There are still opportunities in the finance market for someone with an innovative new credit product. But launching and, more crucially, scaling such a product is not without its challenges. In this Q&A, we explain some of the issues a new credit offering will have to surmount – and highlight a potential solution.
I’d like to launch an innovative new credit product
Good idea. You’ve spotted a gap in the market.
While the financial world has seen significant innovation in areas like payments, investments and onboarding, the credit arena is still niche in comparison and there is room in the market for fresh, new, players.
That’s good news. But what’s the catch?
We prefer to call it a challenge. You might think that the greatest challenge when trying to innovate in the credit market is regulation.
But it’s not. The key to developing a successful product and underwriting credit is gathering and interpreting the right data about your customers.
Data?
Yes. Data. You will need to gather and interpret data about your customers and then use that data to train your underwriting algorithms and scale up.
Ultimately, the data you gather and interpret should allow you to answer one crucial, specific question about each potential customer: What are the odds that this specific borrower will repay my innovative loan on time?
The correct evaluation of data helps a competitive lender to find the right balance when making underwriting decisions about its potential customers.
But here’s the conundrum. You cannot scale up your lending portfolio before you have a proven underwriting algorithm and you cannot develop a proven algorithm before you scale up and collect enough performance data.
OK. Where can I get such data?
Unfortunately, it’s not that simple. There isn’t one accepted solution for gathering and interpreting data about your customers.
In fact, there are several approaches. Unfortunately, many of them have flaws which can make them either inaccurate or too slow.
Oh. Talk me through the different approaches
One way to start is without real-world data. Instead, you build your product on generic information which describes a typical borrower from different angles – how they behave accessing other types of loans, how they manage their bank account, why they need the loan and what income source they’ll use to repay it.
But here’s the drawback with this approach. While these data points add value, they don’t answer the basic question about whether a specific borrower will repay your innovative loan product on time.
Right. So what more do I need?
You need to collect actual “performance data” based on your specific customer’s patterns of repayment. You do this through your marketing and underwriting process.
You then experiment with a small segment, acquire performance data, train your underwriting algorithms and scale up.
OK. How does that work?
Most importantly, you need to ensure that your sample borrowers are representative of all your potential customers.
Then you need to ensure that your sample borrowers are not skewed by conservative lending decisions on your part or customers who are over-eager for the product.
You also need to allow for how your borrowers will react to financial changes like higher interest rates, inflation or lower growth.
Unfortunately, the problem with this approach is that it can be an expensive and lengthy process.
Wouldn’t it be quicker to hire an experienced underwriter to shorten the learning curve?
That sounds like a great idea. But this approach is also not without its weaknesses.
The underwriter’s expertise was probably acquired working for someone else selling a less innovative product. You’re trying to be different, perhaps focusing on a niche segment or looking for a different sales channel.
The expert’s insight is therefore likely to be less effective for your new product.
OK. What if I hire a top-notch data science team to develop a fancy algorithm that can make optimal decisions to beat the market?
You could do that. You could hire a first-rate data science team, experienced in all types of machine-learning algorithms, even one with extensive experience applying neural networks to real-life cases.
But there also could be a potential shortfall in this approach. It is likely that your credit portfolio will not grow quickly. Or, that if you are conservative in your lending, your data will lack defaults. And defaults are the markers which your supervised learning algorithms will use to improve their predictions.
A conventional data scientist will probably lose interest in such a “low default portfolio” and will move on.
So if all these approaches are potentially flawed, what is the answer to getting the right mix of data and performance?
The answer lies in a holistic rather than a singular approach – a confluence of human expertise and machine learning and a combination of the collection and interpretation of customer and performance data.
The process needs to be fast and flexible. It should allow your product to operate from day one. It should collect and use your data to develop and finesse your product. And it should allow for human intervention.
That sounds perfect. And possibly too good to be true. Is there such a solution?
Yes, there is. In fact, that’s what credit analytics solutions like Paretix do.
Paretix, for example, offers exactly the holistic model described above. Its bespoke models leverage the lender’s domain expertise and uniqueness to develop their own scorecard. It gives a business expert oversight on the algorithms with the option to challenge them. It is collaborative and allows different teams to work together on developing the model and credit policy. It can be implemented in weeks rather than months and iterations can happen in days rather than weeks.
Allowing me to concentrate on developing and promoting my product?
Exactly. Design your product, set your course and establish your lending policy. Then let Paretix do the rest.
