---
title: "Why agentic analytics starts with a well"
description: "Agentic analytics increases the need for trustworthy semantics and governance because AI systems must interpret business definitions, trace outputs, and support follow-up questions beyond static dashboards. Nick Eayrs argues that fragmented metrics across BI tools, proprietary semantic layers, and duplicated pipelines erode trust, slow decisions, and increase engineering cost. The proposed foundation is an open, interoperable semantic layer where certified metrics such as revenue, churn, and customer lifetime value are explicitly defined, reusable, and accessible through SQL to BI, AI interfaces, notebooks, and agents. Organizations should establish governed data foundations and a catalog first, certify metrics next, then add AI and evaluate its outputs. Examples include NTT Docomo reducing manual log-analysis time from 66 hours to 6 hours and Net One Systems cutting support response time by 75 percent."
---

# Why agentic analytics starts with a well

[Databricks](https://yomu.fyi/company/databricks) · Catherine Brown · Sep 18, 2026

**Type:** Explainer

## Summary

Agentic analytics increases the need for trustworthy semantics and governance because AI systems must interpret business definitions, trace outputs, and support follow-up questions beyond static dashboards. Nick Eayrs argues that fragmented metrics across BI tools, proprietary semantic layers, and duplicated pipelines erode trust, slow decisions, and increase engineering cost. The proposed foundation is an open, interoperable semantic layer where certified metrics such as revenue, churn, and customer lifetime value are explicitly defined, reusable, and accessible through SQL to BI, AI interfaces, notebooks, and agents. Organizations should establish governed data foundations and a catalog first, certify metrics next, then add AI and evaluate its outputs. Examples include NTT Docomo reducing manual log-analysis time from 66 hours to 6 hours and Net One Systems cutting support response time by 75 percent.

## Context

The discussion contrasts static, report-driven business intelligence with AI-enabled analytics that can answer follow-up questions, predict outcomes, explain causes, and recommend actions. It argues that this broader capability raises the importance of shared business definitions, lineage, access controls, auditability, and open interoperability.

## Approach / What changed

Establish a governed data foundation and catalog, define and certify reusable business metrics in an open semantic layer, expose them through SQL and interoperable interfaces, then add AI and use evaluation frameworks to monitor and refine its outputs.

## Takeaways

- Fragmented semantic models across BI tools can produce conflicting definitions of metrics such as revenue, forcing teams to debate which number is correct instead of deciding what to do.
- An open semantic layer lets BI, AI interfaces, notebooks, agents, and data science teams use the same governed foundation rather than duplicating transformations and maintaining synchronization pipelines.
- The recommended implementation sequence is data foundations, certified metrics, AI capabilities, and evaluation; NTT Docomo reduced manual log-analysis time from 66 hours per month to 6 hours per month.

## FAQ

### Why does AI need more semantic context than traditional dashboards?

AI must interpret terms such as “targets,” the relevant period, and the organization’s metric definitions when answering follow-up questions. Without that context, it may generate generic answers that are not trusted.

### Why should the semantic layer use open standards?

Open standards and interfaces let BI, AI, notebooks, agents, and data science teams build on the same governed data. They also avoid locking business logic into one proprietary tool and reduce repeated extraction, transformation, and custom synchronization work.

### What sequence does the source recommend for adopting agentic analytics?

The sequence is to establish data foundations, build the semantic layer and certify business metrics, then add AI. Evaluation frameworks should monitor and refine the resulting outputs.

**Tags:** [AI](https://yomu.fyi/topic/ai), [Analytics](https://yomu.fyi/topic/analytics), [Data Governance](https://yomu.fyi/topic/data-governance)

- Source: [Databricks](https://www.databricks.com/blog/why-agentic-analytics-starts-well-governed-data-layer)
- Source URL: https://www.databricks.com/blog/why-agentic-analytics-starts-well-governed-data-layer
- Ingested by Yomu: 2026-09-17T14:06:28.981Z

[Read original post](https://www.databricks.com/blog/why-agentic-analytics-starts-well-governed-data-layer)