Despite being a "field RA," I don't usually spend much time in the field. Most of my time is spent in our office in Thanjavur, drawing maps like this:
|Each one of these pins is a village.|
These maps will help our partner organisation, IFMR Rural Channels and Services, a company that provides complete access to finance through the KGFS model, decide where to set up branches for two of its strategic business units, Vellaaru KGFS and Thennaaru KGFS. Over the next five months, each is planning a rapid expansion across two districts of Tamil Nadu—Ariyalur and Pudukkottai—and the project I'm working on will try to understand the impact of that expansion. But before the study can begin, and before the branches have been built, each business unit needs to decide where to build its branches—and because the KGFS model aims to include as many people into the formal financial sector as possible, it has to choose its branch locations strategically. Using data that come from CMF surveys, we hope that together we can set a more optimal distribution of branches across these districts.
Securing the best branch placement requires a tricky blend of global and local optimization. Each branch has to not only be easy, accessible, and make perfect geographic sense to its client base, but it also needs to mesh perfectly with every other branch in the district. Branches can’t be spaced too close, or too far apart, otherwise people will get left out. But in the name of achieving a perfect distance between branches, we cannot place them in out-of-the-way villages, or on the other side of bridgeless rivers. That would undermine the very kind of real, sustainable inclusion that we’re trying to achieve.
The data and technology I have makes global level optimization entirely possible. And I try to push my dataset to make branches locally optimal too. But on this dimension I find that I’ve hit a ceiling. One important lesson I’ve learned from mapping is that when it comes to understanding the particulars of a place, there is no substitute for field knowledge.
It seems that for every rule, or average, or generalization one can make from aggregated data, there exists at least one exception. For example, we’d like to build branches that are close to their target population. That is, we don’t want customers to travel far. And although satellite imagery can give you a good first impression of the distance between places, it can also be highly deceptive. Two villages that appear close together on a map may be practically far apart if the road between them takes a roundabout route. Or if the road is a dirt road rather than a main highway. Or if there is a forest, or river, or other obstacle in between. Or if the bus connectivity is not so good. If we could anticipate all the reasons why the information we’re trying to capture, then we could build those variables into our survey. But there’s always an N+1 variable. That’s why, despite having a robust dataset, there are some questions we cannot answer without going to the place and seeing it for ourselves.
These trips to the field are now quite surreal for me. When I visit villages that I recognize from my service area mapping, I can't help but imagine a giant colored pin standing in the center of town. When I recognize a village name on a passing signboard, I feel like I've spotted a celebrity. It’s the same feeling of awe and embarrassment—the awe that comes from seeing something that only existed in your mind as a distant object actually materialize before you; and embarrassment that comes from knowing so many factoids about people without having ever been formally introduced. Over time, though, I imagine that I won't feel as starstruck as I was when I first saw Thirumanoor, Keezha Palur, Athanakurichi, or many of the other towns of Ariyalur that have figured so prominently in service area mapping.