Data analytics can be extremely powerful when it comes to helping business leaders make strategic decisions. Optimizing all areas of your company is easier when you can derive insights from key metrics. Now that all the market leaders in the BI software space have integrated AI-powered conversational interfaces, there’s even more interest in giving all line-of-business team members access. But while the software capabilities make adoption easy, many organizations need to get their data in order before they can derive maximum value from it.

Avi Perez is the CTO and co-founder of Pyramid Analytics, a leading BI platform that rolled out a natural language decision intelligence solution called “GenBI” last year. But even in the age of AI analytics, as he recently told DATAcated’s Kate Strachnyi, “You’ll be shocked when I tell you that the same data problems exist – it’s just at a different scale. We still have the problems of ‘garbage in, garbage out.’ We still have the headache of, ‘Is the data correct? I’ve got a copy of the data, you’ve got a copy of the data.’”

As the ecosystem has matured, “The software and the industry has moved on a huge amount, and this is where the GenAI frameworks are going to move it even further,” he continued. “But I think the problems have metastasized, and they’re going to keep doing so, because we’re going to collect even more data in the future around everything we’re already doing.” 

A recent AvePoint study found that the top two concerns business leaders have about implementing AI in their organizations are data privacy and security (71%) and the quality and categorization of internal data (61%).

So what do we need to do to avoid issues like these so we can actually benefit from AI-enhanced BI? Here’s how to get your databases, processes and culture ready.

Break Down Data Silos

AI thrives on context. The more it can correlate data across systems and departments, the more accurate and actionable its insights become. So, breaking down the data silos that may exist between the tools and business units in the organization is essential if you want AI analytics to give you the full picture, rather than just fragments.

However, there’s a thin line between enabling access and creating risk. Giving unrestricted access to enterprise data to all users, or even worse, public LLMs is not safe. The solution is smart access governance.

Instead of connecting the LLM directly to live databases, organizations should implement a protective architecture that allows users to interact with data through controlled layers.

A common method is to use query translators or semantic layers that sit between the user and the database. The LLM will receive the metadata and business logic it needs to generate queries without ever interacting with any raw data. Teams also benefit from this approach, as they can share the same interface and derive insights from a unified view of the data without needing direct access to any of the underlying systems.

Merge Your Data Sources

Once access and governance are sorted, the next step is to unify your technical data infrastructure. Fragmented data is one of the biggest barriers to effective AI, as these models need wide access to data without delays from manual ETL (Extract, Transform, Load) or replication.

For that reason, 86% of organizations are making data unification a priority, according to a Dermio survey. Dermio is a data lakehouse platform that allows businesses to run high-performance AI queries directly over their cloud data lakes without having to move or duplicate them. It is exactly the type of platform that can help organizations merge data sources at the infrastructure level.

Data lakehousing emerged years before the rise of advanced generative AI, but has been a real game changer in helping AI models access large volumes of diverse data quickly and efficiently.

For companies with legacy systems or hybrid environments, however, full centralization may not be practical. In that case, data virtualization can be used to allow AI models to analyze data across systems without needing to replicate or relocate it.

Prioritize High-Quality, Well-Categorized Data

The AI model you use is only as good as the data you feed it. Poor-quality data leads to poor AI output, regardless of how powerful the AI model is. As the saying goes: garbage in, garbage out.

Granted, AI models have gotten so advanced that you may get away with less-than-perfect data in some scenarios. However, if you want to derive insights that will actually make a difference and drive decision making, the data needs to be clean, relevant, and well organized.

This starts with establishing consistent data taxonomies across the organization. The data you collect from finance, marketing, and all other departments should share the same framework for naming conventions, definitions, categories, and formats. Without this consistency, AI models may struggle to interpret the data correctly, leading to flawed or conflicting outputs.

You can also leverage a data lineage tool like Alation to track where data originates and whether it meets quality standards. 

Conclusion: Fix the Data, Then Deploy the AI

The promise of AI-powered analytics is very compelling, but only when built on a foundation of clean, connected, and well-governed data. Too often, companies rush to implement AI solutions without fixing the underlying data issues. This leads to mixed results in the effectiveness of the solution, and delays the process for when the company finally decides to go back and address the root problems.

For organizations that take the time to get their data infrastructure right, AI delivers on its promise.