4 Common Mistakes Managers Make About Data Analytics

Link to the full text http://click.aliyun.com/m/22761/

The rhetoric about data and data analysis abounds. Companies continue to be warned about the right strategy for collecting and analyzing big data, and the consequences of not doing so. As The Wall Street Journal recently mentioned, companies have a huge treasure trove of customer data, but most of them don't know how to use it. This article will explore why. Working with companies trying to extract practical information from huge data sets, we identified four common mistakes managers make when using data.




◆ ◆ ◆

Mistake 1: Not understanding the concept of fusion The first challenge that



hinders the value of big data is compatibility and fusion. A major feature of big data is its diverse sources. However, if the data is not in the same form or difficult to integrate, the diversity of its sources will make it difficult for companies to cut expenses and create value for customers. For example, in a project we work with, the company has a wealth of data that records customer transaction volume and loyalty, as well as specialized online browsing behavior data, but rarely cross-retrieves the two types of data to determine a certain browsing behavior, i.e. A precursor to a deal. Faced with this challenge, companies have created "data lakes" to house large volumes of unstructured data. However, the data that these companies are able to leverage is currently in disarray and is just text, which means that when the data is just plain binary numbers, it is very difficult to store it in an orderly manner. It is even more difficult to integrate them from different sources.



◆ ◆ ◆

Mistake #2: Not Recognizing the Limits of Unstructured Data The second biggest challenge that

hinders the value of big data is its unstructured nature. Special progress has been made in the mining of textual data, whose context and technology bring insights similar to those of structured data, except that other forms of data such as video are still not easy to analyze. For example, despite state-of-the-art facial recognition software, authorities were still unable to identify the two suspects in the Boston Marathon bombing from a trove of video that was still processing photos of the suspects taken from different angles. .

While obtaining information from unstructured data is challenging, companies have made significant progress using this data to initially improve the speed and accuracy of analyzing existing data. For example, in oil and gas exploration, big data is used to optimize ongoing operations, as well as data analysis for seismic drilling. Although the data they use may increase in speed, variety, and volume, ultimately the data is used for the same purpose. In conclusion, the initial hope that unstructured data can be used to form new research hypotheses is untenable unless companies have "practiced" the expertise to use unstructured data to optimize an answer to a question.


◆ ◆ ◆

Mistake No. 3: Thinking that correlation analysis is significant

The third challenge—which we believe to be the most important factor hindering the value of big data—is that the large overlap in observational data makes it difficult to identify cause and effect relationships. Large-scale data sets often contain many similar or completely consistent information, which directly leads to erroneous association analysis and misleads managers' decision-making. Recently, The Economist pointed out that "in the era of big data, interrelationships often emerge on their own." The Sloan Management Review emphasized in its blog that although many companies have access to big data, these data are not "objective". ” because the problem is to extract actionable information from it. Likewise, association analysis performed by a typical machine learning algorithm used to analyze data does not necessarily provide causal analysis, and thus no actionable management advice. That said, the trick to making big data profitable is being able to move from merely observing correlations to correctly identifying which correlations are causal forms that can serve as the basis for strategic moves. To do this you have to look beyond big data.

Google Trends is a classic example of big data, which uses Google search terms to aggregate records. However, it also illustrates that data used only for association analysis is meaningless. At first, researchers said the data could be used to reflect the spread of the flu. Later, however, the researchers found that because the data represented the past, using the data only slightly improved coping behavior if the current situation was correlated with past patterns.

As a more specific example, suppose a shoe seller advertises to consumers who have visited its website. Analysis of the raw data suggests that consumers will be more willing to buy shoes when they see these ads. However, these consumers have already shown interest in the seller before seeing the ad, and are therefore more willing to transact than the average person. Does this ad work? Hard to say. In fact, big data here does not take into account causal inferences about the effectiveness of marketing communications. To know whether the ad is effective, the seller needs to conduct random testing or experiments to select a subset of consumers not to touch the ad. By comparing purchase rates between consumers who saw an ad and those who didn't, companies can determine whether seeing an ad makes consumers more likely to spend. In this case, value was created not primarily through data, but through designing, executing, and interpreting significant experiments.

This is an experiment, not an analysis of a large data set of observations to help companies understand whether a link is merely correlated or becomes a judgment call because it reflects an underlying causal relationship. While it is difficult for managers to improve profitability with even a petabyte of data that records consumer behavior, comparing customers who participated in a marketing campaign with those who did not—based on the results of the experiment—could allow marketing Personnel deduce whether the activity is profitable or not.

Carrying out field experiments, drawing correct conclusions, and taking appropriate countermeasures is not an easy task. But successful companies already have the ability to design, conduct important field trials, evaluate their results and take targeted action. It is this "trial-and-learn" environment, and the ability to act on an understanding and awareness of whether it can be generalized, that makes big data valuable.

However, such experiments do not necessarily require big data due to the diminishing returns of an increasing number of data samples. For example: Google revealed that it often uses a random sample of 0.1% of the available data for data analysis. Indeed, a recent article showed that big data is actually bad because "the bigger the database, the easier it is to support the assumptions you're making." In other words, because big data provides overlapping information, companies can The entire dataset can also get the same information from a thousandth of its dataset.


◆ ◆ ◆

Mistake #4: Underestimating the Skills Needs of the Workforce



Full Article Link http://click.aliyun.com/m/22761/