Data Science has not left any industry untouched. Traditionally, the Banking and financial services domain was where heavy data driven decisions were made. But now, healthcare, retail, supply-chain, gaming – almost every field you can think of is basing their strategic decisions on data. As per salesforce, by 2020, 57% of business buyers will depend on companies to know what they need before they ask for anything.
While data points us in the right direction, machines overtime have been able to execute these decisions without/minimal manual interference. But let’s break down what a manual decision may look like in a simple case study of whether or not to approve a loan application:
So an analyst would spend on an average 30 minutes to evaluate an application and additionally 2 days through a vendor to finally approve the loan. We are now able to replicate the human intelligence onto machines; let us see how:
- TASK I -> Match loan application against a checklist of safety net rules:
This can be automated as soon as the applicant fills in his/her application online, such that it doesnot even enter the rule decisioning engine. These are absolute parameters that the system can do a one-one comparison of and make a quick decision. Example: Income greater than Rs. 6,00,000 and Cibil score > 750.
- TASK II & III -> Calculate probability of default on an excel sheet & reviewing the cut-off
Calculate the probability of default using algorithms such as logistic models or decision trees. In a manual mode, the analyst can copy past the variable values such as income, cibil score, etc on the excel sheet and the default probability is updated for the applicant basis the equation. In an automated mode, a rule decision engine is integrated with the loan application system such that the algorithm picks up the variable values the second that the customer fills in the application fields, therefore, calculating the probability of default instantly, comparing to the cut-off value and approving/rejecting the application.
- TASK IV -> Physically verify if the individual is who he says he is:
This can be automated with innovative solutions such as integrating the rule decisioning engine to the E-KYC database. The name, address & phone numbers can be matched with the database and verification OTP can be used to conclude the verification. Technology now also facilitates finger print verification online.
- TASK V -> Physically verify where the applicant works:
This can be done by crawling on social media platforms such as linked-in and developing some social media scores that ascertain the stability of an applicant’s job.
While the 1st 3 tasks have been automated not long after the IT boom, with deep learning techniques unstructured data can also be used to make decisions that help automate a process. Therefore, in the last few years most financial institutions have been able to transfer the verification tasks also to machines.
To summarize, by automating all our decision points, we have managed to approve a loan almost immediately. In addition, automating helps to remove human bias, human errors and the challenge of storing excel sheets and documents.
Does this mean that the rule decisioning engine takes away the job of the credit analyst?
No! The engine has taken away the ‘repeat’ tasks for the analyst such as looking at a certain income level, cibil score,etc. But it is the human intelligence of observing anomalies (which are not repeated) that helps upgrade the skills of an analyst. The analyst can now focus on developing the machines to make automated decisions, creating alerts and even enable manual interference where necessary. For Example: Even though the income safety net is at $2000, an analyst might notice that an application with rounded off income values such as $2001 are usually fraudulent and therefore, needs to be sent in for a manual review.
By allowing machines to learn from us, we are able to evolve further and focus on building more innovative solutions.