Use cases on how Machine Learning for micro-segmentation & personalized targeting, optimal agent-customer pairing, early risk identification & quantification, etc.

We have heard it from many experts – data is the next big thing. Advancements in data analytics capabilities are changing all industries – especially the BFSI sector. Firms are increasingly relying on ‘data-driven’ decisions to evaluate underwriting risks, improving customer retention, streamlining operations, and ultimately improve their bottom line. Although firms are actively investing in using machine learning tools in such domains, there is one overlooked area that has the huge potential of application of the data science – the collection mechanism.

Traditionally, lenders have been using simple analytics to understand who their ‘at-risk’ customers are and then target them individually to improve their likelihood of payment. Although this method works, the use of advanced analytical methods can take the concept to whole another level. Following are ‘use cases’ of machine learning tools to improve collection methods –

• Micro-segmentation: Using advanced data analytics tools, lenders can understand their ‘at-risk’ customers more deeply. Raw data from customer transaction and payment history, communication history, recordings of agent-customer calls, etc. can provide great insights into the likelihood of default for every individual customer. Thus, every customer can be treated as part of its ‘own category’ and personalized strategy can be applied to maximize the chances of payment.

For example, lenders can decide which agent to assign based on the experience and history of the agent, which channels to use for communication, what content to use, if the lender should offer some debt restructuring settlements, if yes then what should be the optimum settlement offer, etc. Machine learning algorithms can offer a great level of granularity in helping lenders make decisions – the tools can also suggest the best time to communicate with customers, to maximize the effectiveness of efforts! More on this topic in the next paragraph.

• Optimizing Targeting Approach: Traditionally, firms have not been paying much attention to decisions like the best messaging strategy, best channels to use, best timing, etc. However, now the power of data-driven decision making has enabled lenders to optimize their targeting strategies. Through optimized targeting using websites, messaging, mobile applications, voice calls, and in-person meets, lenders can greatly reduce early delinquency possibilities. For example, through analysis of past agent-customer voice recordings, machine learning algorithms can predict which type of ‘communication’ works and which does not.

• Smart Agent Customer Pairing: Firms have been using the ‘bucketing strategy’ for agent-customer pairing. For example, very high-risk individuals would be appointed to the most experienced agents. Advancements in machine learning can be used to make this approach more granular. Based on agent behavioral and performance attributes and customers’ behavioral tendencies, the agent-customer pairing can be greatly improved. Data analytics can help in analyzing unstructured data like agent and customer phone calls to get better ‘fit’.

Concluding Remark:

Now, as much as we sing praises for agile talent, one thing cannot be denied. It is impossible to convert the entire workforce to a remote, on-demand one. Top administrative, financial, and operational positions cannot be outsourced - long-term workers who know the company inside out are necessary for these posts for the stable management and smooth running of the company. But free-standing tasks, especially in a company working on foreign turf or expanding their realm, can be accomplished with far better results, less time and lower cost when it is outsourced to a global, agile network of workers.