According to the informe The State of Data Infrastructure Sustainability de Hitachi Vantarainfrastructure and data management are vital to guarantee the quality that drives AI projects with positive results in companies. Furthermore, for companies, main concerns when implementing projects with AI are cybersecurity (33%) and the data quality to train AI. Specifically, its lack, a problem for 32% of those surveyed when preparing the report.
Despite this, there are few companies that are taking measures to improve data quality, which limits the success of their Artificial Intelligence projects. Indeed, for AI projects to have favorable results, there are two key factors: the use of high-quality data (35%) and adequate project management (39%). In both cases it is very important to have quality and well-managed information, as well as trained professionals in charge of the projects.
But the necessary data is only in the required place, and when it is needed, in 30% of the cases. Furthermore, AI models developed by companies only achieve accuracy 32% of the time. Despite this, only 28% of companies are working on improving data quality with accurate training of AI models. 23% do not review the quality of the data used, and 39% do not label it appropriately, something essential to improve their governance.
The report also shows that many companies do not have an analysis of parameters such as ROI or sustainability. In fact, 65% do not consider the latter a priority when implementing their AI plans. Another 63% also do not prioritize ROI when implementing solutions. On the other hand, 86% of companies in Spain are focused on the development of large language models instead of opting for smaller and specialized models. The focus on large models is noticeably higher than the European average, set at 64%.
Compared to the low priority of these two parameters, there are others such as security (46%) and speed (45%), which receive more attention from companies. In this context, security is a priority due to the risks it entails. 75% of organizations recognize that losing a significant amount of data could have catastrophic effects on their operations. Additionally, 79% of respondents are concerned about the use of AI to provide advanced tools to hackers.
Regarding data, although 33% of entities recognize that it is one of the main factors for AI to be successful, many do not have the necessary infrastructure to support consistent standards related to its quality.
79% of companies test and adjust their AI solutions in real time and without using controlled environments. This increases the risk of vulnerabilities and security flaws. Only 7% claim to use sandboxes to experiment with AI, raising questions about the potential for security breaches and flawed results from using unreliable data.
Furthermore, as companies advance AI initiatives, those surveyed for the report recognize that third-party support is essential to work in critical areas. Like hardware, whose effectiveness depends on its security, 24/7 availability and efficiency to meet sustainability goals. 20% of respondents believe that help is necessary to develop scalable and future-proof hardware solutions.
Besides, 29% of IT managers need external help with ROT data storage (redundant, obsolete or trivial), and 27% also need support for the development of AI models and data virtualization solutions. Regarding the lack of qualifications, another important obstacle to the success of AI in companies, a 60% of IT managers develop their AI skills through experimentation and 38% through self-study.
