Finding our Farmers; or, Shamba Iko Wapi?
In the era of digital loans, knowing your client’s location is critical.
In rural parts of Kenya where most smallholder farmers reside, determining a farm’s exact location remotely can be nearly impossible.
Reconciling these opposing statements can be challenging, but instrumental in increasing the flow of capital, products, and services to smallholder farmers throughout Africa.
At FarmDrive, we use alternative data to improve the operations of financial service providers (FSPs), so they can better serve smallholder farmers. Our alternative credit scoring algorithm and data products help FSPs reach new customers, deepen customer relationships, automate Know-Your-Client (KYC) verification processes, manage risk, and deliver targeted financial services.
To do this, we incorporate a variety of data sources in our algorithms; from soil quality and weather, to ID registries and repayment histories. Many of these datasets include a geospatial component. For example, in order to determine the soil quality, weather conditions, and market accessibility of a farm, we must first know its location. Thus knowing the location of our clients’ farms is crucial for the success of our algorithm.
Ostensibly, it may seem easy to collect geolocation data. However, while online maps and GPS on smartphones have become commonplace in many parts of the world, smartphone ownership has yet to become ubiquitous in rural Kenya. With no clear addresses and few GPS-enabled phones, finding a particular farm often relies on the use of landmarks known by most people in the community.
FarmDrive takes advantage of this fact, by considering one landmark that is universally known across Kenyan farming communities — primary schools. Our algorithm uses the locations of thousands of primary schools in order to predict the location of a farmer with a few simple questions on our SMS platform.
With the hope of optimizing the performance of this geolocation model, we conducted some field research which resulted in a number of interesting conclusions.
Through conversing with smallholder farmers, we found that most are familiar with several primary schools in their community, not just one or two. More importantly, farmers have a pretty good sense of how long it takes to walk to each school. This is shown by the comparison of google maps walking distances to farmer walking distances here:
This is good news for our algorithm because the more schools a farmer is familiar with in their surroundings, the easier it is to hone in on their specific location.
Another discovery we made was that it is more difficult to predict the location of farmers living next to natural barriers such as bodies of water or mountain ranges. These natural barriers limit the number of schools surrounding the farmer’s location, decreasing the accuracy of the prediction. In the field we encountered several locales in which this was a problem, including one farming community next to Lake Naivasha, and another community adjacent to the Aberdares mountain range. With this insight, we have been able to adjust the algorithm such that effects of natural barriers on geolocation determination are countered.
The results from our field study have helped us fine tune our geolocator such that it can provide significant value to FSPs and even agriculture sector players. For example, with our geolocator, FSPs have greater insight into the agri-businesses of the farmers to whom they lend, and market offtakers can quickly source produce without first embarking on costly and time-consuming scouting trips. Such small, but consequential improvements will open the flow of credit, products, and services to the backbone of the agriculture sector: smallholder farmers.
