Artificial intelligence or AI is transforming the financial industry in multiple ways. Given its inherent nature of risk assessment and mitigation, finance is perhaps the best playground for AI to stretch its legs and show off its capabilities.
While we're quite aware of AI's presence in the quantitative trading and investing space, how is it affecting the way firms make credit analysis decisions? Let's take a look!
Machine Learning and Credit Scores
Given the propensity younger consumers have towards using digital forms of money, instead of cash, it seems logical that a machine would be far more capable of parsing through the mountains of data that such transactions will generate.
Currently, credit decisions are performed with the aid of the evergreen credit score. However, reducing a person's borrowing history to a single number is a compromise and the original premise of developing the credit score was to reduce the amount of data into a manageable decision for a human being.
Given a machine's ability to process huge amounts of data the very need for a credit score seems limited by the number of humans carrying out credit decisions. A particularly ripe field for AI is analyzing prospects with little to no credit history. This is where companies like ZestFinance, via their ZAML platform, are looking to give lenders a leg up in the decision making process.
Indeed, ZestFinance claims auto lenders who used machine learning to inform underwriting decisions reduced their losses by 25%!
Developments in Underwriting Models
Another active player in the small business and consumer loan space is underwrite.ai . Underwrite sources portfolio level data of cured loans from credit bureaus and performs an initial classification based on certain data points like profitability or status.
The next step is to train their models to compete against one another in order to produce the most accurate prediction of profitability. New data which is then fed into the system comes out classified on various tiers depending on desirability.
An interesting point to note is that such underwriting analysis was usually carried out via linear regression based models. Despite the lesser accuracy, this method involved lesser time and cost.
This is changing, however, and as costs of computing reduce, expect to see more of these models being adopted by lenders.
With performance improvements of 25-50%, it certainly makes a whole lot of sense to do so!
Beyond Finance
Large financial institutions cannot be far behind when it comes to analyzing credit. Companies like DataRobot help institutions develop machine learning models aimed at blockchain, digital wealth management, credit card fraud and lending.
The firm's concentration, unlike the previous two, is not merely finance but exists across all sectors. Given the deep focus DataRobot has on machine learning this is logical. Despite this, the most exciting development is the firm's ability to develop models predicting fraud in payments.
This has major implications for both AI as well as spheres like cryptocurrency which does have the cloud of fraudulent transactions hanging over it. Building predictive models which identify such transactions will go a long way in legitimizing both fields.
When it comes to underwriting, the benefits are already palpable. Alternative lending firm, Crest Financial claims DataRobot's software enables it to make more accurate underwriting decisions and makes these decisions in a shorter time as compared to human intervention.
Challenges and Concerns
Despite the huge benefits, not everything is roses and sunshine when it comes to AI. Some of the major concerns revolve around privacy and the black box nature of machine learning. The former is a legislative and consumer-centric concern and the latter is a business concern.
Machine learning models have come a long way since the 1980's when developers had no idea what was happening and the models, once plugged in, simply took over and ran amok. These days, developers know exactly how the model will behave, given a certain set of circumstances.
There is always the possibility of a model buckling under a never before seen, black swan like, set of circumstances. Despite the marketing of companies in this field, the fact that black swan events are often exaggerated by machine learning models cannot be ignored. However, this does not mean AI is too risky.
These days, developers know exactly how the model will behave, given a certain set of circumstances.
That would be too simplistic. Perhaps the best way to view this would be to contrast it to human performance under similar circumstances and draw a parallel.
When it comes to questions of privacy, however, there are no such easy solutions. It is within the nature of a machine that its output quality will be in direct proportion to the quality of input. It also further follows that the a bigger data set has greater odds of being "better" quality than a lesser one.
Concerns over lending decisions being derived from social media data and other private events are legitimate and how lenders are currently using this data, if at all, remain murky. While firms are quick to stress the lack of quality of social media and personal data, as compared to loan level data, the matter cannot be dismissed easily.
It remains to be seem how legislation evolves to combat predatory actions. After all, governments have been completely caught out by the storm over Facebook and Google's handling of personal data. Perhaps, this is a great opportunity to be proactive instead.
Conclusion
All in all, AI is a step forward in human evolution and is unavoidable. While painting it as purely good or bad is simplistic, it remains to be seen how we react and preempt any issues which will arise over privacy concerns and the black box nature of algorithms.
For now though, the upheaval has gathered momentum and the next decade will surely see a massive disruption of a number of financial services.