Before You Ask GenBI, Build the Foundations

GenBI is everywhere right now. In recent months, it’s been all over the place, many people are talking about it, and everyone in the BI space is trying it. Over the past year, GenBI has evolved at high speed, faster than most of us expected. New features appear almost daily, and AI tools like Copilot in Power BI, Data Agents in Fabric, Power BI MCP Servers… they promise magic: ask a question in plain language, get an answer instantly. No more hours of report building or wrestling with data.

But speed isn’t everything. As with anything done well, it’s not about moving fast, but about structure and care. In Japan, true craftsmanship comes from mastering the basics. Consider it similar to forging a katana, a traditional Japanese sword that is made by numerous intentional iterations, each of which strengthens the blade. GenBI follows the same principle: You don’t just grab a piece of steel and start swinging. The beauty and strength come from the forge, slow, deliberate, precise. GenBI is no different. Without solid foundations, that shiny AI layer is just a mask, and it can mislead you with wrong answers.

Forging Data Foundations

Real GenBI work starts long before Copilot is turned on. Think of it like forging a katana: the beauty and sharpness of the blade emerge only after folding, hammering, and careful refinement. It’s the same with proper dimensional modeling, clear grain definitions, and hardened transformations, they’re not just old practices; they are the hammer and fire shaping insights that last. Skip them, and your AI blade may look sharp for a moment, but it will break under pressure.

When fact tables are messy, dimensions inconsistent, or business definitions unclear, AI doesn’t fix it, but it amplifies the confusion. A star schema and a well-prepared semantic layer are the invisible parts of the forge: quiet, painstaking, but essential if you want a blade that cuts clean and true.

Preparing the Semantic Model for AI

Once the foundations are ready, the semantic model becomes the bridge to GenBI. Ambiguous names are removed, clutter hidden, and relationships aligned for both correctness and performance. No technical namings are used but only business-friendly names, with synonyms and descriptions added to tables, columns, and measures. Every explicit measure is defined, and every element is polished.

Refined this way, the semantic model is like a katana freshly honed. Its strength isn’t obvious at first glance, but when wielded in a right way, it’s precise, sharp, and aligned with business needs.

Polishing the AI Edge with Business Empathy

In many ways, preparing for GenBI is similar to preparing for (managed) self-service BI when it comes to the semantic model. However, while the semantic model defines the structure and language of the data, AI instructions, business empathy, and well-crafted prompts shape behavior and intent. Here’s the twist: good definitions, synonyms, and clear naming become even more critical for an LLM to understand the data properly. GenBI adds a behavioral layer, guiding the AI to interpret information meaningfully. Tools like “Prep data for AI” in the Power BI semantic model and Data Agents let teams encode intent directly into the system.

To succeed, you must understand the questions the business actually has. Then, constantly refine the AI instructions in the semantic model. This requires ‘Omoiyari’, empathy for the business user. Anticipate their needs, understand the context behind their questions, and structure guidance accordingly. When done right, an LLM’s answers are not just technically correct, they’re relevant, practical, and useful.

Crafting prompts is just as important. A good prompt helps the LLM to work well, just like a samurai skillfully wields a katana. The prompt guides the AI, directs its reasoning, and maximizes the value of the underlying data.

Governance Is Not Optional

Many organizations underestimate the governance implications of GenBI. Turning on AI without clear rules isn’t ambitious, it’s risky, just as we still see with (managed) self-service BI today.

From what we’ve seen in projects, GenBI works best when it’s clearly positioned in the adoption framework. Users must know when AI is appropriate and when it isn’t, and also how to write a good prompt or have access to a predefined prompt library. Some data domains may be open for exploration, while others need strict oversight. Data Agents introduce new assets that require ownership, monitoring, and lifecycle management.

If governance is applied thoughtfully, GenBI solutions become scalable and reliable; without it, they falter and struggle with user adoption.

GenBI and Traditional BI Are Stronger Together

Some think GenBI will replace traditional reporting. We believe it won’t. Reports remain the backbone of structured decision-making. They provide curated topics, controlled definitions, and narrative context like for example storytelling with visuals. GenBI, on the other hand, adds a conversational layer, making exploration faster and more flexible.

The best results come when both coexist. Quickly check KPIs with AI, then dive deeper in the report. Notice a trend on a dashboard? Ask GenBI for insights to explore it further. Both tools together make analysts more efficient.

Behind the hype

GenBI is not a quick fix. It's not magic on top of poor data. Teams must be rigorous about foundations, semantics, governance, and user understanding as a result of this architectural change.

The final step rarely determines success, it is everything that comes before it that matters. Like a katana, GenBI is only efficient and effective when built on solid preparation and strong foundations. Organizations must carefully prepare their data, semantic models, and prompts for GenBI to deliver its full potential with precision, insight, and control. Just as a samurai needs skill, patience, and understanding to wield it properly. Are you in need of the right skills and understanding to turn your GenBI endeavors into a real success? Talk with Datashift, we are always open to talk and ready to help.