BigQuery Now Talks Back: Google Makes GenBI a Reality

Natural language questions, multi-turn conversations, and native AI functions. What we predicted a week ago is now in production.

---

A few days ago I wrote about GenBI and how it would change the data analyst’s job. The thesis was simple: instead of writing SQL, you’ll talk to your data in natural language.

Well, Google just launched exactly that in BigQuery. And it’s not an experiment: it’s in production, available to anyone with an account.

What Google launched

BigQuery now includes conversational analytics with these capabilities:

Natural language questions. You type “what were Q4 sales by region?” and BigQuery generates the SQL query, executes it, and returns the result. No code required.

Multi-turn conversations. The system remembers context from previous questions. You can say “now show me just Europe” and it understands you’re referring to the previous query. Like talking to a human analyst.

Custom agents. You can create and deploy conversational agents via API, trained with your specific data context. It’s not a generic chatbot: it understands your schema, your metrics, your business.

Looker integration. Looker dashboards can now be queried conversationally. Business users ask, the system responds with visualizations.

Native AI functions. Three new SQL functions:

• AI.IF: conditionals based on semantic understanding
• AI.CLASSIFY: automatic text classification
• AI.SCORE: predictive scoring integrated into queries

Why this matters

This isn’t Google playing with demos. It’s the real implementation of the GenBI paradigm on one of the most widely used data platforms in the world.

For data analysts

Your role just changed. I’m not saying you’ll lose your job tomorrow, but tasks like “translate business questions to SQL” are about to decrease dramatically.

What’s increasing: designing the data models that feed these tools, validating that responses are correct, and solving complex cases AI can’t handle. If you’re interested in how the role is evolving, I wrote about the analyst to data engineer transition.

For business users

For the first time, you can query data without depending on anyone. No need to open a ticket, wait for the analyst to have time, explain what you want three times.

You open BigQuery (or Looker), ask, get an answer. In seconds.

For companies

The “overwhelmed data team” bottleneck shrinks. Routine questions become self-service. Analysts are freed for higher-value work.

But be careful: this only works if your data is well organized. Conversational BigQuery doesn’t fix dirty data, chaotic schemas, or poorly defined metrics. Garbage in, garbage out, even if you ask in plain English. If you want to dive deeper into why data quality is fundamental, read about Data-Centric AI.

AI functions in SQL

This is what interests me most technically. Google has embedded LLM capabilities directly into the SQL engine.

AI.CLASSIFY lets you classify text without training models:

SELECT
  comment,
  AI.CLASSIFY(comment, ['positive', 'negative', 'neutral']) as sentiment
FROM reviews

AI.SCORE generates predictive scores:

SELECT
  customer_id,
  AI.SCORE(features, 'churn_probability') as churn_risk
FROM customers

AI.IF enables semantic conditionals:

SELECT
  description,
  AI.IF(description, 'mentions delivery problems') as has_shipping_complaint
FROM tickets

This eliminates the need for separate ML pipelines for many use cases. Analysis and prediction happen in the same query.

What they don’t tell you

Requires well-structured data

Conversational analytics works on your existing schema. If your schema is a mess of poorly named tables, undocumented fields, and implicit relationships, the system won’t understand your questions.

Before getting excited about “talking to your data,” make sure your data is in a condition where someone (human or AI) can actually understand it.

Costs can spike

Every conversational query involves AI model calls. BigQuery charges for processing. If you have 500 users asking questions all day, the bill might surprise you.

Monitor usage before opening the door to the entire organization.

Validation is still necessary

Just because the system generates SQL doesn’t mean it generates correct SQL. For simple questions, it works well. For ambiguous or complex questions, it can misinterpret.

You need a validation process, especially for important decisions based on this data.

How to get started

If you already use BigQuery:

1. Enable AI features in your project (requires activating specific APIs)
2. Start with simple queries on well-documented tables
3. Validate responses against manual SQL queries
4. Iterate and learn the system’s limits

If you don’t use BigQuery:

This isn’t exclusive to Google. Snowflake has Cortex, Databricks has similar assistants, and there are independent tools like Text2SQL.

The trend is clear: all data warehouses are moving toward conversational interfaces. If your current platform doesn’t have it, it will soon.

The future that already arrived

A week ago, GenBI was a trend. Today it’s a production product from one of the largest companies in the world.

The speed at which this is moving is brutal. What was a prediction in January is reality in February.

For those of us who work with data, the message is clear: adapt or fall behind. The tools are changing. The role is changing. Those who learn to work with these new capabilities will be more valuable. Those who ignore them… well, you know how that story ends.