Quarterly Journal on Management
From the publishers of THE HINDU BUSINESS LINE
Vol. 2 :: Iss. 3 :: February 1999
Goldmeine!
K. Padmanabhan
Computers have traditionally been used to automate a number of routine tasks such as monitoring accounts, processing payroll or taking care of inventory management. Some recent developments, such as ERP, allow for the integration of a number of operations taking place throughout the organisation, and for the presentation of current data. This data often helps in operational decision making, such as ordering of raw materials or making arrangements for additional cash requirements.
However, data in these traditional systems cannot effectively be used in making tactical and strategic decisions. This is because the process used to arrive at such decisions cannot easily be put down in a simple formula or a neat computer program. All the same, making such decisions without adequate data is extremely difficult. Systems that address this problem, and which assist in human decision-making, have evolved over the past few years. The idea behind these systems is fairly simple. The data present in various forms in the computer is organised, processed and presented meaningfully to human beings in a way that simplifies decision-making. This type of system is called a Decision Support System.
Generally, Decision Support Systems are one of two kinds: data warehousing systems or data mining system. Data warehousing systems provide the basic or summarised data of a similar situation that has arisen in the past. Relatively little analysis is performed with the data, though analysis tools (also called Online Analytical Processing or OLAP tools) are often provided. Data mining, on the other hand, involves extensive analysis of data for detecting patterns.
To illustrate data warehousing, let us say that you wish to plan an advertising campaign. Typical decisions to be made would include the geographical limits of the campaign (regional, national), the media to be used, and the type and duration of the campaign.
Traditionally, these decisions are made based on relatively very little data. At the most, data related to the reach of the media (number of subscribers or viewers) and 'guesstimates' about the impact of the previous campaigns might be available. Much would depend on the individual judgement of the person in charge. While it is true that there is no substitute for the decision making capabilities of human beings, it is also true that these capabilities can be considerably enhanced by the provision of accurate and appropriate information.
Again, going back to the advertisement campaign, the areas where sales is poor, areas where the competition is strong, effectiveness of previous campaigns in terms of increase in sales per rupee spent, impact of the duration of the campaign on the effectiveness of the campaign and so on, would be the kind of information that would be very useful for the decision makers. This kind of data could even help cover up for the relative inexperience of the decision makers and could result in the company reaping rich benefits for the money spent.
Another example of data warehousing would be the information given in any sports event that is shown on television nowadays. Data on the previous encounters between the-two teams, data on the skill and effectiveness of individual players and so on are displayed on the screen. Though this data does not make the sports event totally predictable, it does provide a much better basis for estimating the team that is likely to win and what the score could be. This sort of information is readily provided by a data warehousing system.
Moving back into the realm of business, if such data about the past is presented, significant improvements in decision making could be made. To provide this sort of information, different types of data from possibly different application systems need to be brought together. Data warehousing systems are the answer and would help decision makers get this data easily and effectively.
But then, can't computer systems, especially when integrated into an ERP, routinely provide this information? Is the investment for building a data warehouse system really necessary? The answer is, the base data would be available in the operational systems currently being used. However, most existing systems would require a considerable amount of processing to provide certain specific types of information. For example, information about the profitability in a particular region, the advantages of using a particular medium etc., would require extensive processing and complex programming. The response time of the system would also be very slow so much so that this kind of processing would not even be attempted in the first place.
The overall technology of collecting data from multiple operational systems and combining them in a form that is easy to make decisions is called Data Warehousing. In general, data warehousing involves the collection of data from multiple sources both inside and outside the company, processing it into a form that is suitable for decision making, and presenting it in a manner that enables humans to understand the data easily.
One of the more important uses of data warehousing is found in the current trend towards customer oriented businesses. It is essential for a company to understand who are its most profitable customers, what types of products does a typical customer buy, and the customers who are likely to buy a specific product. Determining the profitability of various customers requires information on the various products bought by the customer and the profitability of each of these products. It is in such scenarios that data warehousing is very effective. Each product could have a different basis for determining its profitability, as the type of cost associated with the selling of each product would be very different. To look over the entire gamut of the different services or products offered to the customer, so as to obtain the total profitability of the customer requires the provision of a data warehouse.
Data warehouses can be created using data not only from a company's internal systems, but also data from external sources. For example, in India, CMIE provides information about various companies in various businesses and in various geographical locations over a given period of time. This data could be integrated into a company's data warehouse, and the performance of specific companies, over a given period of time, vis-a-vis other companies in its line of business and vis-A-vis other companies in the same area could be determined. The,advantages of such information to decision makers become quite obvious.
A data warehouse can be looked at as a multi dimensional database. For example, the accompanying table provides data on the purchases made by customers at various points of time. This could be represented in the form of a cube as shown in the chart.
Here, the length corresponds to various time periods such as April, May and so on, the breadth corresponds to various products such as A07 and so on, the height corresponds to various locations where the sale was made such as east, west and so on. Given this, you can see that different questions can be answered easily. If you want the total sales across all products in the eastern region, you just need to sum up across one particular slice of the cube. With the help of query tools, and a well-designed data warehouse, a manager can get data on all the past transactions in a form that can be readily used.
Quite often, managers require additional information. This could be in the form of a trend analysis and so on, which will enable the manager to do a considerable amount of `what if' analysis. For this purpose, the available data needs to be interpreted in different ways. Most data warehouses have a tool (called an Online Analytical Processing or OLAP tool), which enables the manager to do a number of statistical operations and also to indulge in "what if' analyses. Hence, if trends have been increasing at the rate of roughly 10% each year, you can find out what the market would be like in a particular year. All this can be obtained from one section of the cube.
Some of the decisions that are often taken with the help of data ware- housing are:
- Product lines to be continued (or discontinued), based on their profitability.
- Price sensitivity of dealers or customers.
- Effect of a bankruptcy motion on the price sensitivity.
- Management of defects and defectives. (Where are most of the complaints coming from?)
- Expected settlement amount for a particular claim.
- Suppliers giving high quality products (supplier management).
-
Awareness about the company within the industry, the business group or
geographical location.
Data mining, on the other hand, constitutes the discovery of patterns in the data. Human beings tend to make generalisations based on experience. Similarly we would like the computer to make such generalisations and provide us with some patterns from the details available. This would enable us to form multiple images (or models, as they are called) about the different characteristics of a given data. This is typically used to establish profiles among the customers to characterise them by their behaviour. We can identify customers who are willing to go in for risky products or customers who want only the safest investments and so on. This profiling of customers can then be used to market specific products to them.
There are different types of data mining. The oldest and most traditional form is through statistical analysis. More recent forms include what is called association, which helps in detecting patterns of combination items. This would help, for example, in the identification of product affinity. (If a customer prefers product A, he is also likely prefer product Q).
A number of data mining tools are available in the market now. However, data mining has not yet reached a high level of maturity and only an expert can work with the tool and interpret the results.
Overall, the computer, which, till recently, was considered a tool to churn out data and process it, can now be used to interpret data and understand some of the hidden patterns. This would help all of us understand the environment we live in and exploit that environment to the maximum extent possible to further our business.
K. Padmanabhan is Vice President - Technical, Tata Consultancy Services. He can be reached at 044 625 6445.
Contents