This blog is part of a 4-part series on transforming a product-siloed organization into a customer-centric one. In part one, we discussed Data Capabilities, in case you missed it, you can catch up here. In this blog, we are moving on to the second capability in consideration of customer-centric decisioning, Insights Capabilities.
Pillar 2: Insights Capabilities
Convert your data into advanced AI and ML models and then rapidly and effortlessly operationalize them so they can serve as the cornerstone for faster and smarter decisions.
Pillar 2: Insights Capabilities
Convert your data into advanced AI and ML models and then rapidly and effortlessly operationalize them so they can serve as the cornerstone for faster and smarter decisions.
- Customer-level analytics remains a decisioning cornerstone in the customer-centric world.
- Unlike account and product-level models, customer scores use input data from across the entire range of products held by that customer and will predict a holistic, more broadly-based, level of risk. A customer score, therefore, is not optimal for any one product; rather it reduces the multiple product scores into a single, holistic, overall value.
- Model performance outcomes for customer-level scores can be challenging as sensitivity to what a “bad” customer means usually differs from line of business (product) to line of business.
- Developing customer-level risk management scores can be accomplished at either a two-level modeling approach or a one-level modeling approach
- A two-level modeling approach begins by first developing product-level scores (credit card, installment loan, mortgage, demand deposit), followed by feeding both product level scores and customer level characteristics to develop a customer level score.
- A one-level modeling approach consists in developing a single customer-level score by collating all customer and product-level characteristics as inputs. In such an approach, product-level scores are not part of the customer-level score.
- Having a customer score in place does not make individual product scores redundant as the two types of scores are not mutually exclusive. There is a place for both scores, to be used in different situations. Some decisions, such as authorizations and repricing, can fall at the product level and would be based on the applicable product score. On the other hand, decisions such as how to best manage the total customer exposure or whether to include a customer in a particular promotional campaign are customer-level decisions and therefore would use customer scores.
- In the case of product-level scores, it would be desirable that these are “enhanced” product-level models. In comparison to traditional product-level scores, enhanced product-level models leverage customer-level predictors from other products, while traditional product-level scores are primarily limited to predictors from the specific product being addressed. The inclusion of customer-level data to predict product-specific performance generally improves the predictive value of the score.
- The development and deployment of decision optimization techniques improve and accelerate the realization of customer-centric benefits and decision models can be deployed that cover such decisions as:
- Optimal global exposure
- Optimal product-level limits and associated price
- Optimal set of collections actions
Updated 4 years ago
Version 3.0