Using Data-Driven Geospatial Analysis to Optimize the Location of New ATMs in Banking Industry

Data-Driven Geospatial Analysis for New ATMs location optimization
AvatarMay 19, 2016

The data-driven banking industry has been an early beneficiary of big data and its many business intelligence (BI) applications. A particularly useful application is the selection of ideal geographic locations for new ATMs.


For banks, opening a new ATM involves substantial cost, in terms of real-estate investment and ongoing operation. Fortunately, banks can optimize this decision by analyzing abundant internal metrics, such as market potential, location of existing ATMs, and historical cash withdrawal trends, among dozens of others. External sources also provide a wealth of geospatial information, ranging from foot traffic to weather patterns, which affect cash inflows/outflows. These two factors – the expense of a new ATM and the abundant data available for forecasting an ATM’s performance – make optimizing a new ATM’s location an ideal use case for big data analytics.


Before the era of big data, banks used geospatial analytic tools like MapInfo software, which requires manual data intervention on sets of shape files to analyze a defined geographical area. This approach has many shortcomings, including:

  • Dependence on a desktop-based installation;
  • Scarce skill and limited availability of specialized developers;
  • Limited ability to scale for large queries; and
  • Less efficient use of very large data.

To improve the use of available data, banks need a method of geospatial analytics that allows for faster, parallel processing of data libraries on a large scale.


A banking client recently sought to improve its decision making on the location of new ATMs. It wanted to leverage an internal data warehouse, which contained key metrics on the client’s and its competitor’s ATM performance. The client also wanted to employ efficient big data technology, so we selected Vertica on the Hadoop platform. Vertica is a massive parallel processing (MPP) database by Hewlett-Packard Enterprise, and the open-source Hadoop software framework is used for distributed data storage and high speed processing.


The business goal was to identify optimal locations for new ATMs and to enable other geospatial capabilities by using customer and other geographic data. This increases an ATM on-us ratio and results in higher lead conversions using data driven intelligence.


Our approach entails validating the ATM’s proposed catchment using a combination of geospatial data and business metrics, scenarios, and drivers. The proposed catchment could be defined by a combination of pin codes, city boundaries, or a free-flow drawing produced by the bank’s analysts.


Once a potential area is defined, we use Vertica to analyze raw business and geospatial data on the particular catchment. We enhance Vertica functionality with Java-based user defined extensions to create functions, which serve as an analyst toolbox. The analysis produces estimated ATM key performance indicators (KPIs) for the proposed location based on historical data. The process entails the following steps:

Geospatial Picture1

The KPIs are automatically attached to the final shape file, which can be imported into any geographic information system (GIS) application to visualize KPI’s and make effective decisions.


This process can be performed for single or multiple, proposed catchments. The cost and estimated KPIs can then be compared and contrasted to validate the best location for the new ATM. The system calculates over 160 KPIs. The following table summarizes some of them.


Geospatial Picture 2

The implementation also involves building a master data model consisting of metrics on ATM use, point-of-sale (POS) transactions, debit/credit cards, customers, and other data points.


If you are considering new BI applications and analytics tools, let’s schedule a time to discuss what options are available.