Stop Filtering, Start Talking: The Era of Conversational Analytics in Google Cloud

Making sense of data should never feel like searching for a needle in a haystack. Yet, for years, Business Intelligence (BI) followed a rigid and frustrating pattern: a user had a question, opened a dashboard, adjusted filters, and hoped the answer existed. If it did not, they filed a ticket and waited.

That workflow is officially obsolete.

With the general availability of Conversational Analytics in Looker and the preview of Conversational Analytics in BigQuery, the distance between a question and an insight is disappearing across the entire Google Cloud data stack. Whether you are a business user, an analyst, or a data engineer, you can now interact with data the same way you interact with people: by asking questions. This is not just a chatbot feature. It is a fundamental shift in how analytics is consumed, governed, and delivered.

🎥 Prefer watching instead of reading? You can watch the NotebookLM podcast video with slides and visuals based on this blog here.

1. What Is Conversational Analytics?

At its core, Conversational Analytics is a “chat-with-your-data” experience powered by Gemini for Google Cloud. Users ask questions in natural language and receive answers as tables, visualizations, summaries, or even predictions. What is new and important is that this experience now exists at two complementary layers of the analytics stack:

1. In Looker: For governed, business-ready insights built on a trusted semantic model.

2. In BigQuery: For rapid exploration, SQL generation, predictive analytics, and unstructured data analysis directly at the source.

🌟 This dual approach acknowledges a simple truth: business users and data professionals do not work the same way, and they should not be forced to.

2. Two Paths, One Philosophy: Governance vs. Exploration

Conversational Analytics in Google Cloud is not one-size-fits-all. It is intentionally designed to serve different personas with different needs.

2.1 Looker: The Trusted Business Layer

Looker’s Conversational Analytics is designed for decision-makers and business teams who need answers they can trust.

What truly distinguishes Looker in the AI era is its semantic model, LookML. While generic AI tools often “hallucinate” or guess at definitions, Looker provides a single source of truth where every field and metric is defined and calculated consistently.

• Accuracy: When you ask a question, Looker translates your natural language into precise SQL based on the trusted logic defined by your data experts.
• Consistency: The semantic layer simplifies complex business logic into a language everyone can understand, ensuring that “revenue” or “churn” means the same thing across the entire organization.
• Transparency: Because LookML acts as a code-based representation of your data structure, it fosters collaboration and ensures that AI-generated insights are governed and accurate.

💡 Want to learn what Looker is and explore its features in depth? Check out our previous blog 👉 Looker: Unleashing the Power of Modern Business Intelligence

2.2 BigQuery: The Analyst’s Accelerator

While Looker excels at curated analytics, analysts and engineers often need to work closer to the data, before models are built or when data is messy. This is where Conversational Analytics in BigQuery comes in. Embedded directly in BigQuery Studio, it allows data professionals to ask questions about datasets using natural language:

• Generate and execute SQL instantly
• Visualize results on the fly
• Perform predictive and statistical analysis
• Work directly with raw, semi-structured, and unstructured data

🌟 Instead of measuring insight by the number of lines of SQL written, analysts can focus on interpreting results and answering “why,” not on debugging syntax.

3. Grounded Intelligence, Not Guesswork

A common concern with AI-driven analytics is hallucination. Google Cloud addresses this differently at each layer, without compromising trust.

3.1 Grounding in Looker

In Looker, grounding comes from LookML, the single source of truth for business logic.

3.2 Grounding in BigQuery

In BigQuery, grounding comes from:

• Table schemas and metadata: Unlike generic LLMs that infer structure from text, the BigQuery conversational agent is directly anchored to your actual table schemas and metadata. Before generating any SQL, it understands:
  • Column names and data types
  • Table structures and relationships
  • Available datasets and views


• Verified Queries: This is the strongest safeguard against inaccuracy. You can provide the agent with a curated library of Verified Queries, trusted, production-grade SQL that defines your most critical metrics, such as how your organization calculates ARR, churn, or Daily Active Users.
  
  

  Instead of inventing logic from scratch, the agent uses these queries as reference patterns, ensuring that new answers:

  • Reuse approved calculation logic
  • Stay consistent with executive dashboards and reports
  • Align with metrics already trusted by the business


• User-defined instructions and patterns: BigQuery also allows you to guide how the agent reasons about your data through persistent instructions and contextual rules, such as:
  • Vocabulary alignment: Teach the agent that “clients,” “accounts,” and “customers” all map to the same business entity or column.
  • Business logic rules: Define constraints like:
    • “Always filter to the last completed quarter when analyzing trends.”
    • “Top performers refer only to sales reps with more than $100K in revenue.”
  
  These instructions ensure the generated SQL reflects your internal language and decision-making logic, not generic assumptions.

🌟 By combining schema awareness, verified SQL logic, and business-specific instructions, BigQuery’s conversational analytics produces queries that are accurate, repeatable, and governable. The AI does not replace your data model; it respects it. And instead of guessing, it behaves like a well-trained analyst who already understands your data, your metrics, and your rules.

4. Key Capabilities Across Looker & BigQuery

Conversational Analytics in Google Cloud is not a single feature; it is a capability spectrum that adapts to who is asking the question and where the data lives. Looker and BigQuery share common principles, but each brings unique strengths.

4.1 Data Agents: Curated AI, Not Generic Chatbots

In both Looker and BigQuery, conversational experiences are powered by Data Agents, custom AI assistants based on your data and business context.

4.1.1 Looker Data Agents are built on top of Explores and the semantic model.

• Contextual Intelligence: You can teach the agent that “loyal customers” means users with more than five completed orders.



• Cross-Domain Analysis: A single agent can connect to up to five Explores, enabling questions that span Sales, Inventory, Marketing, and Operations in one conversation.



• Controlled Sharing: Agents can be shared with specific teams, ensuring consistent definitions and aligned insights across the organization.

🌟 To see how the Conversational Analytics in Looker works, check the video below:

4.2.1 BigQuery Data Agents are designed for builders and data professionals.

• Deep Grounding in Your Data Estate: Unlike generic AI that hallucinates, BigQuery agents are anchored in your specific Knowledge Sources. You can connect an agent to specific tables, views, and User-Defined Functions (UDFs). This allows the agent to understand not just the data schema, but also the custom calculations your team has already built.



• Persistent Context & Instructions: You can program the agent with specific behaviors to ensure it acts like a trained analyst on your team.



• Beyond the Console (The API): These agents are not locked inside BigQuery Studio. Using the Conversational Analytics API, you can integrate these agents into your own custom applications and internal portals. This allows you to build “data apps” where users can ask questions in natural language, and your curated BigQuery Agent handles the heavy lifting in the background

🌟 To see how the Conversational Analytics API works, check the video below:

🌟 To see how the Conversational Analytics in BigQuery works, check the video below:

4.2 Advanced Analytics: When Questions Go Beyond SQL

Not every analytical question fits neatly into SQL. Google Cloud addresses this at both layers, differently but intentionally.

4.2.1. Looker Code Interpreter

When users ask questions involving:

• Forecasting
• Cohort analysis
• Statistical correlation
• Outlier detection

Looker can translate natural language into Python code, execute it securely, and return results. Under the hood, this leverages familiar libraries like pandas and scikit-learn, making advanced analytics accessible without exposing users to code. This is ideal for business-driven deep dives that still need governance and transparency.

4.2.2. BigQuery ML Integration

In BigQuery, advanced analytics is native.

Conversational prompts can automatically trigger BigQuery ML functions, such as:

• AI.FORECAST for demand and sales prediction
• AI.DETECT_ANOMALIES for traffic or revenue anomalies
• AI.GENERATE_TEXT for summarization and insight generation

Examples:

• “Predict sales for the next three months”
• “Find anomalies in yesterday’s traffic”
• “Summarize customer feedback across support tickets”

This makes predictive analytics part of the querying experience, not a separate workflow.

4.2.3. Unstructured Data: Where BigQuery Goes Further

One of the biggest differentiators is unstructured data analysis. BigQuery’s conversational agent can reason across:

• PDFs
• Images
• Text documents stored in BigQuery object tables

This unlocks insights from data that traditional BI tools struggle with, such as contracts, reports, feedback, or scanned documents, without complex preprocessing.

5. Best Practices for Conversational Analytics Success

AI is only as good as the context you give it. To get reliable, high-quality answers, your data foundations must be intentional.

5.1 Looker Best Practices: Optimize the Semantic Layer

1. Apply Clear and Business-Friendly Labels
Looker relies on labels to map questions to data fields. Use the label parameter to replace technical names (like user_creation_date) with readable terms (like User Sign-up Date).

2. Write Thorough Field Descriptions
Use the description parameter to provide the AI with critical context. A description like “The total number of unique users who visited the website” helps the AI map user queries to the correct metrics accurately.

3. Curate Explores to Reduce Field Bloat
Exposing too many irrelevant fields can clutter the AI’s options. Hide technical fields or primary keys using hidden: yes and consider creating streamlined Explores specifically tailored for conversational use.

💡 Want to learn about the best practices for Looker development? Check out our previous blog 👉 Best Practices for Looker Development: A Practical Guide for Data Teams

5.2 BigQuery Best Practices: Ground the Agent

1. Maintain Strong Metadata
Table and column descriptions matter. They help Gemini understand intent and reduce ambiguity.

2. Leverage Verified Queries
Feed the agent your trusted SQL patterns (ARR, churn, revenue logic). This ensures consistency and prevents AI guesswork.

3. Use Views and UDFs Strategically
Encapsulate business logic in reusable constructs so conversational queries stay aligned with engineering standards.

Think of this as training the AI the same way you train a new analyst, with examples, context, and guardrails.

Conversational Analytics Is a Strategy, Not a Feature

Conversational Analytics marks a fundamental shift in how organizations work with data. We are moving away from static dashboards, rigid filters, and long analytics backlogs toward a world where insight is accessed through conversation, direct, contextual, and immediate.

With Looker, business users gain a trusted, governed way to ask questions and explore data without breaking definitions or losing confidence. With BigQuery, analysts and engineers gain a powerful accelerator, one that transforms natural language into SQL, predictive models, and insights directly at the source. And with Gemini, these experiences are no longer fragmented; they are connected across the entire data stack.

Contact us today to help you enable and operationalize conversational analytics on Google Cloud, from configuring BigQuery and Looker to grounding Gemini with your business logic, data models, and governance standards, so your teams can use it confidently and at scale.