December 18, 2012

Why OLAP Always Mean Cube Modeling, Secrets of OLAP You Never Know

The original post URL:

When talking about OLAP, people always means cube modeling. Most of us will wondering, if we don't have modeling, how can we go on data analytics? Why OLAP always means Cube-modeling? Can't we just go on business computing without cube-modeling? Maybe there are some secrets you don't know about OLAP.

The typical usage of traditional OLAP tool is always to build module first and then analyze data as if it was a convention. Is the modeling unavoidable? Why we build model? Why we spend time and efforts in this stage instead of analyzing the data directly? Why the performance is still an excuse to put off the innovation despite the hundred-fold increases in hardware performance throughout all these 20 years? What is the root cause?

For the countless questions, the answer is only one. It discloses the secret of OLAP you never knew: The traditional OLAP suffers from insufficient computational capability, and even the limited computational capabilities available are gained at the cost of modeling. Modeling is the means to make up for the product drawback by forcing customers to pay for it.

This discussion here will expose this secret by pointing out 3 facts: First, modeling is not necessary at all. Second, modeling brings about disaster. Third, the real look of OLAP.

The discussion here will start with an example. A retailing business needs to analyze the results of new sales policy implemented in the latest 3 months, and develop the corresponding actions.

Firstly, it is impossible to design an all-roundly complete analystics model beforehand. For example, the new policy may mainly affect the sales achievement, team relation,or even the sales force management. The sales achievement may probably go through a host of possibilities of sudden rise and drop. The sudden rise of sales achievement may result from either the large orders placed with short lead time, or the increase in the volume of orders. There are many branches and fluctuations. Every time, users have to judge momentarily on the unpredictable tendency of branches, which means it is impossible for them to design a clear route to analyze.

Secondly, the model restricts the freedom of analyzers. Modeling means analyzers can only take actions in a stipulated scope. For example, the analysis cannot be completed if you want to import the data of fellow retailers for comparison during analyzing, because the temporary data import is not allowed for models...For another example, without the slicing, rotating, drilling, and other traditional OLAP functions, it is impossible to compute the big clients accounting for 60% of the total profits. The original intention of OLAP is the arbitrary interactive computation. The typical procedure is to firstly make an assumption on the obscure goal, secondly, verify or falsify these assumptions, then correct the assumption continually, and ultimately reach the right decision. The traditional OLAP does not support the arbitrary interactive computation on data, and its limited function hinders the freedom of analyzers, and therefore cannot make the truly valuable decision.

At last, modeling cannot hold the changing requirements. The commercial opportunities are evanescent. Facing the ever changing demands, OLAP must provide the right analysis decision in the shortest time. For example, the continual decreasing of order volumes draws the attention from executives. It is pressing to determine whether there is a relation between the order decreasing and the new policy. When the traditional OLAP encounters such new requirements, users can only request the technical experts and business specialists to design the new models, and then keep on adapting the models to the practical business, sometimes the model has to be reworked when it turns out to be incapable in the analysis stage. During this period, a great deal of time, money, human resources, and physical resource would be spent. When all analyses have been done, the business opportunities have slipped away.

Therefore, the model is not a necessity. On the contrary, modeling makes OLAP lost customers and market. According to findings from Google, since 2004, attentions on OLAP drop by 85 percent. In fact, most users just take OLAP as an expensive presentation tools.

The market is thirsty for OLAP product with improved computational capabilities, that is, a brand new product not requiring modeling, capable to carry out direct analysis on the history data, and provide the timely decision on the arbitrary interactive computation. It must be featured by:

Support the friendly interaction. It will lower the requirements on technical background to ensure that even the normal business personnel can grasp it easily. The analysis result in any step must be always clear and visible. According to the intermediate result, users can manipulate the data straightforwardly and intuitively on menu. OLAP is intended to serve the purpose of business decision for the business specialist users. For the business specialists with limited IT experience, user-friendly interaction is just what they need.

Free multi-step computation. The OLAP tools must be able to arbitrarily decompose the complex problem into several simple steps. Seek the solution to a complex problem by solving lots of simple problems. The step-by-step computation is the key to solve the complex problems. Traditional OLAP tools are confined by the model and unable to solve the complex problem, not to mention the arbitrary operations. For example, the problem of "Compare the product sales fluctuations in the half year" can be decomposed into these simple steps: Filter out the order data in the half year; Group the data by month; Summarize the sales in each year; Perform the year-over-year monthly comparison.

Support the structural data. In the business system, a huge amount of structural data is scattered here and there, for example, in database, or Excel and text file. The ideal OLAP should be capable to provide the direct support for the massive computation on the structural data, such as grouping, summarizing, associating, and filtering. Although SQL can support the structural data well, SQL also requires a relatively higher technical background on users. To implement a bit more complex computation, users will have to write a great deal of scripts hard to write and maintain. The ideal product is the one that is comparable to SQL in computational capabilities with relatively lower requirements on technical background, so that the business specialists can use it easily.

Optimize the business computing. Unlike the scientific computing, the business computing often requires the thinking from business perspective. For example "which products among the top 10 best-sellers last year still remain the top 10 best-sellers this year". In business computing, users shall seek the interactive set of the 2 objects intuitively, instead of writing a large section of SQL/VB/JAVA scripts, and the way to represent the intersection set between objects easily using the business terms. The similar examples also include the associative relations to represent the multi-level relation, for example, the problem "for the sales managers responsible for the products among the top 10 best-sellers, how much contribution their clients made to it". In addition, the complex problem of commercial computing is always related to orders: ranking, link relative ratio, year-over-year comparison, etc., for example, "products whose sales volume keep rising in consecutive three months". All these problems need representing in a concise and easy-to-understand way.

The esProc/esCalc family products from esProc are just the desired software, of which, esProc is good at multi-step computation and solving the complex problems; esCalc relies heavily on the friendly interaction and ideal for the business personnel. They have the advantages below:

Modeling free - esProc/esCalc does not require modeling because its professional computational capability enables its users to analyze on business data immediately to provide the decision support promptly, and seize the evanescent business opportunities ultimately.

Arbitrary computation - esProc/esCalc provides the arbitrary computational capability to implement the truly arbitrary OLAP computation. The enterprise can thus be provided with the valuable data for decision-making. It boasts the friendly interaction, arbitrary multi-step computation mechanism, and optimization on business computation, such as the complete set computation, ordered computation, and object references.

Relatively low technical standard - esCalc/esProc has lowered the technical requirements on the analyzer and designed especially for the business specialists. Business specialist can complete the whole set of OLAP work independently, such as installing and deploying, determining analysis goal, decomposing task, retrieving data, verifying the conclusion, presenting the conclusions, and print & export.

The computational capability of esCalc/esProc is worth looking forward to in the hope that traditional OLAP tools could be inspired to confront their drawbacks. Let them commit to build the OLAP standard of the next generation collaboratively.

Related Articles from Raqsoft:

Business Intelligence Suppliers: Are You Ready for 2013?

Script Computation of esProc Optimizes Analysts' Benefits

Various Data Environments Support of esProc Makes Statistical Computing more Flexible

No comments:

Post a Comment