|
4Thought delivers insight on how today’s decisions will impact tomorrow’s direction 4Thought is a predictive modelling tool offering the following:
The following case studies give more detail on how 4Thought has been used for both time series and profiling analysis. The Modelling Challenge
Businesses face an ever growing need to understand the way they work and the way their markets behave. To be able to make more informed decisions and deliver products that offer greater value, managers must know their business and their customers better than their competitors do. The key to this understanding is the collection and analysis of information. Neither of these tasks is easy, but in the information age, they’re getting easier. Computer technology has allowed data to be collected readily and consistently, and has opened the door to new analysis techniques which can expose deeper insights than have ever been possible with traditional statistical techniques. New insights turn into business value as soon as they are applied:
The link between information and real business value is the ability to analyse data to reach a conclusion on which decisions can be made. With new mathematical modelling techniques, the link is stronger and easier to build than ever before. Top of the list of new techniques is neural network technology. Neural Network Technology 
The original idea behind neural network technology is to try to understand how the human brain works - how it can chew over a mess of data from the external world, how it recognises coherent patterns, and how it associates these patterns into chains of cause and effect. This field of study has been progressing for over forty years, and has some way to go yet before it will be able to synthesise anything exhibiting the slightest hint of intelligence. In the early eighties however, a small breakthrough occurred with the invention of the multilayer perceptron (MLP). It turned out that the MLP exhibited a number of interesting properties which give it real practical value as a general mathematical modelling tool. These are:
Consequently, neural networks are now being used quite extensively in scientific applications such as robot vision, speech recognition and control systems. An important application area for neural networks is in business modelling and forecasting. This is because the datasets associated with businesses and markets are frequently noisy and co-linear, and very often are only available in small quantities. The Modelling Process There are four stages to the modelling process:
Using business models for forecasting lets you make predictions based on what actually causes demand to rise and fall. Causal-based forecasts are therefore much more reliable and realistic than other methods of forecasting. Data modelling is the technique which moves us from "just guessing" to being able to predict an outcome with confidence. Small amounts of data or "missing" data in files doesn't have to stop the modelling process—it just means that the models will be of poorer quality. Larger amounts of data should improve the quality of the model and thus its predictive ability. There will always be other factors affecting a response, and a person will never be able to include all of them into an analysis. As long as the significant factors have been included, and common sense is there to evaluate the results, then data modelling can be a formidable tool for analysts to investigate their business. 4Thought - Neural Network Business Modelling and Forecasting 4Thought is a business modelling tool. Extremely versatile, it is used in hundreds of different applications without requiring any modification. It is also flexible and adaptable. Users can approach their analysis in the manner best suited to the problem and data available, without restrictions from the software. Working extremely quickly, 4Thought gives users necessary interactivity as they chase new business discoveries. Two of the many applications that use 4Thought are described below. Demand Forecasting Through Time Building models of demand over any regular time step, (hourly, weekly, monthly, annually, or another specified period) gives the user a powerful forecasting tool, and highly sophisticated market intelligence. Looking at historical data, 4Thought’s neural network finds the relationships between the demand volumes, and factors you believe might influence them, for example, season, time, weather, competitor activity, advertising and promotions, and so on. This can be conducted at any level, usually analysing daily or weekly data in the first instance. Once the inter-relationships between all of the different factors and volume is established, the resulting model can be used for both producing realistic forecasts of known predictive ability, and to look at the isolated effect of each individual factor. Taking cross-sections out of the model provides a host of valuable marketing information such as the effectiveness of advertising, the underlying seasonality and the price sensitivity of a product. In a similar fashion, the influence of a combination of different factors on other variables such as costs can be monitored, thus yielding information on the effect of key profit drivers. Customer and Direct Mail Profiling The principles are the same as time series modelling in that 4Thought is used to find the inter-relationships between one variable and many different factors. Usually, only a small sample of a database is analysed. For direct mail, this will be a set of prospects mailed with a particular mailer, and for customer profiling, a mixture of actual customers and prospects. In either case, a model is built of whether or not the prospect is a respondent/customer, using those factors associated with each record, (typically geo-demographic details, and type and number of previous purchases). This kind of model is particularly useful. Taking cross-sections through it allows you to get a better idea of your best kind of prospects, and by applying the model to the rest of your database gives a "score" for the prospects’ likelihood to respond. Similar models can be used to identify key drivers, quantify the effect of change and impact of local advertising, discover regional price sensitivities, or benchmark performance of different channels and media, while taking regional differences into account. As with time series analysis, the dependent variable (that which you are measuring), and independent variables (those factors you believe might influence the dependent variable), can be anything you choose. Other examples include retail branch profiling, benchmarking, measuring return on investment, and calculating the value of a particular market. Case Studies 4Thought is not only used for demand forecasting and profiling, but also benchmarking activities (e.g. measuring efficiency), and a host of industry specific applications. In every case the basics are the same—seeing how a series of factors influence an outcome. Naturally, precise details vary according to the specific application, the available data, and to some extent the experience of the modeller. Irrespective of the application, three criteria must be met: Is there a clear financial benefit to be gained by building a model? Is there sufficient data available, and is it of a reasonable quality? Is the resource (manpower and time) available? Unless the answer is yes in each case, you will be wasting your time trying to move the project forward. The following case studies give more detail on how 4Thought has been used for both time series and profiling analysis.
|
