It’s getting crowded in fintech. That’s not a new revelation or anything, but it’s a trend that’s had a lot of serious consequences — especially as we start 2021 and look ahead to the next twelve months (or more). The fintech field — which covers a wide range of sectors and services — has exploded in recent years as online services become significantly more accessible and easier to build and deploy.
This has had some interesting consequences. For one, the whole sector is heavily data-driven by design, and it has been quicker to embrace analytics and data science than others. When you look at services like online lenders who can provide a quote in a few minutes, or online banks that can quickly approve customers, it makes sense why they would adopt data science and machine learning — it provides a great way to differentiate yourself from the competition. But what happens when everyone is at the same place, and the external conditions change? Or when you can no longer build better machine learning (ML) models to get ahead?
2020 was difficult for the sector because it invalidated a lot of historical data and it made it harder to successfully predict risk in many ways. On the other hand, it did show us that data is still the answer, and in a sense, the industry has reached an inflection point.
What happens when everyone else is using the same technology to build ML models? Coming into 2020, the market started seeing the results. Model optimization (Model Selection, Hyperparameter tuning, etc.), which has been the go-to for gaining an edge when it comes to achieving enhanced performance and better insights, has lost a lot of its impact. This method focuses not on changing the data but on how it’s being measured and processed. Generally, it can provide some performance improvements at first, but pretty quickly you would reach the point of diminishing returns. Moreover, these methods are becoming more and more commoditized, and nowadays even a relatively novice data scientist can train and fine-tune a model with just a few lines of code.
More importantly, last year showed us that when we’re too complacent and focus just on tweaking the models we already have, a shock to our datasets can throw everything out of balance. You can have the most optimized ML models, but if you’re not feeding them relevant data, it won’t matter how well you tuned them — the results will be irrelevant. You need both sides of the equation to remain successful.
So how to distinguish yourself from the competition — in terms of better service and user experience — when you can’t get that edge with model tuning? The fintech industry is at a crossroads when it comes to getting more out of their ML and data science potential. On the one hand, it’s a necessity to continue improving. On the other, the old ways offer increasingly diminishing margins.
If last year showed us the problem — the need to find a better way to improve our ML models’ predictive capabilities — 2021 will be all about the solution: alternative data. It’s easy to talk about the awesome things alternative data can do for organizations, but actually finding it is (or was, until recently) much harder. 2020 saw major improvements in the access to and democratization of data as new platforms and data providers made data — which has become essential — significantly more accessible.
This puts fintech and data science at an inflection point. The algorithmic part of data science is becoming less impactful, but better data in the form of alternative sources — which is much more effective — hasn’t been accessible until recently. However, greater access to the necessary data and much better tools to leverage it are quickly pushing fintech in a new direction. Most importantly, it will let organizations get enhanced performance, as well as better insights, and adapt to a rapidly changing landscape.
Think about a small business lender that needs to better understand and predict loan risk in their market. 2020 was an eye-opener, as their models likely had trouble accounting for sharp changes in the economic landscape and their borrowers’ financial stability. Their models couldn’t reconcile what they had seen from historic data with what they were experiencing in the present. The most recent data was so drastically different from their historic datasets that their models were failing — not because of their design, but because they were essentially blind.
The lender needed a new way to measure borrowers’ potential risk that could account for the difference in historic data. Let’s say the company is evaluating a small grocery store in a traditionally financially stable area. 2020 was a rough year for the region overall, and in-person sales — on which the store largely relies — dropped significantly for a good part of the year. The borrower, who had historically been considered a safe prospect, suddenly seems much less so, but it was still hard to tell.
The lender could use a wide variety of alternative data to determine whether they should extend a loan and at what rates. For example, they could use:
Instead of relying on data that is largely outdated at this point, the company could build a much better and comprehensive picture of their borrowers in minutes using the right data platform.
2021 is already shaping up to be a much different year. Most sectors have learned their lessons, and the economy is adapting to the rapidly changing circumstances. For fintech, this means the race is back on. Organizations will need ways to continue offering better services in a much riskier environment and roll with the punches that may emerge throughout the year. As their models continue to evolve, they’ll need to combine creative fine-tuning of their models with alternative data that can give them real insights and differentiate them from the rest of the pack.