Building an accurate scoring model from unconventional, raw text

MPower, a Zurich-based, B2B renewables start-up offers finance and technology to emerging market energy suppliers, who distribute vital plug-&-play solar kits on a lease-to-own basis.

The data science team needed an accurate scoring model to assess the creditworthiness of applicants – mostly unbanked, with no credit history – and broaden energy access through smart, responsible lending.

One of the challenges of building a data model is finding the meaningful features in a data set because features act as the building blocks of a model. However, feature engineering is particularly challenging when dealing with unstructured, raw, free-form data.

SparkBeyond’s platform assisted MPower in feature engineering by harnessing text mining to find patterns in written submissions, such as crowdfunding loan listings  – searching for features in raw text data.

Working to a tight deadline, a joint SparkBeyond/client team used Discovery’s ability to seamlessly join a wide variety of unstructured, raw data – crowdfunding loan listings, handwritten applications and social networking characteristics – to reveal a wealth of valuable customer insights.  

The platform swiftly interpreted written application answers, automatically grouping potential customers according to sector and type of business. Geolocation analytics merged personal profile data with applicants’ financial responses, and added macro-economic data by time and location.  

SparkBeyond Discovery's powerful analytics generated robust behavioral and reputation profiles, giving the start-up a credible assessment of an individual’s willingness (rather than ability) to repay their loans.  

Overall, the model building and evaluation process was hugely accelerated, and with Discovery’s clearly explainable insights, the start-up was able to communicate its credit score findings to applicants – and potential investors.  

SparkBeyond increased the client’s revenue growth, and their future funding rounds will be underpinned by accurate cashflow and revenue projections.

Looking ahead to 2022, they plan to roll out a wider range of solar-powered products across the continent, with the ultimate goal of making clean energy affordable to everyone.  `

The start-up’s Head of Partnerships told us, “We can efficiently implement what we learn, and use the platform to see how different countries correlate – and whether it makes sense to apply lessons from one country to another.” In Sub-Saharan Africa alone – where more than 50% of the population lacked access to electricity in 2019 – the economic and social potential is vast.