{"id":97159,"date":"2025-11-05T08:00:41","date_gmt":"2025-11-05T11:00:41","guid":{"rendered":"https:\/\/abes.org.br\/?p=97159"},"modified":"2025-10-31T10:08:18","modified_gmt":"2025-10-31T13:08:18","slug":"qualidade-de-dados-a-base-para-decisoes-inteligentes-com-ia","status":"publish","type":"post","link":"https:\/\/abes.org.br\/en\/qualidade-de-dados-a-base-para-decisoes-inteligentes-com-ia\/","title":{"rendered":"Data quality: the foundation for intelligent decisions with AI."},"content":{"rendered":"

Incomplete, inaccurate, or outdated data compromises the value of AI \u2014 quality needs to be a strategic priority, not just a technical one.<\/em><\/p>\n

\"\"*By C\u00e9sar Ripari<\/em><\/p>\n

We live in a world where 90% of existing data has been generated in the last two years. This volume continues to grow \u2013 it is estimated that the rate is around 2.5 quintillion bytes per day. To put this number into a more approximate scale, we are talking about 2,500 PB (Petabytes) or 2.5 EB (Exabytes). This would be something like 500 trillion high-quality photos or 68 years of continuous satellite observation data around the earth.<\/p>\n

However, good decisions don't stem from bad data. Assertive AI isn't the result of bad data. When an organization works with incomplete or inaccurate data, outdated information, inconsistencies between systems, and restricted\/complex access \u2013 among many other challenges \u2013 the cost is high: rework, lost revenue, regulatory risk, inefficient campaigns, incorrect inventory, avoidable churn, and guaranteed losses. In a\u00a0I estimated<\/a>\u00a0According to the Gartner Institute, the average cost of poor data quality is estimated at US$12.9 million per year per organization.<\/p>\n

For companies, this scenario represents an extraordinary opportunity. Instead of making decisions based solely on intuition or recent experiences, it's possible to base them on proven facts, analyzed and validated based on customer behavior and preferences over time. The ability to collect, integrate, analyze, and extract insights from data has become an essential competitive differentiator. Companies in a wide variety of sectors \u2013 from retail and manufacturing to telecommunications and especially entertainment \u2013 have already realized the power of data. However, the quality of insights depends directly on the quality of the data. It is essential to ensure that reliable data is available to everyone who needs it \u2013 respecting privacy and governance \u2013 and going beyond the IT team, involving data scientists, analysts, and decision-makers.<\/p>\n

In other words, data quality is not a technical issue\u2014it's a strategic agenda for efficiency, productivity, and governance.<\/p>\n

Bad data can come from any area of the company, from sales to engineering. But following certain principles can make the data ready and reliable for AI:<\/p>\n

1. Data should be diverse (coming from a wide variety of sources) to reduce biases in AI systems.<\/strong><\/p>\n

How much does a dataset include samples from different origins, contexts, categories, and distinct characteristics so that it can reflect reality, avoiding biases? For example, in a financial institution, preventing a loan from being denied due to a person's place of residence.<\/p>\n

2. Data must be updated to enable accurate and well-founded predictions.<\/strong><\/p>\n

How up-to-date is the data, considering the time interval since its collection, necessary transformations, and availability to the user? Some data does not change over time \u2013 for example, a person's date of birth. However, other data needs to be constantly updated, and preferably close to real-time \u2013 for example, financial market data or e-commerce inventory \u2013 since they depend directly on constant and rapid changes. There is also data that allows for greater tolerance in its updating, which can vary from hours to days. The important thing is that the data update time is known and validated by the data consumers throughout the organization, thus ensuring that decision-making is as accurate as possible.<\/p>\n

The degree to which a piece of information corresponds to the actual value or state of an object, event, or condition. Data accuracy considers three aspects:<\/p>\n