance (92%), improved quality/accuracy (90%), improved productivity (86%), and cost reduction (59%).

Natural language processing

A top priority for banks is providing high-quality service that makes the banking experience efficient, pleasant and free of errors. The capability for a customer to simply ask “what’s my account balance?” or “transfer money to Kickstarter” and access 24 x7 personalised service is very appealing.

24×7 personalised service is very appealing.

Customer service chatbots that employ natural language processing (NLP) and artificial intelligence to understand and interpret human language, both verbally and in text, is advancing. But can a chatbot understand the intricacies of human language, correctly interpret the intent and then take the correct action?

While ML can offer improved model and business results, the banks should be aware of the additional complexity as well as the security, testing and monitoring needed. A plan around monitoring, auditing, exception handling, fall back, compliance and governance is important.

A critical aspect in financial modelling is accounting for compliance and regulation by effective documentation, with an impetus on explainable algorithms, thorough security scanning and proper governance. This invariably calls for centralized data handling and modelling. In contrast, in a manufacturing modelling setup a decentralised data and model processing eases out the handling of the vast amount of log data from sensors and processing, selectively parsing it to the main distribution system using edge devices.

Implementing AI/ML in fintech

For a recent graduate with limited credit history and resources, securing a car loan or even a phone contract can be problematic. Those in the middle/low end of the income spectrum may face difficulties obtaining credit at a fair rate. Often these individuals turn to a credit union for a loan to help build credit and make easy repayments. This underbanked sector is a small but important humanitarian market and the process to provide loans at reasonable rates is labour intensive.

Although credit unions play a large role in the personal unsecured loan market, these are membership organisations that can’t easily raise equity. Gojoko is a UK company that supports a network of credit unions and community banks with a fintech ecosystem.

ML will be a critical and valuable component in scenarios like this in assessing the risk profile of the customer segments in order to mitigate, spread out and be prepared with a backup plan.

The future of fintech

Fintech often works under the premise that there is no way of knowing what a popular banking feature tomorrow will look like. Therefore, financial companies must build for technology and a way of working that allows them to evolve with a changing world. Challenging and changing banking’s traditional model of high-volume transactional manual processes by automating and creating new data-driven business models employing artificial intelligence and machine learning can change the archetype of banking for the better.

From bringing underbanked and unbanked households into mainstream banking systems to launching new products to improving the lives of millions of people, AI and ML is already helping transform financial services operations into something that’s more future proof.

By Sameer Dixit, general manager data, analytics and AI/ML and Jaideep Vijay Dhok, general manager, BFSI of Persistent Systems