Author: Prateek Samantaray
Organizations are working relentlessly to make the customer experience at every touchpoint of a transaction as seamless as possible. And the conversation around AI has intrinsically correlated itself with intuitive, human-based customer experiences. The need for identifying and solving customer challenges intuitively and objectively while having access to all possible information pertaining to said needs is of utmost importance in the present economic climate.
Pankaj Kulshreshtha, CEO at Scienaptic AI believes that financial institutions today may be ahead in incorporating the latest technology but there is a gap in how this technology translates into customer experience. Their systems of records (Bureau, databases, core, alt data) need to connect with systems of engagement (CRM, LOS, digital apps) and all of this needs to be powered by systems of intelligence i.e. decisioning systems for every touch point powered by AI. “We hear of instances where an individual with $700,000 of petty cash sitting in his bank account is being offered a $500 credit line. These are instances where despite the seeming intelligence of financial systems, they don’t seem intuitive enough.”
So how do financial institutions solve this problem? The key here, Pankaj believes, is continuous adaptation. “You need to keep in mind that this is not the only time you create a process. You can’t create a model, pack it in a box, and keep it away. This is an adaptive model that needs to keep learning”, he adds.
Implementing continuously adapting models and looking at multiple data sources is a rhythm consistent among lenders of all sizes. Paul Quintero, CEO of Ascendus, a New York based small business lender sees the value of AI as a means to continually improve upon your capacity and your data. “Models can only give you predictions on the basis of data you give them to measure. That is usually historical data pertaining to the past five years or so. Unfortunately, the pandemic happened in the last five years. And a recession happened in the five to ten years prior to that. If you start relying on models built solely around this data, you’ll be stuck, and you’ll end up declining a lot more people. So the first dimension to using AI is to continually refresh the models because every model includes data, and data includes time and experiences that can be put to good use if all facets of information are taken into account.”
John Witterschein, VP, Consumer Credit at Bethpage Federal Credit Union, one of the largest credit unions in the USA, believes traditional scoring methods have many drawbacks and explainable AI powered credit decisions along with alternate data can be a game changer for near thin files.
“While implementing AI for our credit underwriting processes, we came across cases where we would have approved an application in normal circumstances but were declining them now, and vice versa. In essence, looking at the complete picture makes a lot more sense. Even when you don’t have data to make decisions based on payment and credit history, and you leverage alternative data to see that the applicant has address stability, mortgages, and occupational licenses, it makes sense to approve a loan”.
Paul believes that AI is especially important for smaller lenders in order to keep up with new economic conditions. “For a business of our kind, we know that we’d be declining a lot of people if we were just relying on scores alone. We want to refresh, build the capacity of our team to serve, and make sure we’re adapting continually to new economic waters. On top of this, we can use all databases and predictive tools we want, but in the end, it boils down to using a tool that is rational and relates to the type of things we were trying to educate and empower our clients about.”
Murli Buluswar, Head of Analytics, Consumer Credit at Citi, reasons that the focus for lenders in the next decade would be singular: gaining the ability to have a single, all-encompassing customer analytic record. “A 360-view of the customer that stitches together every interaction and transaction at a customer level and helps you understand their needs and their latent needs, and helps you think through more intelligent ways in which you could build relevance and engagement with your customers is important. While on the credit card side, we have a lot of data to go with in order to come up with a better picture of the customer, the challenge to this immense opportunity is to separate the signal from the noise. The key lies in taking advantage of every interaction and transaction, drawing meaning from that, and understanding what that implies in building more relevance and intimacy with our customers”.
This is where AI steps in.
A key reason to adopt AI in underwriting processes is what is called the proverbial “Need for Speed”. Companies have to keep up with the needs of the ‘impatient consumer’, and while speed is of the essence, so is the need to control risk, especially during current times, when a downturn is imminent. “The current timing is such that we have to help companies figure out the risk and growth trade-off specifically. You have to be smart enough to make sure that while you are considering riskier loans, you try to get some of the good loans to control the risk. This line of thinking requires you to have powerful underwriting tools that allow you to make such effective trade-offs", says Pankaj.
“As a risk manager, you can’t just apply an overall strategy and decide that since times are going to get tough, you will reduce everyone’s credit lines by say 80%. A strategy like that will end up making you lose good customers who could have given you good business. In this scenario, you need to look at customer cohorts at a more granular level and identify the cohorts that you want to create an impact for. And to do this level of analysis, you need sharper tools that use more data to come up with better scores and pinpoint which person you can still lend to.” says Paul.
But the apprehensions around AI are still persistent. The concerns regarding regulations remain, despite the advent of simplified, FCRA-compliant AI models that are no longer limited to being black boxes. In addition, a key concern among companies is the identification of problem statements that require AI intervention.
In the current scenario, with bureau data or credit scores being opaque and not enough to tell the whole story of a borrower, underwriting tools like Scienaptic have created an impact. Scienaptic creates an AI-powered credit decisioning engine for financial institutions by leveraging alternate data and AI models. This helps companies identify potentially good borrowers and potentially ‘bad’ borrowers on a much granular level.
John Witterschein vouches for AI in lending decisions “For existing members, we look at deposit data as part of our manual decision; if it comes to a manual decision. So we started thinking, why can't these decisions be automated, leveraging all of this data that will speak to a borrower's creditworthiness, right? If you got a good education, have a home, and have a bank account with good cash flow, those are things that are going to demonstrate that you're going to have the ability to pay. And those are data points we want to leverage. And that's why we went down the road to look at potential vendors out there. And we came across Scienaptic.
And our experience with Scienaptic has been tremendous. Three years of data from our application were used to train their AI models, and the results were impressive. They demonstrated how one could increase one's automated decision rates, more importantly, increase one's approval rates. What was particularly interesting to us was that we looked at some cases where we would have approved the application, which we were declining now. And we looked into some of the alternative data that suggests that maybe they aren't very purposeful, and what’s even better was that you could explain all of it.”
With data available in abundance, the reliance on credit scores as a sole indicator of creditworthiness seems archaic. It is an approach that relies on limited historical data and biased datasets and severely limits a lender’s potential to lend. In this scenario, it is necessary to reimagine the human element in lending processes, develop intuitive solutions beyond rule-based decision-making, and leverage a broader range of data that helps lenders know their customers better. This approach has a consensus among all modern lenders, be it large credit unions like Bethpage or small business lenders like Ascendus. Knowledge of a customer on a granular level, as Pankaj puts it, will help create a seamless customer experience and help tailor a solution that matches customer requirements. This seems all the more prudent in the current times when recessionary indications are pushing organizations towards more intelligent lending that focuses on low-risk growth.
At Scienaptic, we employ the experience of helping more than 70 small and large scale lenders reach better credit decisions with AI. We understand the perils and challenges of manual underwriting, and we’re on a mission to change that by helping lenders tap into pools of data hitherto unexplored. As a result, Scienaptic has been powering 7 million annual credit decisions leading to a 40% increase in approval rates and a 15% decrease in default. With Scienaptic, organizations worldwide are increasing their approval rates at low risk, decreasing defaults, and using intuitive AI powered intelligence to make personalized credit decisions for each borrower, fundamentally transforming the lending industry.
Murli Buluswar sums it up “ We should try not to get too bogged down by whether something is AI or something is different degrees of machine learning. You’ll need to know that the interaction between humans and machines in a process will vary immensely depending on the context. The challenge lies in identifying problem statements that matter on a scale to apply the power of analytics and AI to. It is necessary to make an assessment of areas where we feel we can drive non-linear value through the power of analytics. We have tried to identify five criteria to do so:
Is there any significant customer or regulatory exposure in that problem or opportunity?
Are there manual decisions being made on a scale?
As a consequence of those manual decisions, do we have partial coverage in solving that particular problem?
Is there any significant data exhaust that has been created by that process that hasn’t been tapped into?
What are the financial limitations to end-to-end fully automating that process and frankly taking the human out of the loop? Both this can reduce risk exposure and cost-saving.
Applying these five conditions helps us identify problem statements that matter on a scale, where analytics and AI can be applied, really”