What is a world-class risk team and how do you build one?
I’m often asked that question as Chief Risk Officer of BlueVine. Like other Fintech lenders, we are intensely focused on risk. Building a world-class risk team has been and will always be a key to our success.
Two elements are important in creating a robust risk organization: technology and your team. Naturally, having state of the art infrastructure and data storage and management systems is critical. But it’s not just about having the best hardware and software systems.
Your data science team plays an important role, their analytical abilities, their ability to work together, to identify the non-trivial patterns and to build a dynamic risk analysis system — often these are even more critical than the technology stack your company has in place.
Of course, there’s another important factor: your data. It’s impossible to overstate one point: for a risk team to succeed, you need data that’s clean, coherent, meticulously tagged and maintained.
Your data is your “source of truth.” Whether you analyze performance, train new models or simply monitor the portfolio, your data is your most important consideration. If your data is incomplete, you will come up with incomplete or inaccurate conclusions. If your data is full of errors, your analysis and predictions will be incorrect.
Your data can give you a clear picture of past decisions and offer valuable insights into the path forward. But your data can mislead you if not treated properly. So this point is important: You must be able to trust your data.
Good, trustworthy data sets must be able to answer two questions easily and accurately:
- What happened at a given point in time?
- What is the agreed upon all outcome?
I recently compared what a risk organization is trying to accomplish to the self-driving smart-car. The more the car drives, the more data it accumulates. It’s able to figure out the best routes, the best reactions, the most effective tactics in dealing with a growing range of driving conditions, whether it’s bad weather or potholes on the road. In other words, the more driving it does, the smarter the so-called smart car becomes.
That’s also the way our risk infrastructure strives to operate. The more financing transactions we process, the more data and insights we’re able to accumulate, and the faster and more precise our underwriting becomes.
However, there is one a significant difference between autonomous cars and credit-risk. With smart-cars, you can accumulate and process driving data in order to improve your algorithm faster and more consistently. That’s not the case in lending. It takes time for loans to mature, to detect cases of fraud and even to determine the health of a business’ finances. You may be able to develop an algorithm that enables a smart-car to react effectively to a rainstorm. That’s harder to do for small businesses that could end up facing serious changes in the market and the economy which could impact their ability to repay their loans.
Also, with a smart-car, if the system malfunctions, a person can step in to take control. The feedback can be immediate. With business financing, you may not realize immediately that something is wrong, that a client’s financial situation has deteriorated to the point where the loan cannot be paid.
To be sure, there will always be losses and defaults. As our CEO and founder tell us, losses from bad loans are a fact of life for lenders and financing companies. We will always encounter borrowers who cannot, or will not, repay their loans. Our goal is to keep the number of bad loans to a minimum. We hope to accomplish this through an increasingly precise analysis of customer data.
So how do we do this?
There are several key principles that we always keep in mind. The first one is to put a heavy emphasis on the data we have at the time a specific action or decision was made.
Let’s take the example of a new client. You make a decision on whether to offer that client a loan or a line of credit based on the information the client provided. This includes bank statements and other financial records, a personal credit history and outstanding financial obligations.
Let’s say we decide based on the information provided to offer a loan or a line of credit to a business owner. We begin sending funds and start collecting financial data. But after a few months, the client starts to miss regular payments. Checking the client’s bank records we find one cause: weakening cash flow.
Is there a way for us to predict that change before the client started missing payments?
That’s our goal: To make fairly accurate projections of a business owner’s financial health.
In moving toward that goal, we face a major challenge: to make projections based on data we have at the time a client applies for funding. This would include bank statements a borrower provides, as well as other information from other sources, including public records and even social networks.
One thing we need to guard against is creating computer models and algorithms with data that’s not available and cannot be available at the time of decision-making.
Making sure that your data is stored in a way that easily allows point-in-time data extractions at scale is critical. So is clear and accurate tagging of your data and classifying customers and potential customers in ways that are consistent with your business goals.
For a lender, these goals are typically defined by the sales team and the risk team.
Let’s go back to the example of the client who started missing payments and whose bank records eventually showed weakening cash flow. Now, the business owner may have defaulted after paying back 95% of the original loan plus interest.
In that case, the borrower may be considered acceptable or “margin positive” by the finance team, even as the risk team lean toward classifying the client as “negative.”
Deciding which definition to use is a business decision. It is important for both risk and finance teams to be aligned. When you optimize, you need to optimize as an organization. It’s important for risk, marketing and other teams to share the same goals and to agree on how to classify clients and would-be clients.
This process involves data and having a solid and sophisticated system for collecting, processing and analyzing information. But it also involves teamwork and human input. This is particularly true for young and growing fintech companies.
Ido Lustig is Chief Risk Officer at BlueVine, a cloud-based platform for invoice factoring, revolutionizing the 4,000-year old financing system that allows businesses to receive cash advances on outstanding invoices. BlueVine also offers Flex Credit, business line of credit financing based on 6-month and 12-month payment terms. Based in Redwood City, Calif., BlueVine has raised $273 million in equity and debt funding to date. Lustig spent many years working at PayPal where he was most recently Group Manager – Risk.