The Indian banking landscape is seeing a massive transition with the advent of financial inclusion through RBI. As the government shifts focus toward cashless society (evident from government demonetisation and digital India schemes), it also pushes a bouquet of digital payment options in the form of schemes, apps and services like small savings accounts, Agency banking channel, Aadhaar number, Pradhan Mantri Jan Dhan Yojana, BHIM, AEPS payments and UPI. Though the journey has just begun it may be wise for the banks and regulators to think of ways to take automation to MSMEs and the rural hinterland. Technology, a decade and a half ago changed the landscape of banking in India and is again set to revolutionize the entire process of financial inclusion and this time AI will augment the idea. 

Banking the Unbanked 

In India, a large fraction of 1.3 billion people is still not able to access basic banking services. For instance, in 2017, the Global Findex data showed how 5% of Indians accessed a financial institution account from their phone or the Internet, and only 2% of the population owned a mobile money account. Comparing this to sub-Saharan Africa, where 21% of adults had a mobile money account in 2017 and a 50% increase since 2014, one can see how India lags even compared to other developing regions across the globe. Similarly, for Digital payments it was observed how 97% of adults in Kenya making a digital payment in 2017 and 60% in South Africa, compared to 29% in India. 

There exists a huge gap which can be filled by allowing a fresh influx of technological backed banking scheme. However, with rural India seeing a growth of 15 per cent in mobile internet in 2017 with mobile internet users’ nearly touching 187 million, the prospective of mobile banking will be foolish to deny. Corroborate this with the argument that providing the right information to consumers (both financial and non-financial) is likely to increase consumer satisfaction. An example of a bank that is offering value-added services is ICICI Bank in India which is exploring social media such as Facebook wherein a user can link their debit card to their Facebook profile and use this social platform to recharge their pre-paid mobile phone, transact (borrow and lend money) with their acquittances & friends. On the same lines, in Turkey, DenizBank is offering various banking services through Facebook where users can link their Facebook account with their bank account to execute money transfers and use their credit card for various other activities online. Similarly, services like FinChatBot and Teller allows fintech to communicate automatically through message and present personalized advice to their customers. 

Enter AI and Machine learning 

In rural India where there is hardly any credit history for prospects for small loans, AI can create credit score/credit-worthiness scores using data from various centrally governed social security numbers, farming turnover, affordability (mobile being used, mobile bills, recharge frequency), social network (social media, cell phone call logs), travel information (GPS data, google timeline) and other such features using predictive modelling and Machine Learning algorithms.  

ML algorithm can eventually build credit profiles for those who were never exposed to the banking system and remain excluded from ‘financial exclusion’. Loan Frame uses ML to access the credibility of their customers. In another case, companies such as Monsoon Credit Tech use AI to determine the credibility of MSMEs. Using AI, many countries have started giving cash-flow based loans to MSME by learning patterns from various unstructured data sources including transactions, purchases, financial statements, tax statements and various other documents. These data help ML to predict the financial situation of the company and prescribe repayment methods. For instance, Kopo Kopo (by Grow) in Kenya which is exploring capital requirement of MSME is automatically decided and kept apart for repayments without creating a dent in the cash flow. Many other Fintechs around the world have confined every documentation for approval over a mobile app for loans by analysing streaming cash flows of MSME using data from various digital wallets. 

With the advent of PM Jan Dhan accounts which are linked to Aadhar numbers, ML can be used to predict and prescribe the right products to the customer. The transaction behaviour of these Jan Dhan accounts clubbed with techniques described above, ML can be used to find out the right products for people in remote areas. Clubbing AI with Blockchain technology can break the barriers to accessing financial services for rural areas. Blockchain technology clubbed with AI can be used to create digital DNA of customers who may not have relevant documents for availing banking facilities. Oradian along with Blockchain platform Stellar (by technologically integrating the Stellar platform into their core banking system) brought low-cost micro-payments in Nigeria. Oradian will allow around 300,000 citizens to transfer money between MFIs over the Stellar network without paying heavy charges thus creating huge tractions for such services. This network is available across 200 branches and is accessed by over 300,000 customers, mainly female customers across rural areas in the country. 

The Fintech and bank, hand in hand, could save millions by using AI w.r.t cost of acquiring new customer, reducing churns and lowering default rate - a win-win situation. The bank can further use AI techniques like (Deep) Reinforcement Learning to make AI smarter with every new transaction and feedback. Given the fact that AI and ML can use data from sources that were in past never imagined, sensitive banking decisions can be easily made. One example is data from GSTN being used by AI and ML to create a convincing case for financial support for MSMEs. Not only a credible case being presented, but the time taken to create such cases can be reduced manifold from months to few days, thus helping MSMEs to have quick access to financial support and reducing red-tapism. 

AI breaks language barriers 

AI can be of assistance to Fintechs and banks in creating better customer experience without recruiting a myriad of agents. AI can easily use NRLP (Natural Regional Language processing) to access rural areas (where communications are largely in regional languages/dialects) from a toll-free mobile number at any given hour of the day. This would not only enhance customer experience but can be used to even communicate in local dialects. It will save travel cost and time for people residing in rural areas and are looking for various banking services. Moreover, using voice fingerprint techniques for speaker recognition, the prospective customer can even subscribe to various banking facilities. 

Authorize the Voice  

In one example, the patented Phoneprinting technology by Pindrop’s innovative software can identify 147 different features of a human voice from one call to create an audio fingerprint of that caller and looks for unusual activity, potential fraud to trace unsolicited callers. It integrates with the company’s internal systems and identifies people's voices, locations, and devices. This is added to a database for future reference and help separate legitimate callers from scammers. 

Santander and HSBC both started voice banking technology on their mobile apps in collaboration with Nuance Communications which are intended as an additional layer of biometric security for the customer. By analysing over 100 factors such as speed, diction, accent and pronunciation, individuals can authorize themselves and make payments, report lost cards, set up account alerts and answer questions about spending. 

Recently, Bank of America launched a financial digital assistant called Erica while other banks like UBS, Credit Suisse and JPMorgan are using virtual-advisors (intelligent chatbots) that makes the use of cognitive and Machine learning to guide customers with financial planning and investments. One such breakthrough is Amelia that was able to manage 65% of the most common customer queries in under four minutes instead of the average 18 minutes in case of manual query resolution. Importantly, for the banking industry, Amelia can perform all of the key customer-related processes without ever wandering away from the rules & regulations. 'Luvo' is another online virtual assistant developed - using IBM's Watson AI system - which was recently rolled out by the Royal Bank of Scotland (RBS) and NatWest to interact with customers and address queries and perform simple banking activities via a chat tool powered by AI. 

Credit Rating without a history of banking 

Going deep into rural requirements, AI clubbed with computer vision and deep learning can utilize farmers’ geo-spatial data, pictures of their farms/fields (clicked from a cell phone camera), historical yield, historical trade records, weather and soil data, a technology used for farming, remote-sensing data to come up with right financial advice for farmers by taking into account risks and sales potential. These data can also be used by Fintech to optimize insurance and other financial benefits. Moreover, farmers who are illiterate and may not understand the fine prints of documents can be assisted with reading a contract using the text to speech feature and to fill their forms using speech to text algorithms. In another example, MSMEs are being reached for financial services by accessing the movement of products from shelves (using deep learning and computer vision) to estimate financial stability and repayment potential of customers. Services like FIBR assists Fintech to use AI (computer vision, predictive analytics and natural language processing) for providing financial services for MSME sectors where no previous banking traces can be found. Other examples would be Tala, that uses non-traditional data to predict credit score of those who are not covered by credit bureau like CIBIL by looking at data ranging from ‘social networks, mobile texts and call data, online transactions, app usage frequency and purpose and various other individuals attributes’. With the rollout of GST which had made it mandatory for tax filing to be online, these analyses can be generalised across industries, geographies, demographics or combinations. AI systems can further analyse tax data and build a quite accurate probabilistic model by merging these data with postal codes, cell phone records, travel pattern, social network etc. 

Pre-emptive measures for fraud 

Online payments fraud on the hindsight involves the use of credit card to make unauthorized purchases. For fraud dispute detection systems that involve machine learning, businesses must label the payments, explicitly indicating for each of their payments whether it is fraudulent or not or use Machine Learning techniques to set rules for labelling the transactions if not done by the system. 

• To warn the system (pre-empt) to be cautious of the next transaction as the transaction may be raised as a disputed transaction. The system would thus raise an alarm even before the transaction takes paces and place an extra authorization for the next transaction to happen.

• To help in decision making whether a transaction that has been as raised as a dispute is actually a dispute or a fraud (dispute despite being provided with evidence). 

• Assign a score to customers in order to find out those customers that have a high probability of raising dispute that may lead to fraud dispute (dispute despite being provided with evidence). 

Association rules such as ‘If disputed transactions are preceded by a common merchant’. These may seem non-important as these are undisputed, but rule mining would reveal that the next transaction would be raised as disputed thus revealing a pattern of sequence that may lead to fraudulent transactions. Analysis of existing fraudulent transactional data can help confirm the suspicion and provide a score. This score can be first used to create an alert and more importantly as a feature to train credit transaction classification models for dispute. 

The extensive feature set includes the transactional details, customer scores and a few engineered features like Shift from spending pattern, Shift from a general mode of transaction, Distance from Common transaction area, Common fraudulent previous merchants that have high probability frauds. 

Citing a few examples, banks like RBS Royal Bank of Scotland, Danske Bank, OCBC are using AI to predict fake transactions in real-time.ML can allow computers to learn a suspicious pattern and raise alerts and by automation of the triaging process for alerts. ML is quite fast in learning behaviour and finds out an anomalous pattern that could indicate various financial crimes. Few examples being abnormal transactions between borders, abnormal transaction values, abnormal transactions time, sequence and pattern of transactions. 

Banking Advancements across Globe 

Taking an example of TransferWise (an international money transfer portal) which lowered the cost of remittance down to around 1% thus allowing millions of people to transfer money across the world without heavy charges. Extending the thought, a huge remittance potential lies in rural areas as more than 100 million people migrate from one village to other for work and can be brought under the banking system for proper remittance. A study by National Remote Payments Survey by National Council of Applied Economic Research (in partnership with Nielsen’s forecasting technique) states that domestic remittance is around 1000 billion per year where rural India contribution is around 60-70% but only 30-40% of these remittances are done using a proper channel. 

Open banking - a.k.a open API is being implemented across the UK and EU - would allow a bank “to access native fintech solutions in a plug and play type of way”. In a simple case, Experian Connect API gives power to customers to view their credit score instantly without any hassle. Banks and customers can now not only use it for tracking their credit score but can also plan its course for improvement. The website states that “Consumer-empowered sharing allows you to create products and services for previously unreachable markets”. It further states that it allows “landlords, property owners, real estate agents and other small business professionals the ability to view a credit report online” thus enhancing the reach and scope of credit scoring. Similar transactional behaviour can allow an ML algorithm to learn more about spending pattern of an individual and thus predict better recommendations for products and credit score, to say the least. This also allows easy portability of bank accounts and banking services (the way mobile number portability works) as Open banking would allow various system and databases to talk to each other seamlessly and customer data won’t be the property of one bank. 

Paytm in India (during demonetization) was a lifesaver for MSME. AI and ML if clubbed with open API would allow fintech to offer various other banking services using data from the likes of Paytm. ML can peruse customer data like spending habits to advise financial planning. ML can enhance identity profiling in multiple ways as discussed in the sections above. AI will not only decrease transaction-fraud detection by taking various features other than customer details (such as geospatial data, keystrokes, pressure, speed, timings, monetary movement, device details etc) to verify the identity of a person in real-time thus validating real-time transfers. AI can use network analytics to understand and predict the upcoming fraudulent transaction by learning the pattern from fraudsters. AI can be used to read financial pattern by looking at mobile bills, SMSs, call records, mobile browsing history, frequency and distance travelled thus creating deeper user profiles. 

In an example from Africa, ML allows small rural workers to get their transaction approved without delays and get their money transferred to their account from as far as America almost instantly. Data collected through the IoT can aid fintech in making a better decision by using ML to dive deep into customers’ spending patterns, transaction pattern thus boosting rural banking 

The best attribute of Open Banking is eliminating multiple mediators and connecting banking services directly to customers thus reducing costs of money transfer. 

Conclusion 

As banks focus on efficiency, customer delight, experience, their use of innovative technology is crucial. Artificial intelligence (AI), or the use of computers to carry out the processing and decision-making tasks previously carried out by humans, is one form of innovative technology banks are adopting at a very high rate to achieve these objectives. The steady increase of AI in banking, however, will likely have both positive and negative impacts on the banking industry. Given the fact that India is accelerating towards digital mindset and given that STEM population in India stands at 2.5 lakhs per annum, there is no denial for future of banking revolution to ride on the back of AI. Visibly, as efficiency rises, AI will have a tangible and optimistic influence on banking.