Every major software vendor is now selling the same promise. Point AI at your business, ask a question in plain English, and get a trusted answer back. Salesforce, Microsoft, Snowflake, Databricks, and a dozen others have built their AI messaging around it. The promise is real, and the demos are honest about what AI can do. What the demos leave out is what it takes to get there.
Why the demo works and your business doesn’t
A polished demo works because it runs against one clean dataset, with metric definitions already loaded, answering a question chosen in advance. A real mid-market company rarely has any of those conditions in place. Your data is scattered across an ERP, a CRM, a ticketing system, and a stack of spreadsheets that hold the logic tying everything together. “Revenue” means one thing to finance and another to sales. Margin by customer depends on cost allocations that live in one analyst’s head. Point AI at that, and it cannot see across your systems, it has no way to know which definition you use, and it answers anyway. The result reads as fluent and confident, and is sometimes wrong in ways the person asking cannot detect.
The work the shortcut assumes is already done
This is not a flaw in the AI. The questions a business actually cares about live in the overlap between systems, and the AI has been handed a single piece of the picture. The work of assembling the full picture, integrating the systems, reconciling the definitions, and encoding the business rules, is exactly what the shortcut assumes is already done. The vendors know this. Their own documentation calls for a hand-authored semantic model, packaged metric definitions, and a domain-specific glossary before any of it works. The natural-language chat box is the last mile. The road underneath it still has to be built.
The foundation that makes AI trustworthy
That road is the same foundation that has always made business intelligence trustworthy. Raw data is integrated and cleaned across systems, then modeled so that every metric is computed one agreed way and the same question returns the same answer every time. Once that foundation exists, it serves both your governed dashboards and your AI from one trusted source. The dashboard answers the high-volume questions you already know you need, the same way every week. The AI answers the long tail of exploration, putting that capability directly in the hands of the people who know the business best.
You need both BI and AI
This is why AI analytics and traditional BI are complements, and why you need both. The companies getting real value from AI run it alongside their reporting, on a foundation they own and control, with no lock-in to a single vendor’s cloud.
Our white paper, AI Needs a Foundation Too: Why “just point it at your data” doesn’t work, and what does, makes the complete case from the vendors’ own documentation, including the single-vendor lock-in trap, the economics of AI at scale, and a six-question checklist for any vendor selling you the shortcut.
