If you have worked in a government debt collection environment with millions of debtors and billions of dollars of outstanding debt, you will immediately grab the complexity of this question. You can read about it in many published papers, “one-size–does-not-fit-all”. Debt collection has long moved away from a static process along a well-defined escalation path. Unfortunately, the one-size-does-not-fit-all discussion describes what “you should not do” and not how you should actually tackle the complexity of a massive debt book. There are different debtors with different attitudes, different cultural backgrounds and so on. Each different debtor, each debt, each situation needs to be treated differently. So for a start, in order to organize hundreds of permutations the debt/debtor data needs to be classified/segmented.
Thereafter the task – sounding so simple – is to find, for a given debt/debtor (or rather class thereof), the best possible collection step (atomic intervention such as a correspondence, call or visit) and then hope that the payment will follow. It was often suggested to me that Predictive Analysis would solve such an issue with ease. Yes true; debt collection is a great field for Predictive Analysis as it is clearly a big data problem. Those clever predictive algorithms must surely be much more intelligent that humans trying to model – what they perceive best – into a rule engine.
But wait; there was more content that those human rule modelers put into their rules. There were all those boundaries: Legal boundaries (e.g. you need to follow those step and the threshold must be met), policy boundaries (e.g. respect religious holidays or a deceased debtor situation), cost prohibitive boundaries (would you really send your mobile workforce to collect 50 dollars), quality boundaries (without a proper, verified address or telephone number many activities will not work) and workforce boundaries.
Therefore, it is clear that the predictive outcomes have to work hand-in-hand with these boundaries. In addition, Predictive Analysis only works in case you have a good basis to compare past outcomes with, otherwise new methods will never be selected as the algorithms cannot judge the likelihood of their success. So new methods have to be constantly tested. For some time ago, this concept is referred to as “Champion-Challenger” principle. Champions are the existing, well-known methods of operation and challengers are the new ones, which might proof to be better in a changed environment or in certain situations.
In the world of Predictive Analysis, the methodology of constantly challenging existing activities is not only an option, but it is necessary and crucial to the overall optimization success. Collections systems must enable staff easy setup of those new challenger strategies based of best understanding of the collection data and trends. In an ever-changing environment, this process is very similar to ad-hoc campaigns where campaign strategies need to start with an interactive definition of a campaign focus group. Once the focus group is defined new collection steps/rules need to be defined for the challenger strategy. In other words, the collection workforce must become intelligent debt managers optimizing the system and move away from case decision-making.
To add to the complexity, often the right order or mix of activities can significantly influence the collection outcomes, so instead of simply predicting the best next collection step, prediction of the best strategy (a number of steps combined) seems superior. And finally there is the Minister challenging the collection department asking “How much additional revenue could we expect (”predict”) in case we could overcome some of the described (legal, policy or workforce) boundaries?”
So again, simply, at run time, deriving/predicting the next step seems way too simple. What is needed is a comprehensive predictive scoring system, which is being validated at run time against the current boundaries. The predictive scoring results of each collection strategy/step are used to support manual decision making and in order to predict outcome when simulating boundaries changes.
The following slides summarizes how the different, described parts need to be orchestrated in order to optimize the collection yield:
SAP is deeply committed to overcome the debt collection complexity by giving collection agencies all the optimization tools – Collection Strategies, Rules, Predictive Analysis, Champion/Challenger setup, Campaigns and Exceptions for manual decision-making – on one integrated platform.