It depends. If you are an analytics professional from the banking-side who has taken up the mantle to perform micro-finance analytics, you will possibly see a little or no data. You will also see little use of the tools and artefacts of the analytics that works in banks. Probably, you may conclude it is a dead horse. But, if you are an analytics professional not from the banking-side, you may see different kind of data. You may also see a potential to use of tools and artefacts of the analytics that have rarely been used in the banking domain. And possibly conclude that is indeed a sleeping giant.

Photo Credit : ShutterStock,

Clearly, the world of micro-finance consumers’ socio-economic environment is starkly different from that of the banking consumers’. For one, for the usual banking consumers there is enormous amount of structured data. Consider a banking customer’s data about the banking (savings or current account), credit (loan account) or purchase (credit-card) history. One can run every kind of analytics: from simple cross tables to complex multidimensional OLAP cubes. One can run advanced statistics, such as predictive analytics, to predict who will buy what, by when and how much. Or predict other matters of interest such as loan defaults. Not much of data that is image or text is used. Micro-finance consumers have practically none of the structured data that the usual banking consumer has. And there is an increasing potential in using non-structured data, such as image, to answer some of the basic questions.

For instance, the Google Earth Images of night-time data of Dharavi, Mumbai over the past 15 years shows increasing intensity of light. On the other hand similar data of Gardanibagh, Patna shows decreasing intensity of light – an indication of migration. This simple insight can drive the overall market demand estimation in both volume and value of loans terms.

One very senior micro finance entrepreneur remarked that the way he measured the sales potential of a tea-vendor, a potential customer, was by stationing a person for a couple of days measuring number of cups sold, other items sold, etc. While there is novelty in collecting sales data this way (which for a usual customer is just asking for two-years of P&L), the idea cant scale. Imagine the same data collected through an installed camera rolling, a computer vision algorithm counting everyday the number of cups of tea brewed. The scaling potential of collecting such data is immense since computer vision now can recognize almost any item that we ask it to recognize.

The idea can be extended to almost any assessment for a micro finance household. After securing necessary permissions from any household, a rolling mobile camera and thereafter an AI lead Computer Vision can assess the ‘wealth’ by noting the type of house, fridge, TV, bedding, toilet, roof, etc. Of course, one can also assess the number of household members, their gender and probable age. It is quite possible that the demographic data so collected is actually far richer than the ones that is collected and stored in any other manner. It is also easy for us to see that changes in such data can be captured in a similar fashion thereby ensuring that the Micro Finance institution has the latest demographic data. Such latest data is something even accomplished banks find it hard to get; I was looking to compute the credit-risk score for a large South Indian bank and found that several aged customer’s occupation and incomes, filled several years ago at the time of opening the account, were so outdated that using such variables was rendered fruitless.

And then there is the (African) case of rural / agri micro finance determined by cash-flow (nothing new here). But the way cash-flow was generated per customer was new. The micro finance company flew a slew of drones over crop-lands and installed IoT at critical points to estimate quantity of produce. Armed with this data one could closely estimate the actual cash flows of a farmer. There are early evidences of estimating livestock similarly.

As one can see from the above, AI led Computer Vision can help uncover several “dark data” and estimate the market size, the demographics and cash flows critical for some very fundamental decisions. It can be used also for operations.

Attendance of members in a Joint Liability Group (JLG) is an important determinant of risk and continuity of a member. Today the attendance is taken physically by an agent with attendant risks of poor data; one then also transcribes the data at the branch or head office. One can move the attendance to the app which can process a picture, recognise faces and mark attendance directly to the tables. Several other processes, such as cheque clearing and new bank or loan application form processing can be automated using AI lead computer vision.

To conclude, trying the same ways in which to collect and analyse data as is normal in established banks will not help the cause of a distinctly different environment of micro-finance. That is a dead route. However, if one sits up to use really novel ways to collect and process data, e.g. using AI lead Computer Vision, the micro finance analytics is a sleeping giant. Of course, it is neither easy, nor cheap to quickly imbibe such techniques. But that path will create massive competitive advantage, superior customer experience and increased profits.