Table of contents
Let's talk about partnerships.
They form the very basis of most fintech businesses. After all, they provide the plumbing through which lenders can make finance easy. It wouldn't be a stretch to say that lender-fintech or fintech-platform partnerships are what make these businesses viable. We at FinBox have written exhaustively about this on our blog.
Here’s a reading list:
The important thing is, for such relationships to function, each player must acknowledge the expertise brought to the table by the other. Even when the temptation to exercise full control over the assembly line arises.
Here's what I mean by this -- just like a fintech can't front the capital to give loans without going lengths to first acquire funds and establish a regulated subsidiary, a lender cannot immediately build the infrastructure that eases the delivery of financial services without transforming their legacy systems.
And that brings me to a very specific fintech product that lenders and partner-platforms both have, time and again, been tempted to build in-house - risk assessment models. It’s a seductive proposition but one fraught with complexities that are beyond comprehension of most laymen.
In theory, anyone can build their own risk assessment model. But to be truly successful, they must pay undivided attention to their data intelligence.
What goes into building an underwriting model?
Picture Day 1 at a fintech focused on building credit products. There is no prior data by which to judge the creditworthiness of a customer. Building a robust formal credit score, as it were, is out of the question. They start small – by creating a product that merely does a sweep of device data and supplies these insights to lenders.
The pilot goes well. Now, business starts trickling in. Lenders want more and more insights or parameters that can help underwrite the customer. Ultimately, the fintech takes out a weighted average of all these parameters to build a credit score.
Next, customers begin sharing feedback data. Based on various portfolios, the fintech can build a statistical underwriting model incorporating the performance of all the parameters shipped to customers serving varied customer demographics.
Here’s a more precise, albeit back-of-the-envelope, timeline for creating a risk assessment model -
The job doesn’t end with shipping the model. In many ways, a risk assessment model is a living organism that grows stronger as it is fed more and varied performance data. So, once the model is created, it is time to nurture it – a process that could take years to perfect.
Create a roadmap for including various data sources into the model and evaluate its performance vis-a-vis peers. Identify shortcomings in performance across segments and take steps to rectify them.
Prepare the data to make it ready for modeling.
Perform cyclical extraction of performance data from clients and incorporate it into the model.
Data intelligence expertise for a modern underwriting model
Objectively, the above roadmap (which is resource- and time-intensive) isn’t necessarily hard to replicate for lenders or even platforms looking to foray into new-age risk assessment. However, having a strong data intelligence arm can give an edge. Here’s why –
Large lenders draw on multiple data sources like bureau scores, account history, transaction data, insights from network analysis, behavioral data in addition to insights delivered by their fintech partners. In the large scheme of things, individual fintechs provide smaller modules that can be plugged into lenders’ own risk assessment strategy.
In this context, the data digested by fintechs and expressed as credit scores functions as one of the pillars on which lenders assess credit risk. This underwriting function provided by data intelligence experts is indispensable because only they have the knowhow to process proprietary data collected, for instance, via their own in-device risk engine.
Wide portfolio exposure
Being platform-facing, fintechs function as aggregators of risk data from various portfolios. For instance, they are exposed to varying customer cohorts depending on the target market of their partner-platforms. They act as go-betweens when they compile and retrospectively test feedback data from these diverse portfolios for lenders.
Because they test their models across a large and variegated sample size, the credit scores built tend to be stable in the face of unforeseen macroeconomic disruptions. This is a differentiating factor for players with a strong focus on data intelligence because individual lenders tend to serve a narrow borrower cohort, resulting in myopic credit scoring that often fails to withstand economic jolts.
Not unlike credit bureau insights, fintechs give a button-rung-of-the-ladder boost to new entrants who are starting out to build their own risk assessment models. Moreover, their prior experience with lenders or platforms with a similar make-up can help newcomers get a sense of strategy and portfolio evolution without experimenting and losing money.
Robust data science capabilities
Traditional underwriting models have a strong foothold over rules-based strategy for lending. They are adept at credit decisioning based on certain predefined business rules. However, as data sets continuously expand, it may become necessary to pivot to machine learning capabilities in order to better analyze portfolios.
That’s where focused data science teams excel. They already have the expertise since their core product revolves around it. For instance, they periodically revise their models, perpetuating a cycle of challenger-turned-champion credit scores as the model grows robust.
Furthermore, data science units in a fintech are more experimental when it comes to modern and incrementally efficient ML techniques like deep learning, neural network and model ensembling. They are supported by a hardy ML ops function to help them deploy these new-age models in production and scale them easily.
For larger lenders, on the other hand, a data science unit is a support function to their lending business. So, instead of building and hiring resources for periodic refresh of external data models in-house, they can outsource that part easily to the providers of the data. This would help to remove operational challenges of having to develop, deploy and maintain multiple models .
By virtue of it being the core product, fintech data science teams keep pace with modern techniques,for exploring and extracting new information from raw, unstructured data. This becomes the key to generate new and improved features to be used in credit risk modeling.
At FinBox, we spend a lot of time on feature engineering and improvement of existing features. The process involves scouring big data and applying evolving algorithms to categorize it. Dedicated teams of data scientists extract unexplored characteristics and attributes to complete 360° profiling of customers, which in turn improves the overall predictive power of the credit risk model.
On the other hand, those who view machine learning as a non-core function have a hard time building a comprehensive set of parameters for credit underwriting.
The idea for decentralization of financial services and the wave of FinTech infrastructure is to build strong capabilities with vertical specializations. This means that one doesn’t need to reinvent the wheel every time they launch or build a new product.
Think of it this way - you don’t start a delivery business by building a Google Maps competitor. Instead, you take the existing infrastructure, intelligence and the reliability that comes with Google Maps APIs and build your product on top of it.
While the temptation of building cutting-edge infrastructure is irresistible, lenders and enterprises focused on creating large businesses through FinTech products will be well suited to first explore and exhaust their options of partnering with the domain specialists. Chances are, they have already built, scaled and perfected a solution that won’t cost an arm and a leg to simply integrate.