With the rise of fintech and digitals lending, the words ‘credit assessment’ are now used and heard almost daily, though not everyone fully understands their implication. What does it mean, you ask? Credit assessment is the process of determining how creditworthy (responsible/able to repay a loan) a person or business is. Though the meaning is often simplified so those who do not have a background in finance, risk, or analytics can grasp the concept easily, as done above, credit assessment is a thorough and intricate practice, often guided by numerous rules and policies.
Perhaps it is because of this intricacy that, these days, machine learning and artificial intelligence are often touted as replacements for human intelligence and decision-making. Indeed, it is easy to find numerous research papers and thought pieces claiming that credit assessment can potentially be completely automated both in India and in western markets.
Automated or not, there is no denying that a great deal of credit assessment consists of making decisions under uncertainty, a skill that few can acquire and that requires a lot of practice and trial and error to master, even for AI. Despite the advanced automations that the data science team at Protium is capable of, we still highly value the judgment of our credit underwriters, who we think are the best in the industry.
While the SME credit market has evolved and become more formalized over the last few years, there are still a lot of nuances to assessing the creditworthiness of small businesses in India. The sheer diversity in the types of businesses that are present in India and the significant cash component of their business transactions makes it very hard to assess them through just models or code. We strongly believe that good underwriters are central to the success of any company that wants to be a market leader in the SME lending sector.
For better or for worse, good underwriters are scarce, and they do their best when given adequate time to make decisions. As a result, one of the key objectives we have as data scientists is to leverage our underwriters’ intelligence by embedding it into algorithms and scaling these algorithms with the help of engineering. Explicitly taking these decisions algorithmically also allows us to evaluate the quality of these decisions and continuously improve these algorithms as more and more data flows in.
Among all our customers who seek loans, there might be a subjective bias to decide who is a good or bad borrower in terms of creditworthiness, which is why these decisions are best left to an experienced, and therefore objective, underwriter. In contrast, recognizing an unsuitable credit profile is usually pretty straightforward since such profiles carry tell-tale attributes. For example, if a customer has missed multiple loan payments in the past and has almost no money in his bank account, it is because they do not have the disposable income or experience to sustainably manage a line of credit and make timely repayments. As such, the decision to write off such a profile is practically pre-determined for an underwriter. Detection of such credit profiles is the kind of decision that can easily be algorithmized and automated using a simple rule engine. Such an automation would lead to these customers being algorithmically removed as soon as their applications are received. This, in turn, would increase productivity since our underwriters would now only have to decide between the good and the bad, saving time that would have otherwise been spent weeding out the ugly.
To make sure that underwriters don’t have to rely on memory to check all these parameters, as there can be quite a few, we create automated checklists that automatically flag parameters that are not in line with our policies. These checklists are continuously adapted based on simulation of past data and our credit policies. Although simple, these checklists are powerful tools for decision-making and reduce the chances of overlooking key parameters. A good checklist should be short enough that our underwriters don’t have to spend too much time
reviewing it per case, but not so short that it leads to oversights. Checklists like this ease the cognitive load on our underwriters so they can focus only on aspects of underwriting that are more subjective and need their domain expertise.
The debate of man vs machine is as old as the industrial revolution and just keeps morphing into new avatars as new technologies are developed and gain popularity. In its current avatar, the debate centres around the premise of AI taking over humans in complex tasks that were hitherto thought to be the under the sole purview of the human brain.
The data science team at Protium doesn’t like to be part of such debates simply because we have too much to do. We believe that the key to becoming the most successful SME lender in India is to augment the intelligence of our best underwriters and enable them to make better decisions at scale so that’s exactly what we do. Does that mean we use no automation or artificial intelligence? As we’ve stated above, not really, since we believe that technology can be and is a facilitator for complex decision making; we just don’t believe that it is yet a replacement for human reasoning.
By Dhruv Nigam