For the past three years, "chat with your data" has been one of the most overused and misunderstood phrases in enterprise analytics.
Every major analytics and data platform rushed to add GenAI capabilities to their platforms offering natural language interfaces layered on dashboards, semantic models, or raw tables. Early demos were impressive. Executive curiosity was high. Pilot programs launched quickly.
Today, many of those same vendors are far more careful in how they describe their GenAI analytics capabilities. Language has shifted toward assisted analytics, exploration, productivity, and in some cases explicit disclaimers that GenAI outputs are not system-of-record decisions.
This blog examines why that shift happened, what it reveals about the limits of traditional analytics stacks, and what it means for organizations that still want trusted, conversational business insights.
The First Wave: Ask Questions in Natural Language
The original promise was simple:
If business users can talk to data like they talk to colleagues, analytics adoption will finally scale.
Vendors across BI and data platforms introduced NLP features positioned as conversational analytics, often demonstrated with deceptively clean examples:
- "What were last quarter's revenues by region?"
- "Which products are underperforming this month?"
In controlled demos, these experiences worked well. In production environments, they rarely did.
The Market Correction: From NLP to Assisted Analytics
Over time, several well-known platforms softened their messaging. Not because GenAI failed technically, but because it failed operationally.
BI Platforms Reframing NLP
Vendors such as Tableau, Power BI, Qlik, and Looker began reframing conversational features as:
- Insight summarization
- Dashboard explanation
- Measure suggestions
- Analyst productivity accelerators
The implicit message became clear: AI can help interpret analytics, but it cannot be trusted to define business truth. Yet.
Data Platforms Reframing NLP
Data platforms followed a similar trajectory.
Databricks (via Genie) and Snowflake (via Cortex and Copilot features) increasingly emphasized:
- SQL generation assistance
- Exploration and ideation
- Developer productivity
Both avoided positioning GenAI as an executive-grade analytics interface or as a proven system-of-record decision engine.
Why Vendors Had to Pull Back
The retreat from NLP-first analytics was driven by five structural realities.
1. Semantic Fragmentation
Most enterprises do not have one semantic model. They have dozens. NLP systems produced different answers depending on which model, dashboard, or dataset was queried.
2. Metric Drift
Executives noticed quickly when "revenue," "active customer," or "risk exposure" returned inconsistent results across tools.
3. Hidden Business Logic
Critical logic lived inside dashboards, SQL scripts, and analyst workflows, completely invisible to GenAI systems.
4. Governance and Audit Failures
In regulated industries, analytics outputs must be explainable, traceable, and defensible. Free-form GenAI responses failed basic audit standards.
5. Executive Trust Collapse
Once leaders saw conflicting answers, confidence eroded fast. Usage dropped. Pilots stalled.
The safest move for vendors and customers alike was to reclassify GenAI analytics as assistive rather than authoritative.
The Unintended Consequence: A Strategic Gap
Ironically, this market correction created a new problem.
Executives still want:
- Conversational access to insights
- Faster decision cycles
- Less dependence on analytics teams
But now they are told:
- "Use AI for exploration, not decisions"
- "Always validate results manually"
- "Dashboards remain the source of truth"
In other words, the consumption model did not change.
BI and Data Platform Vendors Responding to Fix the Issues
Over the past year many of these vendors have started to implement tools and techniques that will improve these systems and get them closer to the systems-of-record-for-decisions required by enterprises.
For instance, Microsoft insists on organizations needing to explicitly prepare and model data for GenAI use cases. Databricks similarly insists on using AI/BI Genie with Unity Catalog and Metrics Views, meaning curated semantic definitions and metrics.
While these are good steps forward, many challenges remain in that the way to achieve these curated semantics and metrics require heavy efforts from humans doing manual data modeling and coding YAML files to create these for system use.
Why Codd AI Took a Different Path
Codd AI was built on a contrarian assumption:
Conversational analytics fails not because of GenAI, but because enterprises never encoded business context in a reusable way.
Comprehensive and Unified Contextual Semantic Layer
Instead of bolting NLP onto dashboards or raw data, Codd AI introduces a contextual semantic layer that:
- Encodes business definitions, KPIs, and logic once
- Aligns structured and unstructured data with domain meaning
- Preserves lineage, governance, and explainability
- Constrains GenAI to reason only within approved business context
This flips the model:
| Traditional NLP Analytics | Codd AI Approach |
|---|---|
| Ask raw data questions | Ask business questions |
| Logic trapped in reports or user-supplied context | Logic centralized and governed |
| AI guesses intent | AI reasons over context |
| Assistive outputs | Decision-grade answers |
GenAI-Powered Automation to Reduce Cost and Effort
But Codd AI goes further in that it extensively uses GenAI to automate the entire process of creating your ontology, data model, and metrics, reducing 50-70% of the effort to create these foundational semantic artifacts.
Built for the GenAI World
Codd AI was designed and built from the ground up for the GenAI world. It was built to eliminate hallucinations and provide a governed and transparent GenAI analytical foundation. Codd AI extensively uses GenAI to power model creation, metrics creation, and to embed our semantic layer to power agentic conversational analytics.
The Real Lesson from the Market
The industry did not abandon conversational analytics because it was a bad idea.
It abandoned context-free conversational analytics.
The next generation of analytics will not be defined by better prompts or prettier chat interfaces, but by whether GenAI is grounded in the same business logic leaders already trust.
Final Thought
GenAI did not fail analytics.
Analytics failed to prepare for GenAI.
Platforms that treat AI as an assistant will always need dashboards and BI reports as a safety net.
Platforms that treat context as the system of record can finally make analytics conversational, without sacrificing trust.
About Codd AI
Codd AI is an AI-powered analytical platform designed from the ground up for the GenAI age of analytics. It is designed to overcome the risks of hallucinations by providing an enriched context-aware semantic layer that serves as the foundation for GenAI to interpret questions, understand results, and generate business-relevant insights. Whether you are using our built-in conversational Canvas, Metric Boards, or embedding this into Slack or your BI tools, Codd AI provides the governed and trusted foundation for transforming how your business generates insights and makes decisions.
To learn more, visit us at Codd.AI or schedule a quick intro call.