Abstract
D.Litt. et Phil.
The Self Service Channel (SSC) of First National Bank does not have a consistent and scientific method whereby the potential of a site could be evaluated for the placement of ATM devices. Estimating the potential number of transactions that could be generated at any particular site will enable FNB to determine whether or not an ATM would be viable at that site.
Subsequently, the problem, or reason why this study was conducted, could be formulated as follows: There exist no uniform, consistent and scientific method that can be applied to predict the potential number of transactions that could be generated at an ATM, at any potential site.
Therefore, the primary objective of this study was to develop a transaction-forecasting model (in the form of a regression model) whereby the potential number of ATM transactions at any given site can be estimated.
Secondly, it was essential that this model must be developed in such a way that it is practical and easy to apply in the working environment. Employees of FNB must be able to use the model as part of their daily workflow and therefore it had to be made as user-friendly as possible.
As the basis of this study, the first chapters focus on the study of available literature regarding location analysis and site selection. Literature regarding ATM location analysis as such is extremely limited, and therefore literature regarding the location of retail facilities was used as the foundation of the study. As with ATMs, most retail facilities are dependent on high pedestrian and traffic volumes to increase their sales. Variables that relate to, or that could be an indicator of pedestrian or traffic volumes at a particular site, was used (where possible) as variables in this study.
ATM sites can be classified according to the type of facility where it is located, e.g. shopping centres, garages, etc. Each of the site classifications that were identified in the study, is influenced by different variables, which means that a different combination of variables could be included or used for each site classification. This resulted in the development of separate regression models for each site classification, depending on the specific set of variables influencing ATM transaction volumes at that site classification.
When the variables were selected, one of the criteria that had to be kept in mind was the availability of data. One of the objectives of the study was to provide a model that is relatively simple to apply in practice. Data that needs to be gathered for input into the model (independent variables) must be easy to obtain.
Furthermore, to develop a regression model, the variables used in the regression analysis must be quantifiable. Certain of the variables already had values that could be used. An example of such a variable was the Gross Lettable Area (GLA). The GLA of a shopping centre is measured in m², and this figure was used to measure the correlation between GLA and transaction volumes at shopping centres.
Other variables, more specifically accessibility, visibility, competitors in the vicinity and FNB ATMs in the vicinity were not easy to quantify, and methods had to be developed to quantify these.
After identifying the criteria that was used to quantify the independent variables, a sample of ATM sites - for each site classification - were selected. For each of these sites the relevant data were then collected. Data were collected by means of site visits to each of the sites included in the sample, GIS databases and requests for information from petrol garage owners, shopping centre management and retail store owners where the ATMs in the sample are located. Multiple regression analysis was applied to the variables in each classification that proved to have the strongest influence on transactions at that particular site classification.
Of the 14 site classifications, models could be developed for 5 classifications. These 5 classifications, however, which include shopping centres, business nodes, convenience stores, garages and industrial areas, accounts for over 90% of all the ATM locations at FNB. The other classifications were excluded due to a lack of sufficient information for regression purposes.
These models provide a scientific approach to ATM placements. It will also improve the decision-making process in FNB’s Self Service Channel, and result into cost savings by preventing the installation of ATMs at unprofitable locations.
This study has made a significant contribution to the banking industry and the field of urban geography in terms of new knowledge, models and methods that were developed.
There are various factors that impact on the performance of ATM devices at different locations. However, it is important to keep in mind that any scientific/statistical model could be applied only with limited success in practice, as human behaviour - and anything depending on it - cannot be fully explained or predicted by statistical models. Therefore, when these models are applied, the discretion and experience of the person will always play a role in reaching a final decision.
The models and methods that were developed in this thesis will be developed and refined further on a continual basis to increase its accuracy and prediction ability.