Risky Data
Is Credit Risk the ultimate data product?
I used to work for a short-term lender in London, where the UK business was a subsidiary of a larger US one. I was hired to lead the BI & Analytics team there. This was actually my first time leading a generic data team, having previously led a specialised one relating to payments pricing. I’ve written about this experience from a different perspective before, related to hiring and team management. I’m going to focus in this post on the comparison to a sister team at the company - Credit Risk.
When I joined the company, my team was in a bit of disarray. It had a poor reputation, with a team member having resigned recently. The team wasn’t seen as particularly valuable within the business. It struggled to deliver what was seen as the basics, which were mostly financial reporting requirements. It was the team’s mandate to support Product and Marketing as well as Finance, and it had struggled to do so.
Eventually, after a lot of turnover as well as rebuilding the entire team, we did manage to support those additional functions. We used an outsourced provider to deliver the core data engineering and platform part of what we needed (a company called Avora, who I later joined). This, with me and some data scientists, served as the hub and then I hired analysts to embed into Marketing and Product, as spokes.
In all honesty, this was a tumultuous 18 months! Yes, it was only that amount of time, but a lot happened. During the whole period, I always felt there was a need to prove our value as a team - this is actually a good thing. We managed to deliver some value for marketing through conversion propensity modelling and providing improved and timely analytics.
My team was there to act, in part, like a data equivalent of Finance. I did report to the finance director, which probably reinforced this. We were there in part to measure other teams’ performance and to help provide them with analytical and data science support. We mostly were not in the money-making flow of the company. Where we managed to do work that did enter that flow, like in Marketing, was where we had the most value.
We had a sister team in Credit Risk at the company - their value was never in doubt. Even though their skills and methods were very similar to my team’s, they had a completely different perception. They were focused and specialised on one specific activity at the company: deciding who to lend money to. There was nothing nebulous about what they were supposed to be doing, it was very clear. If their models could do a good job of deciding who to lend to, the company could make profit, because the rate of default was low enough to be covered by interest from repayments.
Credit Risk’s clear and defined role was right in the middle of the money-making workflow of the company. So much so that they got to decide what business the company did or didn’t do. In my mind, this is a very clear service, composed of data products like risk scores and lending decision systems.
The company wants to lend to consumers to profit from interest charged, but they don’t know who to lend to safely to avoid an unsustainable default rate. The company is able to make products and attract consumers, but must choose appropriate customers. This is something unique about lending - no shortage of demand for the product, which is money itself, but risk in making the sale. Credit Risk provides a service to the rest of the company to help them decide who to lend to, based on their risk appetite. They also educate their colleagues on how this works and continue to evaluate model efficacy and improve models.
These methods in Credit Risk are now so dominant, the previous methods (not using data and ML) are seen as antiquated. These previous human or paper-based methods may not even be legal in some jurisdictions any more. They are so far in the rear view mirror that no-one would even dream of going back to them. The value of data and data methods in executing in Credit Risk are completely undisputed.
The way data is being used in Credit Risk is now regulated by law as it’s so pervasive and can drive unintentionally poor outcomes for certain groups. This is data work so important to how an economy functions that it is legislated for by the top levels of government!
There is a multi-billion dollar industry that just collects data from lenders and sells it back to Credit Risk teams (Experian, Equifax etc). This industry is only a fraction of the value that Credit Risk generates as a provider of data products.
Could there be other industries where data methods become similarly specialised and valuable? Medical data and related insurance risk fit a similar kind of model, where you have a boolean decision of whether to insure or not/payout or not, instead of whether to lend or not. The same is true for insurance in general, especially for consumer insurance products. The data available for medical has traditionally been hard to come by, sparse and inconsistent in format - there are now huge companies like Datavant solving this problem.
Is this a model of how data should work in every industry? Probably not - not every industry has these crucial boolean decision points that happen repetitively. These decision points are prime candidates for optimisation with data, allowing data to step into the money-making flow. E-commerce, for example, doesn’t really have these boolean decision points. The decisions are much broader across many simultaneous activities. Should we carry these products? Should we invest in these marketing campaigns? All of the micro decisions affect outcomes of each other, making things less clear. Decisions to do new activities, like carrying new products or expanding into new product categories, often don’t have data to help support decision making.
This isn’t to say that there isn’t value or standardisation to be had in data work in these other industries, it’s just different and the data team steps out of the money-making flow and into an advisory role, like Finance or Legal. The way for data to step back in to the flow, in these industries, is to become the teams that are in the flow - often Product and Marketing. The data folks are then Product and Marketing people doing better Product and Marketing work using data methods.
