Here is a scenario that plays out in organisations everywhere, every week. Two dashboards, both tracking the same product's revenue, show contradictory spikes on different dates. Someone in a meeting asks which one is correct. The room goes quiet. The head of analytics starts digging. A data engineer is pulled away from other work. Two hours later, it turns out the reports were pulling from different systems with different timing rules.
This is not an edge case. It is the normal state of enterprise data. Most large organisations have accumulated years of reports, tables, databases, and documents that were each built to answer one specific question, by one specific team, at one specific moment. Nobody went back to reconcile them. Over time, the word "revenue" came to mean slightly different things in different systems, and most organisations cite data-driven decision-making as a goal while fewer than half say they actually trust the data they use to make decisions.
Databricks, the data platform used by over 60% of the Fortune 500, has published research on how its AI agent called Genie is being designed to handle exactly this kind of problem. The technical detail is less important than what the approach reveals. A generic AI assistant, given access to enterprise data, solves about 32% of real-world business questions correctly. Genie, built with specific techniques for navigating messy, contradictory enterprise data, solves over 90% of the same questions. That gap is significant because it reflects a fundamental difference in how the tool understands your organisation's context versus just your data.
The key insight in the research is that enterprise data questions are different from simple lookup questions. They require cross-referencing multiple systems, understanding company-specific definitions, and being able to spot when an initial assumption is wrong and course-correct mid-answer. A general AI tool cannot do this reliably. A tool trained specifically on enterprise data patterns, and given access to the business logic embedded in your existing reports and documents, can.
This is the race now happening across the entire industry. Snowflake has announced its own natural language data query tools. Microsoft is embedding AI question-answering directly into its data platform. Google has similar capabilities inside its cloud data products. Every platform is moving in the same direction because the demand is clearly there. Customers have been asking for years to simply talk to their data and get a trustworthy answer.
The practical implication is that the organisations most likely to get value from these tools are not the ones with the cleanest data. They are the ones with the most consistent definitions. When the word "customer" means the same thing in your sales system, your finance system, and your operations system, an AI agent can do useful work. When it means three different things, any AI tool will produce three different answers and you are back where you started.
The organisations winning with AI data tools right now are the ones that did the boring governance work first: agreeing on what metrics mean, who owns them, and where the authoritative version lives. That is not glamorous work. But it is increasingly the prerequisite for everything else. The tools are ready. The question is whether the data underneath them is.