Gmid

Analytics Case Study: Collection Analytics Services

Objective

The objective of the Scorecard was to output an ordered score that would distinguish the potential good accounts (which have higher likelihood of regular payments) from the bad accounts for early delinquent (bucket 1 i.e. <30 DPD and bucket 2 i.e. <60 DPD) portfolios. These scores were to be used for prioritization of collection efforts on this portfolio.

Background and Challenges

The challenge was to optimally allocate collection efforts for maximum revenues. The bank wanted to use predictive analytics to enable it to devise targeted approach rather than a flat strategy across the portfolio.

The predictive model was supposed to discriminate the accounts that had higher likelihood of regular payments from the potential bad accounts. The client wanted to use this scoring to divide the universe into segments and then decide multimedia follow up treatment like SMS, IVR and Outbound calling etc at segments levels.

Our Approach

Historical profile and transactional data was used to first identify attributes of the customers that had a governing role to play and then create a predictive scorecard.

The regression modeling technique chosen was logistic, and selection of important fields was conducted using Crosstab analysis, regression selection methods, and information values. The model was developed over 6 month data and validated over the next 6-month data.

The model would evaluate all eligible accounts on the first day of a month with the prediction (score) valid for the rest of the month.

Results

The final output (score) was created at account level such that higher is the score, more likely it is for the account to default in the next month.

The score is calculated on the 1st of every month and used for the next one month to devise differential multimedia treatment at segment level.