Natural-Language BI: The Governance Questions

Natural-language business intelligence — platforms like ThoughtSpot, and the AI-driven query features spreading across the BI tools — promises something genuinely appealing: anyone can ask a question of the data in plain language and get an answer, no dashboards or query skills required. That democratisation is powerful, and it raises governance questions that the demo does not dwell on. When anyone can ask anything of the data and get an authoritative-looking answer, the questions of whether the answer is correct, whether the person should see that data, and whether the number means what they think it means become acute — because the guardrails that a curated dashboard provided are exactly what natural-language query removes.

Why natural-language BI sharpens the governance problem

A curated dashboard is governed by construction: someone decided what data it draws on, how the metrics are defined, and who can see it. Natural-language query dissolves that curation. The user asks a free-form question, and the system interprets it, decides what data to use and how to calculate, and returns an answer — often with the confident authority of a precise number. That creates specific risks. The system may interpret the question in a way the user did not intend and return a technically-correct answer to the wrong question. It may compute a metric differently from the firm’s agreed definition, so “revenue” in the answer does not match “revenue” in the board pack. And unless access is enforced at the data layer, it may surface data the asker should not see. The ease that makes natural-language BI attractive is exactly what removes the human judgement that caught these problems before.

The governance questions to answer

• Is the answer correct, and does it mean what the user thinks? The system must draw on governed data and agreed metric definitions, or it will confidently return answers that are wrong or misinterpreted. A semantic layer underneath is often what makes natural-language query trustworthy.
• Does it respect who can see what? Access control has to be enforced at the data layer, so that natural-language query cannot become a way for anyone to ask for data they are not entitled to and receive it.
• Does the user understand the answer’s limits? A plain-language answer hides the assumptions behind it; users need to know that a confident number can still be a misinterpretation, and the system should surface its reasoning where it can.
• Who governs the definitions the system uses? Natural-language BI is only as trustworthy as the metric definitions and data it draws on; those need governance, or the answers drift from the firm’s agreed truth.

Adopting it well

• Build it on governed data and metrics. Natural-language query on ungoverned data amplifies the “which number is right?” problem; on a governed foundation with a semantic layer, it can be genuinely trustworthy.
• Enforce access at the data layer. Ensure the system cannot surface data the asker should not see, regardless of how the question is phrased.
• Set expectations about confidence. Help users understand that an authoritative-looking answer can still misinterpret the question, and surface the system’s assumptions where possible.
• Govern the definitions. Own the metrics and data the system draws on, so its answers align with the firm’s agreed truth rather than drifting.

Natural-language BI is a real democratisation of data access, and the appeal of letting anyone ask the data a question is genuine. But it removes the curation that governed dashboards provided, which makes the correctness, access and interpretation questions sharper, not softer. The firms that adopt it well build it on a governed foundation of trustworthy data, agreed definitions and enforced access — so that democratised access delivers trustworthy answers rather than confident, authoritative-looking ones that are wrong, misinterpreted, or shown to the wrong person. The technology is impressive; the governance underneath it is what makes it safe to trust.

Who this is for

This reading is for:

• Data leaders weighing natural-language analytics like ThoughtSpot
• CTOs whose business users want to “just ask the data”
• Governance leads worried about self-serve answers at scale
• Boards told AI will let anyone query the firm’s data

Sixteen Pillars helps firms build natural-language BI on a governed foundation – trustworthy data, agreed definitions, enforced access – so democratised access delivers trustworthy answers. Pricing is published at /pricing/. If this is live for your organisation and you would like an independent reading, the place to start is a conversation.

Sixteen Pillars is a technology governance consultancy based in Cyprus. Engagements run remote across the EU, UK, and Middle East, with on-site time where the engagement requires it.