Many institutions we work with want to segment their customers, but very few end up doing it. Yet segmentation can save costs, improve client services and help design and target product offers. It can also help reduce risk Why is segmentation so hard when you're at the bottom of the pyramid?
Common segmentation strategies use "big data" to identify groups of clients and segment them accordingly. They can then be targeted through promotions, media, social marketing, and mailing campaigns based on industry, gender, zipcode, etc. If you are a provider of financial services to low-income households, the first challenge is that you most likely, you don't have "big" data, so the depth of analysis might be limited. However, you might have useful data, which can be a first step.
If you begin by analyzing "small data", for example an indicator of how many transactions per month a client makes, you might be able to identify cross selling or other opportunities. A provider's instinct might be to send text messages out to all such clients promoting new products, based on this information. However, data bases have notoriously bad contact information. Additionally, sms sales are still tricky in the low income space and the sales "muscle" of many of these models is still very much composed of front line staff or agents. Synergy between head office data and segmentation and field-level operations can be difficult, at best.
But this does not mean that segmentation is impossible. Only that it should be based on observable characteristics rather than transactional behaviors. These are much easier for front line staff or agents to act on.
However, data sets are not very rich in observable characteristics. Few mobile money platforms can tell you the gender or business line of a client, for example. A microfinance institution may similarly not know the housing condition of a client that would enable a tailored housing loan offer.
Our solution to this has been working with front line staff on what we call "Bottom up Segmentation". We start with available "small data", but also work closely with front line staff to understand segments based on specific research inquiries that we might have and based on observable characteristics. One group of loan officers explained that income levels could be ascertained by the number of wheels on a client's main mode of transportation (motorbike, tuk tuk, or car), for example. Another group could tell how volatile farmer income was by the presence of machinery on the farm (more investment meant less crop risk).
Front line staff and agents can also project incorrect views of client behaviors onto their analysis. This is something that a skilled facilitator, and robust interview tools, can address. We also corroborate some information directly with clients as a test. For example, gender biases are common. One group of loan officers explained that in diversified households, wives with shops made less income than their husbands with crops. In some months, that was the case, but over the course of the year, we found that the sum averaged out, especially after the cost of borrowing was taken into account. "Bottom up Segmentation" is both an art, and a science. But used wisely, it can open an institution to new business opportunities in the present, while we wait for the "big data" of the future.