Data analytics is an applied field of research that has two main components - mathematics and development. It would seem that unless new mathematical theories or programming languages are invented, there should be no radical changes in this field. But this is not the case at all. Data analytics is evolving, and new ways of extracting knowledge from data, new approaches and methodologies are emerging. We have tried to structure the main trends and are ready to introduce you to them.
Development and application of generative AI
Generative adversarial networks and ready-to-go architectures like ChatGPT methodologically simplify the process of interaction with high technologies. The principle of «do not know how to solve the problem - solve with the help of a neurogrid» becomes more and more applicable to data analytics. Indeed, it is great to have a universal and scalable tool in the arsenal of business, but it must be remembered that the thoughtless application of technology is fraught with risks. So see the following trend
Model risk management
I will allow myself the freedom to combine in model risks two components - mathematical and ethical. Today, many prominent businessmen (especially those at the forefront of technological development like Bill Gates or Elon Musk) talk about the need to control the ethical issues of artificial intelligence development. Of course, this is important, but while we are limiting the existence of the neural network to digital space, as long as the mode of its operation is entirely in the hands of the operator, and the scope of application is clearly limited to the task, in my view, the ethical risks are a little overestimated, while mathematical risks, on the contrary, are underestimated by companies implementing models in processes. A model is a potential threat not to a person, but above all to the correctness of processes and business metrics. This is a tool of qualitative positive influence on business, but in case of errors - the source of losses. And expressed in specific currencies. Therefore, when introducing AI, it is necessary to introduce systems of model risk management, assess the probability of occurrence of error, the degree of its impact on business, and select strategies to minimize the impact of errors on business and financial results.
Again, the trend associated with the previous item. To understand how to minimize the impact of model errors on company processes, you need to understand how the model works. At least at the reporting level. Better at the level of logic.
Unified data infrastructure
DWH, modeling systems, decision-making systems, monitoring and risk management, reporting - all of these systems should not just exist in the company circuit or in accessible clouds, these systems should be integrated into a single circuit. Systems must share both the raw data and the company’s strategy. Having changed the settings in the decision-making system, the risk monitoring system should rebuild its work, risk management should directly adjust the settings of the decision-making system, all changes should be displayed in DWH and monitoring, etc. The effect of it is analytic date synergistic only when the data cycle is closed.
Maximum data utilization
The data must work. If the data science staff does not have time to cover the entire accumulated data set, change the processes. Think you can automate. The green trend works not only in ocean cleaning but also in extracting information from raw data.
With maximum data recovery and efficient data management, you can move on to data sharing. And both internal and external exchange
It is necessary to understand the mechanism of data flows, so a single data infrastructure must have a metadata block with the description of flows. This will simplify development and in general very convenient from the point of catching errors.
Lowering the Data Analytics Entry Threshold
Data analytics can be done without coding. Modern platforms allow to implement in a single infrastructure of work with data platform, which in the web interface allows you to configure the processing of data of any complexity. When choosing a platform, it is good to pay attention to the possibility of uploading the results in the form of a code. Better on understandable date analysts programming language like Python. In this way, not only former data users will be able to perform the development function, but in the future developers will be able to improve, customize the built processes.
People make key decisions, data analytics helps but does not replace people
In the end, not even a trend, and common sense. Any data tool is a tool in the hands of a person, a specialist, whom you trust. No platform replaces a specialist, it helps him to decide more quickly and correctly based on existing opportunities and risks.