A mind of its own: Advanced TMN will speed the service creation process through the use of neural networks and fuzzy logic

As the distinction between local and long-distance markets rapidly disappears, carriers of all stripes are battling for a piece of what is fast becoming a more generic "telecom services" market. This large, ultra-competitive arena demands that carriers offer much more than simple, attractively priced services such as POTS if they hope to retain today's customers-and win more customers tomorrow.

To satisfy end users' increasingly sophisticated demands and stay ahead of the competition, carriers need a business solution infrastructure that can deliver complex, high-quality services quickly. Until recently, many service providers believed they had exactly that, in the form of networks based on the Telecommunications Management Network architecture.

With their ability to combine customer care and automatic service provisioning, TMN-compliant networks have become the industry's preferred solution platform. However, TMN networks may not be sufficient in the ongoing battle for customers because of their rigidity and associated costs.

The biggest share of the market will likely go to those carriers that are first out of the gate with cost-effective services tailored to each customer's unique requirements. Such speed to market may call for a more flexible, smarter version of today's TMN implementation.

Emerging competitive pressures ultimately will demand a platform that can "learn" what is necessary to rapidly provision exactly what each customer wants, and when and where that customer wants it. Moving TMN standards to the next evolutionary stage may involve introducing neural objects and fuzzy logic that will enable computers to learn-and make decisions.

TMN limitations Today's TMN standards support a range of network management activities, including the planning, operation, administration, maintenance and provisioning of the network and related services. TMN's standardized interfaces also define the interconnection of various operations support systems (OSSs) and equipment to supply required network management information.

The major limitation of current TMN solutions is that they cannot define abstract service models accurately. A TMN solution focuses on the data of managed objects, rather than on the behavior of those objects. As a result, TMN's inherent rigidity forces carriers to configure or model every new service or service application.

For example, suppose that one customer orders POTS with remote call forwarding, where copper facilities are available, while another orders a packaged product-POTS with three-way conferencing and videoconferencing over ISDN, where cable facilities exist. The carrier's service modeling effort incurs significant costs in terms of time to market and technical expertise.

When one multiplies those costs-not to mention the data requirements-across the carrier's entire customer base, the magnitude of the problem becomes obvious.

One solution to this industrywide challenge is to accelerate the evolution of the tool used-or to establish a new level of core intelligence in TMN. The idea is to eliminate the time and expense of modeling each application and, instead, enable the platform to "think" on its own and respond to each customer's unique service needs. An aspect of artificial intelligence, fuzzy logic will help enable computers to recognize and respond to patterns, rather than to fixed facts.

Software developers are using fuzzy logic to build neural networks, systems that mimic the behavior of biological neurons or nerve cells. These neural networks consist of interconnected processing elements, or neural objects, that collectively learn to recognize patterns, data or processes, just as the human brain does.

By incorporating neural objects and fuzzy logic, a TMN platform can evaluate a set of factors and devise one conclusion-and evaluate that same set of factors differently and come up with a different conclusion. Essentially, the platform "thinks" and acts by recognizing patterns and applying experiences.

Take the airline industry as an example. Suppose an individual wants to fly from Atlanta to Chicago; the airline books her on a 747. Another passenger wants to fly from Atlanta to Houston; the airline books him on an M80 aircraft. Who or what decided a 747 should handle the Atlanta-Chicago flight and an M80 should handle the Atlanta-Houston flight?

Obviously, the airline industry considers factors that are invisible to both travelers: the average number of daily passengers flying from Atlanta to each of the other two cities, the amount and cost of jet fuel to cover the geographic distances involved, the rate at which each type of plane burns fuel at certain altitudes and other factors. People and computers evaluated two versions of the same customer requirement-transport-against numerous factors and came up with two related but separate conclusions-a 747 and an M80.

Returning to the telecom industry, consider the previous example of one customer ordering POTS with remote call forwarding, while another orders a packaged product-POTS with three-way conferencing and videoconferencing over ISDN. Suppose the carrier serving those two customers has an advanced TMN platform that incorporates neural objects and fuzzy logic. Rather than using skilled technicians to model configurations, the carrier can let the smart platform handle the task.

Basically, the advanced TMN network recognizes that both customers want the same thing-POTS-but it also understands that different factors are associated with each-remote call forwarding vs. three-way conferencing and videoconferencing over ISDN. The platform proceeds to act on the recognized patterns by assembling and provisioning each specific service configuration.

By using and re-using one model, the advanced TMN platform can enable the carrier to satisfy two sets of customer requirements faster and for less money than with today's TMN platform. In turn, that can give the carrier an edge over marketplace rivals.

Evolutionary steps Today, every managed object within the TMN architecture is part of a fixed relationship-either a process-to-process relationship or a data-to-data relationship. This explains TMN's inherent rigidity, which forces carriers to create a separate model for each new service.

To make TMN more flexible and smarter, the first requirement is to expand the TMN logical layered architecture to include a neural abstract level or layer, neural objects and fuzzy logic functions.

A neural abstract level/layer moves TMN's internal architecture to the next evolutionary stage-a true service-modeling environment that includes object awareness, experience and contextual decision-making. Neural objects are core software elements with "data experiences" that are capable of defining new service models.

Fuzzy logic functions provide a framework for fluid relationships in which the neural object can make decisions based on previous data and processing experiences.

These functions can be integrated easily into the functional domain in TMN's service management layer (Figure 1).

By accommodating a learning element and accepting non-fixed relationships, advanced TMN solutions offer several benefits, including the ability to model service concepts instead of customer data, stable definitions of service models and reusable service models.

A service featuring POTS with remote call-forwarding, using the same provisioning rules, will learn via fuzzy logic to offer POTS with three-way conferencing and videoconferencing over ISDN-without having to develop a new model.

Neural technology would be interspersed among the function blocks of current TMN standards (Figure 2).

The evolution of TMN to a more flexible architecture, one that includes a learning element, will require ongoing cooperation among all industry participants-standards-setting bodies, equipment manufacturers, software developers and service providers, including the latter's marketing departments. Cooperation between software developers and marketing people will be particularly important. Competitive success in the all-encompassing telecom services market will demand the rapid delivery of complex services across numerous technologies and industry sectors.

In developing new services that support customers' existing or planned applications, the marketing department must build a solid business case that takes into account several factors, including how many times, how quickly and in what ways the carrier will have to modify this new service.

By using TMN solutions that incorporate neural objects and fuzzy logic, the carrier can provide customized service packages quickly and cost-effectively based on customers' reactions.

To ensure TMN's ongoing evolution and realize the benefits it promises, service providers will have to modify some of their own behaviors as well. One of the most significant challenges confronting advanced TMN is the industry's traditional service deployment strategy.

When a carrier develops a new service offering, the process is a lengthy one, characterized by fits and starts. Chances are, the carrier first gathers all the collected research about customer requirements, expectations and wish lists and turns that data over to a management information systems department. There, platform architects and software programmers are responsible for creating a new product or service.

Once a new offering is created, the carrier does not launch it immediately. Instead, the company picks a few friendly customers and persuades them to try the new service and provide feedback. The management information system department then tweaks the service accordingly.

Generally, the carrier lines up a few focus groups to gauge response to the proposed service and then tweaks it some more. After that, the carrier formally announces the service and sets a date-usually sometime in the next quarter or so-when it will be commercially available.

In the new ultra-competitive environment, in which speed to market is critical to success, a carrier should be able to turn the appropriate data over to a TMN software platform that will create the corresponding product or service. The carrier should be able to launch the product or service immediately into the mass market.

With neural objects and fuzzy logic, expedited service deployment may not be a luxury but a competitive necessity. After all, more than one carrier will have advanced TMN platforms.

Another necessary change for the industry involves the way carriers calculate the return on investment on planned service offerings. Traditionally, such strategies have meant considering the costs of the underlying research and development, as well as the costs of assembling, manufacturing and marketing the offering. Carriers establish a price designed to recoup those costs and make a profit.

Advanced TMN solutions also will expedite the process of calculating return on investment. By turning that process over to a "thinking" entity that has no emotions, no political ax to grind, no departmental budget to protect and no turf to guard, carriers should receive the most realistic answers to their pricing questions.

Finally, the competitive realities of an advanced TMN environment will call on industry executives at the highest levels to articulate a vision for this brave new world. Understandably, any decision-making process that involves the mass market and millions of dollars always has been a cautious, relatively slow one. The products and services supported by TMN neural objects and fuzzy logic will ensure that no carrier will be able to afford that kind of thinking.

With its ServiceCoordinator network management system, Siemens Telecom Networks is one supplier moving in the direction of advanced TMN. The product includes a suite of software applications that provides cable, voice/data/video and wireless service providers a flexible platform for launching new revenue-generating services by leveraging their existing network investment.

The next evolutionary stage of TMN has the potential to revolutionize the telecom services industry and all the customers who depend on those services in their professional and personal lives.

By throwing into the advanced TMN platform a mix of customers' individual requirements and a broad range of knowledge and experience, a carrier can quickly satisfy customer needs and emerge a winner in the new telecom services market.

Another area where neural computing and artificial intelligence may play a role is in data mining.

Marketers use the term "data mining" to describe the process of analyzing a company's internal data for customer profiling and targeting. Until recently, companies relied on supervised learning for their data mining needs, but now they are beginning to use new methods.

Supervised learning-a form of neural computing, also known as neural scoring-involves building regression models and ranking each subject according to his or her likelihood of exhibiting a certain behavior-ordering a particular service, for example-based on historical results. As a complement to neural scoring, companies are beginning to look at neural clustering.

Unlike neural scoring, neural clustering is a form of unsupervised learning. An offshoot of the study of artificial intelligence and neural networks, neural clustering enables users to simply issue the command "self-organize," and the data automatically forms itself into high-, medium- and low-occupancy groups. A high-occupancy group has a large number of subjects with a similar profile.

By studying these clusters, organizations can help answer questions such as "Who are my best telecom customers?" "Who are my target telecom customers for a specific product?" and "What telecom service has generated the most revenue?"

Structured query language, traditionally used in data analysis, approaches such questions by providing a list of records that meet the user's select criteria. Neural clustering can be more effective in answering these questions by identifying each population based on what most distinguishes that group, rather than making the user guess which criteria might be relevant.

For example, the top markets for a certain telecom product line, such as call waiting or caller ID, may be characterized by several distinct clusters that range from young, unmarried apartment dwellers to older women who have been long-time customers.

The input for neural clustering tools is data in its simplest form-flat files of fixed-length records. The user is able to cluster or self-organize the data on all fields, some fields or only one field. Clustering on one field simply allows data to group itself based on the distribution of one attribute. The tools allow the user to specify a maximum number of groupings-but the total cluster space will not be used if it is unnecessary.

Although the underlying principles behind neural clustering are complex, users need only a basic understanding of neural computing to feel comfortable with neural clustering tools and readily explain the concept behind forming data into clusters.

A key concept is initialization, which assigns a random profile to each pre-defined cluster. In clustering customer data, each cluster initially will be assigned random demographic values such as age and gender, and random internal characteristics such as the services one buys and average monthly revenue.

When a real record is processed, all clusters compete-the "winner" being that cluster with the data profile closest to the profile of that record. The values of the winning cluster and its neighbor clusters are all altered toward the values of the real example.

As the file is processed through several passes, neighborhoods and clusters form. With the cluster map, marketers can gain insights without having to formulate specific queries.

Effective use of neural clustering may require changes in organizations and processes. Statisticians may be freed from tasks that can be achieved with neural computing and, instead, may spend more time consulting with less technical users who will easily construct complex models. Information technology organizations may need to accelerate the competencies they are building in neural computing and be prepared to feed relevant preprocessed flat files for clustering on a scheduled basis.

Neural scientists have delivered a robust capability. Together, businesses and information technology communities have the opportunity to build the applications enabled by this new technology.