Lending banks on trust.
And trust comes from knowing the borrowers better than themselves.
Traditionally, lending was performed more as an art with siloed channels, person centric processes and individual biases. This approach resulted in increasing charge offs and non performing assets (NPAs). In response, machine learning is now penetrating deeper into banking and financial markets for empowering lenders in taking (almost) instantenous and accurate lending decisions.
The debate on benefits
The increasing adoption of machine learning have divided the banking technology experts into two camps. The skeptic side argues that machine learning brings limited benefits to banks on the table. They place enormous faith in existing trove of individual credit data, risk assessment models and believe that maximum data optimization and extensive analytical enhancement have already been brought into the system. They see limited scope in extracting more efficiency in process automation.
The advocate camp, on the other hand, counter that there are still undiscovered characteristics of would be borrowers that can be profitably mined with the right machine learning models.
Advancing lending by leveraging machine learning
Smart algorithms already decide the price that an individual customer will pay for an air ticket, which advertisement he sees or the offers that he gets. Similary, lenders can quickly decide who gets what credit and on terms. They can substantially improve their accounts, ticket size and asset quality by…
1. Creating a whitelist of high quality applicants
Customers today demand instant fulfilment. They need to service their financial needs in as few clicks as possible. A lender can use machine learning algorithms to map customer details to data models and identify a whitelist of high quality prospects or applicants. Lending decisions to these individuals can be automated to save time, costs. As the whitelist carry minimal risk of defaults, they can even be proactively approached with credit offers.
2. Identify deserving customers with no credit history
Human experts have spent a lifetime trying and hoping to set a golden standard for identifying the credit worthiness of a borrower. The result was credit scores, based upon voluminous credit data. However, there are still significant sections of society that lack a comprehensive credit file. This leaves out applicants who need and deserves credit but are rejected due to a lack of traditional credit history.
Instead of giving primacy to a single number, machine learning can quickly bring in 10X or 100X variables, resulting in a more holitic borrower 360 degree intelligence. Lenders can make huge improvements in reducing write offs by using machine learning to map an applicant demographics, credit, payment history, employment history and industry etc. This approach boosts a lenders confidence in providing credit to underserved candidates with no prior credit files.
In 2008, when car loan defaults started rising across the industry in North America, one of the biggest auto lenders moved quickly to minimize the impact on its portfolio. However, as they specialized in underwriting borrowers using traditional credit assessment, the approach tightened lending and raised tresholds to the point where 70% of the applicant were turned down for a laon.
With the help of machine learning, the firm’s analyst were able to identify 2700 borrower characteristics, up from a mere 23 indicators. They quickly built and trained the new data model set. They also deployed intelligent robotic process automation for automating approval decisions.
The result? The firm’s lending volume doubled and new approvals increased by 45%. Furthermore, the new credit contracts were performing better than the ones previosuly issued. The auto lender was thus able to achieve it goal of approving more borrower without additional risk.
3. Automate risk based interest caculations, credit lines and approvals
We humans have become experts in making strategies and rules for creating the right data sets for a risk assessment model. However, given the massive number of insights to be analyzed in a credit assessment model, humans are physically and psycological incapable of always making the best decisions. There is always a chance that an important characteristic might be overlooked. Traditional credit scroes are also open to manipulations and biases.
Machine learning can help lenders to automate complex decisions like credit approvals, risk underwriting etc. through rule driven business process created by visual designers. Through robotic process automation (RPAs) powered by deep machine learning, can help create intelligent business rules, integrate with credit rating bodies and back office process for faster loan approval process. Machine learning can also help determine the optimum credit line that can be offered and the best risk (default) based interest to tap into deserving opportunites.
A leading auto finance provider in Asia used machine learning to automate loan approvals with straight through processing to cut down loan processing time from 30 days to 3 minutes.
Getting a headstart
Using Machine Learning in lending processes will drive improved efficiency, reduce costs and deliver a delightful customer experience through instant gratification. But for that to happen, there are certain basic steps to be taken, such as:
1. Analyze internal customer data
Lenders are already sitting through mountains of cusotmer data gathered through onboarding processes. This can be an individual’s or firms’ financial docuemtns, demogrpahic, occupation, administrative and lifestyle details. Smart machine learning alogorithms can then calculate default risk probability, best interest rate for a segment and develop a credit assessment, pricing product that is more nuanced and targeted.
2. Start with small samples
While increasing accuracy in machine learning algorightsms need increasing amount of data, it is advisable to start building data sets with smaller samples. Starting small will allow data scientisits to optimize data models for small samples. Gradual increase in data samples will help lenders to improve data analysis, structure conversions, dimension reduction, numeric calcultions and product engineering.
3. Clean, aggregate and retrain data
Data, especially those fetched from external sources, contains noise. These have weak attributes that can affect the outcome and structure of a data model.
The evolution of machine learning is directly proportional to the amount of training data available. Clean, aggregate and retrain your data sets on regular intervals, preferably every one or two weeks.
Model planning, building and implementation for machine learning is a multifaceted process. It links connections that are unobserved by humans. This empowers lenders to drive credit offerings, customer experience and business growth.