AI Is Not a Shortcut to Better Reporting. It’s a Test of Your Data Readiness, Says Kristýna Merňáková (Adastra)

April 13, 2026

Companies often ask what AI in Power BI will bring them. But according to Kristýna Merňáková, Data Visualization Lead at Adastra, the real question is different: “What will AI reveal about what we need to fix in our data?”

AI in reporting doesn’t just bring speed. It also quickly exposes weaknesses: poorly defined metrics, unclear terminology, or missing context. And for many organizations today, that’s a bigger problem than the technology itself.

Watch the interview (in Czech):

Listen to the podcast (in Czech):

In the podcast, Kristýna explains how two worlds are now colliding: traditional BI built on precise, trusted data, and AI, which operates on probability and context. This clash creates new demands for data quality and how data is prepared for end users.

A key role is played by the semantic layer – a “translator” between technical data and business meaning. It determines whether AI provides accurate answers or starts generating misleading outputs.

Real-world experience shows that without high-quality data descriptions, clear metric definitions, and proper context, AI tends to create confusion rather than value. On the other hand, a well-prepared data model can significantly accelerate work – from documentation generation to report adjustments.

AI is also reshaping the role of analysts. It doesn’t eliminate them but shifts their focus: from manual report creation to managing data quality, designing semantic models, and educating business users who will increasingly work with data directly.

So how should companies start with AI in Power BI? According to Kristýna, gradually – beginning with simple use cases, focusing on data quality, and avoiding critical areas such as controlling or bonus calculations.

Key takeaways:

• How to prepare data so AI truly helps instead of causing harm.
• How “chat with your data” works in practice.
• Why AI gives wrong answers without context – even when the data is correct.

Read the podcast as an interview

(The interview was shortened and edited using ChatGPT)

Ivana Karhanová: Most companies ask what AI in Power BI will bring them. But the right question is: what will AI reveal about what we need to fix in our data? Where do companies hit the limits of reporting in Power BI and beyond? What about Copilot – will it help, or will it expose weaknesses right away? Today, I’m talking to Kristýna Merňáková, Data Visualization Lead at Adastra. Kristýna, welcome to the studio.

Kristýna Merňáková: Hi.

Ivana Karhanová: What’s the reality today? Power BI is widely used as a reporting tool, and Microsoft presents Copilot as something that generates reports, reduces manual coding, and helps understand data. But when you walk into a company, what do you actually see?

Kristýna Merňáková: The reality isn’t simple. It’s complex and multi-layered. AI has been around for a few years, but it only started entering BI last year, so it’s still very new. We’re seeing two worlds collide – AI and BI. BI has historically been about managing a company based on reliable, structured data with clear logic and guarantees. AI, on the other hand, works with probability and non-deterministic outputs. That’s the paradox – and the challenge – figuring out where it actually helps.

Ivana Karhanová: Let’s clarify what Copilot in Power BI actually is.

Kristýna Merňáková: Microsoft offers Copilot in multiple places. There’s Copilot in M365, like in Teams or Outlook, which helps with everyday tasks. What we’re talking about here is Copilot directly in Power BI. It works over a specific semantic model or report, and you can interact with it directly over your data.

Ivana Karhanová: So it’s not a universal tool that generates any report on demand?

Kristýna Merňáková: It allows you to ask questions over data, but I wouldn’t recommend using it without preparation. “Chat with your data” isn’t optimal if you just apply it to existing reports. The data needs to be prepared properly.

Ivana Karhanová: What stage are we in right now?

Kristýna Merňáková: It’s still early. Users are learning how to work with it. Underneath everything, you still have data and a semantic layer – metadata, relationships, descriptions, business logic. AI shortens access to data, but it also raises questions about validity and responsibility.

Ivana Karhanová: If business users query the data, AI needs to understand it. Who provides that understanding?

Kristýna Merňáková: That’s exactly the role of the semantic layer. It has to be prepared so that everything is clearly described. AI works very well with that – but it’s up to analysts to build it properly. Power BI includes a feature called “Prep Data for AI,” which involves selecting relevant data, verifying answers, and adding instructions and context.

Ivana Karhanová: So it’s not a magic button.

Kristýna Merňáková: Exactly. You have to guide it – like a junior. Prepare the data, provide context and instructions. Otherwise, it won’t work correctly.

Ivana Karhanová: What is the semantic layer in simple terms?

Kristýna Merňáková: It’s the layer between technical data and what users see in reports. It translates technical language into business language.

Ivana Karhanová: How important is documentation?

Kristýna Merňáková: Very. And AI can now help generate it, which also improves how AI understands the data.

Ivana Karhanová: What if people ask the wrong questions?

Kristýna Merňáková: AI won’t think for us. The clearer the question, the better the answer.

Ivana Karhanová: Can we recognize when answers are wrong?

Kristýna Merňáková: Yes. Business users know their numbers. If something is off, they notice immediately. In the future, AI may even help suggest fixes to the data model – something that takes days today could take hours.

Ivana Karhanová: Who is responsible for data?

Kristýna Merňáková: It depends on the company, but there should always be a clear owner of each metric and its meaning. Communication between business and IT is key.

Ivana Karhanová: How should companies start with AI in Power BI?

Kristýna Merňáková: Gradually. Start with simple datasets like sales. Avoid critical metrics like controlling or bonus calculations. AI can provide quick insights, but not always perfectly accurate ones.

Ivana Karhanová: Where is this heading?

Kristýna Merňáková: AI will mainly help with building the semantic layer – generating documentation, refactoring, designing models. It will save manual work but remain under human control.

Ivana Karhanová: What will analysts do?

Kristýna Merňáková: They won’t disappear. It’s like AI in medical diagnostics – it helps, but humans make the decisions. Analysts will focus more on data quality, communication, and education.

Ivana Karhanová: AI is often said to work even with poor data. Is that true?

Kristýna Merňáková: No – quite the opposite. AI needs high-quality data and governance. Only then can it help. Otherwise, it just exposes problems. It’s the cherry on top of well-prepared data.

Ivana Karhanová: Thank you for the conversation.

Kristýna Merňáková: Thank you for having me.