M-Pesa – cash flow-based credit decisioning

The behaviours and spending patterns of the prospective borrower are the primary inference driver for the Finexos credit risk algorithm. In this case study, we take a look at the strong empirical evidence to support the fact that transactional data, when analysed correctly, is actually a far better predictor of credit default than traditional credit metrics.

In 2014-2015, our Solution Architect led the implementation of the credit scoring engine for the M-Shwari loan product, which was delivered exclusively over the M-Pesa mobile payment platform. M-Pesa is the leading mobile payment platform in the world, with over 40% of Kenya GDP flowing through it.

In that instance, the borrower’s behaviour on M-Pesa was the only source available; they did not have access to the borrower’s income, employment status, or performance in paying off previous loans. In fact, given the tenuous enforcement of KYC in obtaining SIM cards in Kenya, in many cases the borrower’s name was unknown.

Nevertheless, using the M-Pesa spending behaviour alone, the model developed delivered a 26% reduction in the non-performing loan ratio of the underwriting bank.

Beyond the significant improvement in default rate, cash-flow based analytics delivered the following for M-Pesa:

  • Peak loan originations per day went from 500 to 50,000
  • GDP per capita of Kenya almost doubled
  • 9 million loans originated and paid off in first 5 years of operation
  • The solution made credit available to millions who had previously never before been able to access it

The model developed in Kenya is of course specific to that market. The Finexos solution leverages considerably more powerful and sophisticated machine learning methods than those available in 2014-2015, but the Kenyan results remain an excellent proof point that making credit decisions based on spending patterns is both more accurate and more financially inclusive than traditional credit scoring metrics.