Lalit Mehta, Co-founder and CEO, Decimal Technologies

Default loan prediction has always been an extremely crucial step for lenders. The availability of large datasets and open data sources clubbed with the emergence of advanced technologies like artificial intelligence (AI) and machine learning (ML) have opened new avenues for risk assessment. Along with this, automation of the loan approval processes has created massive opportunities for micro, small and medium enterprises (MSMEs). In the traditional system, risk assessment is based on parameters like demographics, age, gender that may lead to biased decisions. On the other hand, AI and ML algorithms analyse historical data based on previous lending operations, debt, marital status, and the financial behaviour of applicants. This helps in accurately assessing the risks involved for the respective customers. According to experts, AI can reduce non-performing loans up to 53 per cent and boost lender’s revenue by 37 per cent. With AI and ML, companies will reduce risk and speed up the decision-making process from weeks to days.

When banks use imperfect data samples, they often include other judgmental, qualitative metrics, like an assessment of the management team’s competencies and effectiveness, a measure of the firm’s competitive position or an appreciation of the firm’s physical location (prime vs. non-prime areas). The utilisation of such qualitative metrics is  often beneficial – for instance, by supporting more nuanced risk assessments through the inclusion of hard-to-quantify information that would affect risk measurements. But collecting this sort of data takes a lot of time and money, as this work can’t be automated. Furthermore, although these qualitative metrics must ultimately undergo a quantitative fitting phase – to derive weights, for instance – they’re often based largely on intuition and consensus.

Eliminating bias and human intervention

Integration of AI and ML means the elimination of human intervention. Decisions made by humans are influenced by their biases which may end in loan fraud. AI-driven tools run the available data against a group of rules to work out the borrower’s acceptability. This also allows for faster credit risk assessment which results in lower risk for the financial institutions.

As more and more data becomes available online, hackers are finding new ways to realise access thereto. Lenders might find it challenging to always be completely sure of the identity of a borrower when communicating virtually. Here, the use of biometrics alongside two-step authentication can be useful. Voice identification or video-based identification will make sure that the borrower’s data is safe and guarded and can eliminate the probability of human error.

Credit application phase

During the loan application phase, AI and ML are often used to anticipate credit needs by analysing credit line usage and understanding historical data patterns. For instance, an agricultural business is likely to have seasonal credit needs; these needs can be modelled to understand typical versus atypical patterns. By understanding how a company’s recent financial behaviour deviates from past behaviour, banks can detect or create opportunities for expanding their business relationship with the customer – or get early insights into potential causes of concern. In both these scenarios, having early insights enables financial providers to take action with a relevant response – i.e., extending credit proactively or declining a loan.

Banks will also need to evolve their lending criteria as the impact of COVID-19 changes over time. For instance, they will have to frequently review their lending process, risk decision process, and the availability and quality of data used to assess creditworthiness to assess credit worthiness during a post-COVID world. With the right information sets and models, AI and ML could help banks rapidly identify which companies are more or less affected.

Data gathering

Traditional modelling techniques typically rely on two sets of information: financial statement information (often in the form of ratios for liquidity and coverage) and qualitative information. These are extremely valuable data points, but banks now have access to far more information to consider. These tools not only utilise data collected from traditional sources but also leverage alternate data such as a borrower’s social media activity, daily transactions, utility payments, employment history, education and more (It is very important to ensure that data collection is consent-based). The more data the tools have to analyse, the more accurate the insights are.

Conclusion

AI and ML have played a huge role in increasing the process efficiency and reducing the turnaround time for the entire loan assessment process leading to a better customer experience by reducing errors in churning loans and evaluating customers’ digital footprint, search history, and social media activities to assess their credit worthiness. Adopting AI and ML will help financial institutions to gain customers with higher lifetime value thereby grabbing a profitable portion of the market.