Big Data and BI: Failure Reports

In September, Moscow hosted the Big Data and BI Day conference, dedicated to the application of Big Data and BI (Business Intelligence) technologies in business and public administration. 5-7 years ago, such speeches resembled parades of victories and achievements. At present day, the professionals openly share their failures. The idea seems to be more and more true: for mastering the possibilities of new technologies for both businessmen and IT specialists, the experience of failure is perhaps more valuable than the experience of success.

Crypto miners and IoT overestimations

For example, the Mail.ru Cloud Solution identified the biggest and most unexpected problem — 60% of the traffic on their services that were provided to customers for free was from miners who used Cloud Solution's power to mine bitcoins. As a result, specialists had to create and set special filters to cut off spurious traffic.

And here is another one: the Internet of Things (IoT) promises and provides precision farming, smart homes and cities, robotic industries, etc. However, the sensors that form the basis of IoT, regardless of price and manufacturer, often give inaccurate readings, and today this is one of the most serious problems in the development of the industrial IoT. Businessmen should be more critical concerning successful cases cited as arguments when selling BI tools: what worked in one company is not a proof of success in another — with different staff, a different business, and a different IT landscape.

Managing risks and communications

What about risk management? It turns out that despite the developed mathematical apparatus and the software, currently, risk management is more a myth than a real practice. One of the reasons is that people who make decisions on introducing a particular risk management tool are often not familiar with the basics of mathematics and measurement theory. They rarely ask the question: where does the probability of the risk event comes from and how reliable is that? This is just one example of possible faults and there are many more to solve.

The analysts also discussed the problem of false correlations. While using the methods of working with Big Data, one can find out, for example, that the dynamics of suicides in the city clearly correlates with sales of cheese, and the number of degrees awarded by a doctor of science with sales of yogurt. However, the question of the causes of false correlations and how to identify them and exclude them from the conclusions of business analytics still remains unsolved.

The problem of communications is one of the hottest: chief officers, turning to the analytics department, cannot clearly formulate their questions. What to do with the pains that the “internal customers’ feel, but cannot express? How to translate their vague wishes into the language of clearly defined tasks? How this question could be solved when the demand for Big Data analytics will skyrocket in all departments of the corporation?

The most painful issue

Today, the mechanics mostly is the following:

• a platform is selected (with its ability or impossibility to work with certain metrics),
• then BI software is selected,
• then the question arises of a clear and unambiguous formulation of business tasks that should be solved by tools for working with Big Data.

But the common sense requires the steps to be taken in the opposite way: from business tasks to metrics, then to processing tools, and only then to choosing a platform on which these tools can be deployed.

The terminological chaos in publications that has developed around the words “Big Data”, “BI”, “AI” and “ML” is one of the problems that is yet to be solved. The discussion showed that there is no uniform understanding among its participants — those very professionals who make Big Data, BI, AI and ML with their own hands.