Passionate about creating compelling visual stories through the analysis of big data, Chloe Maraina is a leading Business Intelligence expert with an aptitude for data science and a vision for the future of data management and integration. As organizations navigate a landscape transformed by artificial intelligence and an ever-increasing demand for data-driven insights, her perspective is more critical than ever. In our conversation, we explored the delicate balance between empowering users and ensuring data integrity, the paradigm shift brought by autonomous AI agents, the modernization of data architectures like the data lakehouse, and the crucial human skills required to thrive in this new era. We also delved into the foundational technologies, such as semantic layers and analytics-as-code, that are bringing new levels of discipline and consistency to the field.
With more business users accessing self-service BI tools, how can organizations balance data democratization with robust security and privacy? Please describe the practical steps leaders should take to govern not just the data, but the entire decision-making process driven by BI.
It’s a fantastic challenge to have, isn’t it? The goal has always been to get data into more hands, but now we’re grappling with the consequences of that success. The reality is, with cyberattacks on the rise and complex regulations like GDPR and HIPAA to navigate, security and governance have become the most critical concerns for many businesses deploying BI. The key is to shift our thinking. We’re not just locking down data sets anymore; we need to govern the entire decision-making pipeline. A practical first step is implementing an analytics catalog. Think of it as a curated library where users can find certified, relevant dashboards and reports. It guides them, shows them what’s appropriate for their work, and builds trust. This is a much more elegant solution than just restricting access. We’re also seeing a huge focus now on AI governance, which adds another layer of complexity. When an AI tool provides an insight, you absolutely must be able to document the model, the training data, and the confidence level of the output. Explainability isn’t just a buzzword; it’s becoming a regulatory requirement.
Agentic AI can now autonomously monitor data, run analyses, and present findings without human intervention. How does this fundamentally change the role of a human analyst, and what new validation processes are needed to ensure we can trust and act on an AI agent’s conclusions?
This is where the ground is truly shifting under our feet. Agentic AI isn’t just an assistant; it’s a proactive partner. Imagine an agent noticing a supply chain anomaly on its own, investigating potential causes by querying related data, and then presenting you with a full summary and recommended actions. It’s a world away from a human analyst just running pre-defined queries. The role of the analyst is fundamentally elevated. They move from being a data manipulator to a strategic reviewer and decision-maker. Their job becomes less about the “how” of the analysis and more about the “so what?”—validating the agent’s conclusions and orchestrating the business response. This demands a new, rigorous validation process. We can’t just blindly trust the output. Leaders need to establish best practices for interrogating these AI-generated results before anyone acts on them. This means creating checkpoints, demanding transparency from the models, and fostering a culture where questioning the AI is not just accepted but expected.
The data lakehouse architecture aims to be a single platform for all analytics. For a company migrating from a traditional data warehouse, what are the biggest benefits for BI teams, and how does this architecture enable a move toward continuous, real-time intelligence for business operations?
The move to a data lakehouse feels like going from a quiet library to a bustling, dynamic information hub. For years, BI teams were constrained by the rigid structure of the traditional data warehouse, which was great for historical reporting but struggled with the variety and velocity of modern data. The biggest benefit of a lakehouse is that it breaks down the walls between BI, data science, and data management. It’s a single platform that offers the flexibility of a data lake for raw, unstructured data and the performance of a data warehouse for structured queries. For a BI team, this means streamlined access to a much wider range of data without cumbersome ETL processes. More importantly, the lakehouse architecture is built to treat streaming data as a core component. This convergence is what unlocks continuous intelligence. Instead of waiting for a nightly batch update, insights are updated on the fly as events happen. For use cases like fraud detection or dynamic pricing, that real-time capability delivers business value that older BI models simply can’t touch.
As AI generates more insights, AI literacy is becoming critical. What specific skills should training programs focus on to help employees evaluate AI recommendations, and how can leaders foster a culture where it is safe and even encouraged to question an AI’s output before acting on it?
You’ve hit on the most important human element in this whole transformation. A data-driven culture is impossible without a data-literate workforce, and today that absolutely includes AI literacy. Training programs need to move beyond just teaching people how to use a tool. They must focus on critical thinking. This means training employees to recognize appropriate use cases for AI, to understand the potential biases in an algorithm, and, crucially, to know when to trust an AI recommendation versus when to dig deeper. It’s about developing an intuition. One specific skill is mastering data visualization—not just creating a pretty chart, but understanding if the visualization presented by an AI is genuinely insightful or misleading. Leaders are the linchpins in fostering the right culture. They must create an environment of psychological safety where an employee can raise their hand and say, “I’m not sure I trust this AI’s conclusion,” without fear of looking incompetent. In fact, that skepticism should be rewarded, because it’s the last line of defense against flawed, automated decisions.
Semantic layers are positioned as a “single source of truth” to prevent AI from inventing metrics. Could you walk us through the process of building a universal semantic model? What are the biggest challenges in defining and maintaining consistent business rules across a large enterprise?
The semantic layer is our secret weapon against AI hallucination. We’ve all seen AI invent plausible-sounding metrics that are completely wrong. A strong semantic layer prevents this by providing a clear, consistent set of definitions that constrains the AI’s behavior. Building one starts with a collaborative, cross-functional effort. You bring together stakeholders from finance, sales, marketing, and operations to define core business metrics. What exactly constitutes “revenue”? How do we define an “active customer”? These definitions, hierarchies, and business rules are then codified into a shared layer that sits between the raw data and the analytics tools. The biggest challenge, by far, is achieving and maintaining consensus in a large enterprise. Different departments often have their own dialects and definitions built up over years. Getting everyone to agree on a single source of truth requires strong leadership, clear communication, and a willingness to compromise. The maintenance is also non-trivial; as the business evolves, the semantic layer must be versioned and governed just like any other critical asset, often within the data lakehouse itself.
New approaches like “analytics as code” apply software engineering discipline to BI. How does this help teams reuse and collaborate on analytics artifacts, and can you share a specific example of how this methodology would reduce errors and speed up the creation of a complex financial dashboard?
“Analytics as code” is about ending the cycle of reinventing the wheel, which has plagued BI for years. In the past, if you built a complex measure or a geographical hierarchy for a dashboard, it was incredibly difficult for anyone else to reuse it. So, another analyst would just build their own version from scratch. This is not only inefficient but a recipe for errors and inconsistency. By applying software engineering principles, we treat these analytics artifacts—measures, dimensions, hierarchies—as code. They can be versioned, shared in a central repository, and collaboratively improved. For a complex financial dashboard, imagine defining your “quarter-to-date profit margin” calculation once. It’s saved as a code object. Now, every analyst building a financial report can pull that exact, vetted calculation. There’s no risk of someone misinterpreting the formula or making a typo. This dramatically reduces errors and speeds up development because your team is building with reliable, reusable components instead of starting from zero every single time.
What is your forecast for the future of business intelligence?
My forecast is that BI will become more essential and, paradoxically, more invisible. It will continue to be the backbone of business success, but its delivery will be completely transformed. The era of the standalone BI application as the primary interface is waning. Instead, insights will be deeply embedded within the operational applications where people do their work, providing guidance at the exact moment of decision. AI will not replace the human analyst but will elevate them to a more strategic role, where they curate, validate, and interpret the outputs of autonomous AI agents. We’ll see a continued convergence on unified platforms like the data lakehouse, where real-time data and advanced analytics are not afterthoughts but core components. Ultimately, the future of business intelligence is one of continuous, ambient intelligence that empowers every employee, making data-driven decision-making feel less like a separate task and more like a natural, integrated part of their daily workflow.
