For debt recovery, AI offers the opportunity to learn from one’s data what a high-yield / low-effort matter looks like. We are able to extract from the computer its yield-effort views for any new matters we present it so that more money is recovered more quickly and the dead-end matters are left until last.

In this example, for one tranche of 20,000 matters handed over to a debt-recovery agency, an AI model was trained and then applied to the tranche. It was then observed what proportion of the total debt in the tranche would be collected in the next two years after dealing with a given number of matters. This was done assuming that the matters were processed in random order and then compared against the amount collected if the matters were processed using the AI model’s predictions. The results are as follows:

  • For 500 matters, the random allocation generated 4.1% of collections while the model generated 16.1% which is almost 4 times the success of collecting randomly.

  • For 1,000 matters the AI model performed 3.3 times better

  • For 5,000 matters the ratio was 3 times

  • For 1,0000 matters the ratio was 2.1 times

It can be seen that the model is 2 to 4 times more efficient than randomly working the matters. The reason that the multiple decreases as more matters are processed is that the AI model “upfronts” the best matters. This shows that the model was able to identify the most profitable matters that should be worked first. short hair extensions < /div>