I wouldn’t recommend trusting AI-powered BI tools without first addressing how business metrics are defined and governed.
The issue isn’t necessarily that Sales and Finance have different revenue numbers. It’s that both numbers can be correct within their own business context.
Sales might be measuring bookings or closed deals, while Finance is measuring recognized revenue. Those are different metrics serving different purposes.
The real problem begins when an AI system treats them as interchangeable.
Imagine someone asking an AI-powered BI assistant, “What was our revenue last quarter?”
If the system pulls data from whichever dashboard or table appears most relevant, it might generate a perfectly convincing answer using the wrong definition of revenue.
And because the response sounds confident, the user may never question it.
Here’s how I’d approach this before rolling out conversational BI:
1. Establish a governed semantic layer.
Define important business metrics, their calculation logic, source tables, ownership, and intended use. Tools like Power BI semantic models can help maintain these definitions consistently.
2. Preserve legitimate differences between departments.
A single source of truth doesn’t mean forcing every department to use one metric. Sales bookings and recognized revenue can coexist, provided they’re clearly defined and the system understands when each applies.
3. Make AI ask for clarification.
If someone asks an ambiguous question about revenue, the AI should ask whether they mean sales bookings or recognized revenue rather than silently choosing one.
4. Make answers traceable.
Users should be able to see which metric definition, filters, data sources, and reporting periods were used to generate an answer. Ideally, they should also be able to trace the calculation back to the underlying data.
5. Test AI against known business scenarios.
Before wider deployment, I’d have Sales, Finance, and the BI team test the same set of questions. Compare the AI-generated responses against approved calculations and identify where ambiguity or inconsistency appears.
This is especially important for financial and operational KPIs that influence business decisions.
Can AI-powered BI tools manage conflicting definitions?
Yes, to an extent. With well-designed semantic models, metadata, access controls, and clear business context, they can distinguish between different definitions and explain why numbers differ.
But they cannot independently decide which definition an organization should consider authoritative. That requires agreement between the people responsible for the data and the business processes behind it.
I also wouldn’t delay every AI initiative until the entire organization achieves perfect data consistency. That’s rarely realistic.
Start with a limited set of well-governed metrics, validate the results, and expand gradually.
My view is that AI-powered BI doesn’t necessarily require one centralized database. It requires agreed definitions, clear ownership, and a reliable way to trace every answer back to its source.
Otherwise, we’re simply making inconsistent reporting faster and easier to access.
I’m curious whether teams implementing conversational BI are finding semantic modeling or cross-department agreement to be the bigger challenge.