Project Everest

Proposed Experiment

[Proposed Experiment]: SoCon Fiji - Alternative Risk Assessment Solutions - July 2018

Lean phase: Testing proposed solution (MVP prototyping)


Purpose of this experiment

This experiment is to test the assumption that alternative data methods can assist with accurately measuring the risk of a loan applicant.



Alternative data helps to assess the risk of a loan applicant accurately and is reflective of someone’s ability repay back a loan.


Success Metrics

  • Default rate

  • Factors affecting ability to repay


Green light: If our assumptions are valid, we will proceed using alternative data as a source to measure risk.


Success Point: If 80% of our assumptions are proven then we will continue to use the alternative data source being tested to predict risk. This metric gives room for outliers and the influence of external factors while still being statistically significant for a large enough sample size (n >30).


Failure Point: If 80% of our assumptions are disproved then there will not be enough data to suggest that the alternative data source being tested can be used to assess someone’s risk.


Red light: We will look to iterate and improve the experiment through testing another metric from the same data source, or using a different alternative data source entirely.


Method/Experiment Build

  1. Identify customer segment e.g. farmers

  2. Identify alternative data source e.g. metric on water bills

  3. Make assumption on the metric will reveal about the repayment behaviour of the segment e.g. famers who pay water bills will repay their loan

  4. Create two groups:

    1. Farmers who paid their water bills on time (assumption: unlikely to default)

    2. Farmers who did not pay their water bills on time (assumption: likely to default)

  5. Provide loans and monitor if the above assumptions are correct (default rate datapoint)

    1. Assumptions are correct if the metric of water bills accurately measure risk

    2. Assumptions are not correct if the metric of water bills used does not accurately measure risk. In this case we iterate and use an alternative metric from the same source or use an alternative source

  6. Survey loanees to deduce risk factors that affect rate of default (risk factor datapoint)


(See diagram below for visual representation of this process)


Time period

This experiment should run between December to the end of January to have a 2 month period on loan repayments and to ensure a statistically significant sample is gathered (n>30). This time period will be dependent on the amount of the loan provided. 


Further comments

While this particular experiment is specific to alternative data, we believe that this method of experimentation, if successful, could also be applied to looking at other methods of reducing risk (e.g. using financial training). This is something that could be explored in the future. 

edited on 5th September 2018, 23:09 by Justin Hakeem

Andrew Vild Aug 3, 2018

I really like this Ineke! It is a very intelligent way of collecting data that already exists and probably has a strong correlation and causation.

How are you intending on getting this information? Asking individuals themselves or going to the water company and asking for the data - I am concerned they wouldn't be willing to share personal data such as this as it would violate most privacy policies. This is my assumption, if you have done research or talked to people on the ground that indicates otherwise then that's awesome.

I would be happy to have this experiment adopted if it went that one step further to say HOW we're going to get this data.

Reply 0

Rose Martin Aug 4, 2018

My initial thoughts on this would be to get it from the individuals themselves, however I see your point on the privacy thing. This wasn't something totally considered when writing this experiment as we were just ideating logical ways to test whether alternative data would work in the Fijian market; however I'm doing some research on alternative data now and will make a post on it soon so I can get a better idea of things like privacy laws etc. and how we could best obtain data such as this!

Reply 1

Ineke Mann Aug 5, 2018

I would suggest that we could make this kind of information compulsory for individuals to provide in the application process just as they would their financial statements. Our ability to do this is based on assumptions though and we have not talked to any locals about it so it is definitely something we need to understand further before implementing any experiments.

Reply 0

View all replies (2)

Andrew Vild Aug 3, 2018

Status label added: Proposed Experiment

Reply 0