Introduction
The proposition to utilize web-based credit approval systems to enhance the credit approval process is an appealing prospect for banks and other lending institutions. This is because the credit approval process is laden with risks that all lending institutions spend many resources managing.
The risks arise from poor quality information brought about by fraudulent applications, interagency nature of its collection and outdated information. By using a web-based system, the efficacy of the process will improve, reducing the process overheads and increasing the margin for the banks. It calls for the use of a system architecture designed to produce the best possible results from the analysis of credit applications.
There are a number of system architecture options to choose from in the process of designing such a system. Their typical composition includes, a database, model base and a user interface, which interact to produce the analysis report that decision makers rely on . They form the critical components of the Artificial Intelligence (AI) systems used for decision support.
Burnside and Kahn (2004) identified the most common AI tools as, “Bayesian Network, Artificial Neural Networks, Case-Based Reasoning and Rule Based Systems”. In the case of a Web-Based decision support system for credit approval, the system must solve two problems.
The first problem is the comprehensiveness and accuracy of information. This problem arises from databases in use because each database has its limitations. Some provide more wholesome information on clients than others.
The second problem is in the procedural issues that relate to the handling of the information. The preferred solution encompasses, on one hand, a large data collection capacity that spans the entire internet, and on the other hand, a superior processing capacity of data for the availing of critical information credit officers require to make lending decisions.
The system proposed to meet this challenge will utilize Bayesian system running on a web-based platform for frontline screening. It will make possible the identification of clients with the best possible credentials for credit.
The second set of AI tools it will use is Rule-Based System and Case-Based Reasoning to identify potential for fraud. The third tool for it to use is the Artificial Neural Networks to provide massive fraud detection capabilities.
Proposal of System Architecture
Performance Specifications
The required system must meet the following specifications.
- It requires the capacity to handle a large volume of data. Data handling includes data acquisition and filtering. The security needs of the system will also require attention.
- The system should identify and flag incomplete or inconsistent individual applications to detect fraud and inconsistent applications.
- The third element that the system must have the capacity to handle is the ability to detect massive fraud. This stems from the possibility of the generation of multiple identities to defraud financial institutions.
AI tools
The proposed system employs AI tools at three levels.
Bayesian Networks form the first level. They consist of a structure, a set of probabilities, and an inference algorithm, which forms the basis of operations. Expert input goes into the development of the inference algorithm. This makes it useful for frontline screening.
A practical application of this AI tool is the separation of clients who have the characteristics that match those of clients with good relationships, from those whose initial conditions match clients with difficult relationships. This will help to weed out clients who do not bring in the potential for a good relationship with the bank.
The second level consists of Artificial Neural Networks and Case-Based Reasoning systems. At this level the feature that it will have include the capacity to detect fraud by identifying similar patterns in new credit applications, which may be fraudulent.
Artificial Neural Networks provide the basis for deriving relationships between various data sets. On the other hand, Case-Based Reasoning provides the capacity to compare fresh applications with the ones stored in the database. It provides the credit officer with the comparison of the two data sets ensuring that there is no repeating of mistakes made in the past.
The third level will comprise of Rule–Based Reasoning, which will allow forward chaining, and back chaining to analyze applications.
This tool makes it possible for credit officers to determine whether an application will end up in a successful relationship poor whether it will end up in fraud. In addition, it can provide information that will make it possible for the banks to design its marketing effort by starting with the desired final conditions, and using the tool to generate the initial ones.
Physical Architecture
The physical architecture of the system will comprise of the standard computing equipment. There are a number of topology choices for the system.
They include the star topology that links PC’s in an asterisk pattern, bus topology that have all the PC’s connected to a main network line, ring topology whose main feature is the circular interconnection of all PC’s using, “paired physical interfaces” with other network equipment.
Point to point connection applies when there are only two devices in the network, while point to multipoint applies when all the devices connect to a specific device.
Schematic diagrams of star, ring and bus topologies
For the system under consideration, the best system architecture will be a hybrid system featuring the bus topology, and the point to multipoint topology. All the computers used for analysis will connect to the bus while the main storage server will be at one end of the bus with a point to multipoint connection.
The other connection and the end of the bus will be the internet server. In addition to the PC’s, the system will require a number of components that will provide the basis for utilization of internet search capabilities. This includes servers, switches, routers, hubs, and network cables.
Protocol Architecture
At the protocol architecture level, there is a choice of two protocol models. They are the TCI/IP model and Open Systems Interconnection (OSI) Network Model.
There is widespread use of the TCP/IP model, hence it presents the least installation challenges, and has capacity for expansion and cross platform communication. All the layers that the OSI model has are included in the TCP/IP model. This means that there is minimum loss when using TCP/IP.
Comparison of TCP/IP and OSI Model Layers
Therefore, the proposed system features four artificial intelligence tools at the software level, the TCP/IP protocol, and a hybrid topology that includes bus, and point to multipoint topologies. The three elements take care of the software requirements of the system, the physical interconnection and the networking platform required for making the entire system operable.
Selection and Justification of Software Packages
The Business Criteria
Business operations design their systems to achieve any number of business objectives that ensure that the business returns a profit. The two forms these measures take are increasing revenue and cutting costs. The proposed system will lead to increase in revenue by eliminating business uncertainties.
It will do this by providing the bank with the ideal starting conditions in the lending process, checking it against the application brought in and based on the system, making appropriate judgments. With this system, the bank can quickly determine whether a particular credit relationship is worth pursuing in the first place.
In addition, the banks will be able to work backwards to identify who its target customers are, therefore providing a greater degree of focus for its marketing efforts. On cost savings, the system will seek to eliminate fraud, which is a serious loss risk to the bank. This will protect the banks revenue and will ensure the bank does not incur the losses through fraud. This will improve the margin.
Management Criteria
The system will be labor intensive at the onset. It will require various experts and analysis to develop the AI components, while networking engineers will be the ones to install all the networking equipment. In addition, software programmers will develop the system and install the required user interfaces which the credit officers will use to access the system resources.
The justification of all this effort from a management point of view is that it will reduce the effort spent in analyzing applications and related costs, such as litigation costs from fraudulent applications. Other management issues that the system will solve include better marketing strategy because of better identification of the target market.
Technical Criteria,
The choice of the system includes the best technologies available in terms of AI, the best physical topology that provides for enhanced functionality, and the use of TCP/IP, which is the most recommended networking protocol. This means that this is the best possible design given current state of development and limitations. In addition, elements like ease of implementation and maintenance come to the fore.
Other considerations
Other considerations in the criteria included in the system design include the security requirements of the system. As an internet based system, there need to ensure that the system is secure is high. Cyber security of the system should have a high degree of robustness because the system will face cyber attacks from different quarters.
Explanation for the choice given
The choices given for the above system take into account the need to implement a robust system that can handle the sensitive requirements of AI tools and to provide uncompromising performance required for the banking business. The business will not have much time for down time hence the software implementation must be such that it has minimum maintenance requirements, and performs its tasks well.
System Evaluation Measures and Methods
Fisher (2008) identified six measures that a proposed system needs to go through before evaluation. These measures include measurement of functionality, technology, software vendor, implementation, maintenance, and support, and the total cost of ownership.
Functionality refers to the way the software fulfils its assigned role to it in the system. This calls for a careful comparison between software options that can support the required system functions. The one selected and implemented must be the one that fulfils the functions assigned best.
The second measure to consider in the evaluation of a software option is technology. Technology refers to the degree of congruence and compatibility with the programming platform and the database in use.
This means that the software option chosen should be able to work well with the systems carrier platform and the form of database that it will interrogate. It also calls for settling for software that in-house staff members are familiar. It is also prudent to pick software that will operate well within the rest of the organizations systems.
The third major issue to consider during software evaluation is the degree of support that the vendor will provide after the implementation of the software. It is crucial to know whether the vendor will provide enhancements and additional support after selling the software. Fisher (2008) says that when someone buys software from a vendor, he is not just buying software, but he is establishing a relationship with the vendor.
The fourth major consideration during software evaluation is implementation. Again, the vendor is critical. The idea is to find a vendor who will provide support to the implementation team and continue providing support during the entire implementation phase to ensure that the implementation team handles all technicalities.
The next issue that is critical in the evaluation of software is support and maintenance. While having in-house staff to maintain the system is a good practice, the need for a relationship with the vendor where they provide upgrades and regular software check is very critical. This stems from the fact that the vendors provide a wide range of software solutions to different clients.
They therefore have a wider experience with the systems operations and functions. It makes them uniquely qualified to find holes in the system and to close them up in their entire software range before they cause any real damage to all their clients.
Finally, there is need to have in mind the total cost of ownership when evaluation software. The costs here include cost of licenses, cost of maintenance and cost of upgrades. Fisher (2008) recommends looking at these costs over a five to seven year horizon to determine the long-term costs of implementing the software.
In order to ensure that the entire project achieves the objectives, the need for the application of good IT project management skills will come into play. The project will require organization into stages marked by relevant milestones. This will make sure that there is adequate preparation and implementation of each stage including testing to achieve the set objectives.
Conclusion
The process of implementing a web enabled hybrid intelligent system that supports credit approval process in banks has several bearings. From the discussions presented above, we can conclude the following.
A strong business, management, and technical case for implementing software project to address the credit approval problems that banks face, exists. The business case is that it will reduce overheads by improving the efficiency and certainty of the approval process.
It will also reduce losses associated with fraud and will make it possible to cut back on litigation costs arising from defaulting, thereby reducing overheads. From a management front, the system will reduce the overall business management requirement of the system by taking up the work required to process applications. Finally, the availability of the AI techniques to implement this system makes it technically viable.
The system will use a combination of AI tools to achieve the goals. These techniques include Bayesian Networks, Case-Based reasoning, Rule Based reasoning and Artificial Neural Networks.
The arrangement of these tools will be to provide the best result for the functions since each has its strengths and weaknesses. Use of TCP/IP as the most appropriate protocol architecture will bring in the benefits of easy linking with various information networks found on the internet
Security concerns will form a key aspect of the implementation process because the system will operate on a web-based platform. This will expose if to cyber attacks hence the need to invest in cyber security for the system.
In order to evaluate appropriately the suitability of the software choices, the system will call for the application of several measures. These measures include functionality, technology, software vendor and implementation. In addition, it will require taking into account maintenance, support and the total cost of ownership.
These elements necessary for the implementation of a web-based decision support system that will provide a solution to credit approval problems that banks face.
Reference List
Burnside, E. S. & Kahn, C. E., 2004. Artificial Intelligence Helps Provide Decision Support in Radiology. Web.
Evans, T., 2011. Improving Your Credit Rating. Web.
Fisher, C. R., 2008. 6 Key Software Evaluation Criteria. Web.
Hill Associates, 2002. Telecommunications: A Beginners Guide. Berkeley: McGraw-Hill/Osborne.
InetDaemon, 2011. Inet Daemon: Real training for Internet Professionals. Web.
Leotta, J., 2011. Credit Approval. Advameg, Inc. Web.
Lewis, M., 2010. Credit Rating: How it Works and How to Improve it.. Web.
Microsoft, 2011. TCP/IP Protocal Architecture. Web.
Oesterreichische Nationalbank, 2004. Guidelines on Credit Risk Management: Credit Approval Process and Credit Risk Management. Vienna: OeNB Printing Office.
Power, D. J., 2002. Decision Support Systems: Concepts and Resources for Managers. Westport CT: Greenwood Publishing Group.
Servigny, A. & Renault, O., 2004. Measuring and Managing Credit Risk. Illustrated ed. New York, NY: McGraw-Hill Professional.
The Computer Technology Documentation Project, 2011. Network Layers. Web.
Zopounidis, C. & Doumpos, M., 2011. Decision Support Systems. Web.