Running self-service analytics without losing the numbers

Decide who builds what

Before choosing tools, agree which kinds of content business users create and which the central team owns. Most organisations end up somewhere on this spectrum, and many run different models for different domains.

Whatever you choose, write down who owns it. Gartner found in 2025 that 70% of chief data and analytics officers have primary responsibility for building the AI strategy and operating model, so the decision usually lands with the same office.⁴

Certify a small core

Pick the datasets and models that feed executive reporting and certify them: an owner, documented definitions, automated quality checks and a visible badge in the BI tool. Keep the list short enough that the team can actually stand behind every item on it.

Then set one rule for self-service content. Anything shown to senior leadership or used in a regular decision is built on certified data, or carries a clear label saying it is not. People keep their freedom to explore, and the numbers that matter come from one place.

Put the definitions in one layer

Metric definitions scattered across individual reports are what turn self-service into sprawl. A semantic layer holds them once, so every tool and every user reads the same definition of revenue. Databricks launched Unity Catalog Metrics in public preview in June 2025 to move metric definitions out of BI tools and onto the data platform.⁵ The Open Semantic Interchange published the first version of its open specification in January 2026, giving definitions a portable format between tools.⁶

Gartner predicts that universal semantic layers will be treated as critical infrastructure, alongside data platforms and cybersecurity, by 2030.⁷ You do not need to wait for that. Start with the ten terms executives ask about most, as I argued in an earlier piece.

Add natural language, carefully

Plain-language querying is already common. In a Gartner survey of 403 analytics and AI leaders, more than half said their organisations use AI tools for automated insights and natural language queries, and Gartner forecasts that 75% of new analytics content will be contextualised for intelligent applications through generative AI by 2027.⁸

The risk is answering questions confidently from data nobody trusts. Salesforce found that data and analytics leaders estimate 26% of their data is untrustworthy, and 89% of those with AI in production had seen inaccurate or misleading outputs.³ Point natural-language tools at the certified core and the semantic layer first, and widen their reach only as definitions and quality checks catch up.

Measure use and prune

Self-service content grows faster than anyone retires it. Pull usage logs every quarter, retire what nobody opens, find owners for what nobody owns and fold duplicates into certified datasets. I wrote about this in more detail in Nobody counts the dashboards.