How to power AI agents with Fivetran and Google Cloud

Agentic AI is changing how teams interact with data. Instead of relying on static dashboards or waiting for analysts to write queries, business users can ask complex questions in natural language and get actionable answers in minutes. For example, a revenue leader could ask, “Which high-value customers are at risk of churn, what’s driving that risk, and which accounts need immediate action?” and get a complete answer without manually pulling reports from multiple systems.

That’s the workflow I walked through in our recent webinar, Ask Your Data Anything: How Fivetran, dbt, and Knowledge Catalog power AI agents on Google Cloud. I showed how AI agents can query data across systems, apply business logic, explain results, and even trace answers back to the source.

[CTA_MODULE]

7 steps to power AI agents using Fivetran and Google Cloud

Here are 7 steps data teams can take to start building agents that are grounded in trusted, governed, and business-ready data using Fivetran, dbt, and Google Cloud.

1. Connect enterprise data into Google Cloud

The first step is getting data from across the business into a place where AI agents can use it.

Use Fivetran UI/API to automate this process by syncing data from operational systems, applications, databases, and APIs into Google Cloud.

2. Store data in an open, AI-ready foundation

Once the data is synced into Google Cloud Storage, it is exposed automatically via Lakehouse for Apache Iceberg to any applicable GCP service such as BigQuery. Fivetran hosts the primary catalog, Polaris, external of Lakehouse.

This matters because AI architectures are changing quickly. An Open Data Infrastructure gives organizations flexibility and interoperability: BigQuery, agents, applications, and other compute engines can all work from the same trusted data foundation.

Fivetran manages the complexity behind the scenes, including file creation, updates, snapshots, and maintenance, while the customer retains ownership of the data in their own Google Cloud environment.

3. Standardize data with consistent models

AI agents need clean, consistent data to reason accurately.

That means standardizing fields, names, and business logic across systems. For example, an organization does not want one system using “customer_id,” another using “cust_id,” and another using “cid.” Agents need consistency so they can understand relationships across datasets.

Using dbt, teams can create staging models, semantic views, and reusable business logic. This keeps definitions like customer lifetime value, churn risk, customer tier, and support health in one governed place. It also means that when business logic changes, teams can update the model once, redeploy it, and the agent can use the updated version automatically.

4. Use metadata to tell the agent where to look

One of the hardest parts of enterprise analytics is knowing which data to use. Google Knowledge Catalog helps solve this by giving agents business context. Instead of hardcoding every query path into the agent, teams can define glossary terms, descriptions, relationships, and documentation that guide the agent.

For example, when a user asks about churn risk or customer lifetime value, the agent can use the catalog to understand which views, tables, and definitions apply. This helps the agent answer questions more accurately without requiring users to know the underlying data structure.

5. Build a multi-tool agent

At the front end, Gemini Enterprise provides a simple natural-language interface for the user. Behind the scenes, a backend agent executes the actual work across Google Cloud systems.

This backend agent can query BigQuery, read from Knowledge Catalog, access lineage, combine data across systems, and return a structured answer.

That separation is important. The user experience stays simple, while the backend can evolve as data models, tools, and workflows change.

6. Capture lineage from source to answer

For AI agents to be trusted, they need more than access to data. They need to explain where the answer came from.

Google Cloud can trace lineage across tables and views inside BigQuery. But by default, that only shows what happens inside the GCP project.

Fivetran fills the upstream gap. Through the Fivetran Platform Connector, pipeline metadata is written directly into BigQuery, including connector run history, row movement metrics, schema changes, table lineage, transformation activity, and much more.

When Fivetran metadata is combined with Google Cloud lineage, via the OpenLineage specification, organizations can create an end-to-end view of how data moved from the source system, into Google Cloud, through transformations, and into the final datasets consumed by the agents.

That means users can ask not only, “What is the answer?” but also, “Where did this answer come from?”

7. Let agents explain and act

Once the foundation is in place, agents can do more than summarize data.

They can identify high-value customers at risk of churn, explain what is driving the risk, connect support tickets to customer health, surface order issues, recommend next steps, and help teams act faster.

In an operational workflow, an agent could flag an at-risk account, draft an outreach email, notify the account team, or trigger a downstream process. In an analytics workflow, it could replace days of manual analysis with a natural-language question and a trusted, traceable answer.

Turning AI agents into trusted business tools

Production AI requires more than a model and a chat interface. It requires a reliable data foundation.

Fivetran provides the automated data movement, freshness, metadata, and pipeline visibility. dbt standardizes and models the data into consistent, reusable business semantics that agents can understand and trust. Google Cloud provides the storage, query engine, governance, catalog, and agent interface. Together, they help organizations build AI agents that are grounded in trusted enterprise data.

The result is faster insight, less manual data work, and AI agents that users can actually trust.

When agents can access fresh data, understand business context, and explain lineage, they stop being black boxes. They become reliable partners for analytics, operations, and decision-making.

[CTA_MODULE]