How to Do Data Analysis in Looker with AI

Doing data analysis in Looker is getting a major upgrade, moving beyond clicking fields and filters into the world of conversational AI. If you've spent hours dragging and dropping dimensions to build the perfect report, you'll want to pay close attention to Google's Duet AI finding its way into your Looker dashboards. This article will walk you through what Looker's AI features are and how you can use them to build reports, create calculations, and find insights much, much faster.

Graphed

Still Building Reports Manually?

Watch how growth teams are getting answers in seconds — not days.

Watch the Demo

What is Looker's AI? Untangling "Duet AI"

You may have heard the term "Duet AI" popping up across Google's suite of products, from Gmail to Google Slides. Think of Duet AI as Google's answer to helpers like Microsoft's Copilot - it's an embedded, generative AI assistant designed to streamline your workflow within the tools you already use.

Within Looker, "Duet AI in Looker" is the specific application of this technology. Its main purpose is to make data interaction more natural and intuitive. Instead of needing extensive training on where every button is or how every Looker function works, you can simply tell the platform what you want in plain English. This is a big deal for both experienced analysts and casual business users.

For business users - the marketers, sales managers, and operators - it means getting answers without having to perfectly build a query in Explore. For the data teams and developers, it means speeding up tedious tasks like writing Looker's modeling language (LookML) or generating complex formulas.

Specifically, the key features powered by Looker AI include:

• Conversational Queries: Ask questions in natural language and get back a visualized answer instantly.
• AI-Assisted Calculations: Describe a calculation you need, like "percentage of total," and have Looker write the formula for you.
• Dashboard Suggestions: Get recommendations for charts and layouts as you build dashboards to better visualize your data.
• LookML Development Support: Accelerate data modeling by automatically generating, explaining, or refining LookML code.

Graphed

Still Building Reports Manually?

Watch how growth teams are getting answers in seconds — not days.

Watch the Demo

Getting Started: What You Need Before Using Looker AI

Before you can start chatting with your data, there are a few prerequisites you and your team need to have in place. AI in Looker isn't a magic wand that works out of the box on day one, it relies on a solid data foundation that an administrator needs to set up correctly.

Here’s a simple checklist of what's required:

1. Google Cloud-Hosted Instance: Looker's AI capabilities are powered by Google's backend infrastructure, so your Looker instance must be hosted on Google Cloud. If your instance is self-hosted or hosted on another cloud provider, you won't be able to access these Duet AI features.
2. Duet AI Enabled: A Google Cloud administrator must enable the Duet AI API for the specific project that your Looker instance is connected to. They might also need to accept certain data processing terms within your Google Cloud Console.
3. The Right Permissions: Not everyone gets access to AI features automatically. A Looker admin needs to grant specific permissions to users or groups who should be able to use the AI capabilities. If you don't see the features, check in with your administrator.
4. A Well-Defined LookML Model: This is the most crucial part. The AI isn't guessing, it's using your company's LookML data model as its "brain." Your LookML model acts as a semantic layer, defining your business logic - like what "revenue" is, how to join users to orders, and the names of all your metrics. A clean, well-documented, and user-friendly LookML model will lead to fantastic AI results. A messy or poorly defined one will lead to confusion and inaccurate answers. Garbage in, garbage out.

How to Use Natural Language to Ask Questions in Looker

The most immediately useful feature for the majority of users is the ability to ask for data using simple conversational prompts. This functionality is being integrated directly into Looker’s "Explore" interface, where you typically go to build reports.

Step-by-Step Guide to Conversational Explores

Imagine you're a marketing manager and you want to check last week’s website performance without hunting for the right dimensions and measures.

1. Navigate to an Explore

Start just as you normally would. Go to the Explore section that contains the data you're interested in, such as "Website Analytics" from a Google Analytics connection.

2. Open the Conversational Prompt

Instead of searching through the field picker on the left, you'll see a prompt bar or a "chat" icon. This is your command center. Click on it to open the natural language interface.

3. Ask Your Question in Plain English

Type what you want to know. You can start simple. For instance: Show me my total website sessions by day for last week

Duet AI will then take that prompt and try to figure out what you mean. In the background, it's translating your phrase "total website sessions" into the LookML field named ga4_sessions and "last week" into the correct date filter. It then automatically selects the fields and filters, runs the query, and presents you with the result.

4. See the Results and Iterate

Looker will generate the data and a suggested visualization - in this case, probably a line chart. You'll see the sessions plotted for each day of the previous week.

This is where the real power lies. You can now build on this. You might follow up with a refining question: break this down by device category

The AI understands the context of your previous query. It won't start from scratch, instead, it will add a "Device Category" dimension to the existing report. Now, your line chart will show separate lines for desktop, mobile, and tablet traffic.

You can keep going. Ask questions like:

• "What were my top 10 countries by users last month?"
• "Compare revenue for Q1 this year vs Q1 last year"
• "Show me the average order value from Google Ads traffic"

Each time, the AI translates your request into the appropriate Looker components, saving you the time and mental energy of doing it manually.

Graphed

Still Building Reports Manually?

Watch how growth teams are getting answers in seconds — not days.

Watch the Demo

Automating Formulas with AI-Assisted Calculations

Another common sticking point for Looker users is creating custom fields using "Table Calculations." These are incredibly powerful for on-the-fly analysis, letting you calculate things like period-over-period growth or percentages without changing the underlying LookML. But they require you to learn Looker’s specific formula syntax, which can be unintuitive. Duet AI now helps write these formulas for you.

How to Create Calculations with AI

Let's say you've already created a report showing your monthly revenue, and now you want to see the month-over-month growth rate.

1. Get Your Base Data:

First, create a simple explore showing "Total Revenue" by "Month."

2. Open Table Calculations:

In the data table, click the "Calculations" button. Instead of staring at a blank formula box, you'll now see a generative AI option.

3. Describe the Formula You Want:

Prompt the AI with a description of the desired calculation. For example: Calculate the percent difference between this month's revenue and last month's

4. Generate and Use the Formula:

Looker AI will analyze your prompt in the context of your data and generate the proper Looker expression. It might produce something like: (${revenue} - offset(${revenue}, -1)) / offset(${revenue}, -1)

Gone is the need to remember difficult functions like offset(). You can click "Run" to apply the calculation and see your month-over-month growth percentage appear as a new column in your table.

Graphed

Still Building Reports Manually?

Watch how growth teams are getting answers in seconds — not days.

Watch the Demo

Best Practices for Prompting Looker's AI

While the AI is smart, you can get better, more accurate results by understanding how it "thinks." It works best when you provide it a clear path.

• Be Specific and Clear: Vague prompts lead to vague outcomes. "Show sales" is much weaker than "Show total sales amount grouped by product category for the last 30 days." Adding details like aggregation type (total, average), grouping fields (category, state), and timeframes gives the AI definitive instructions.
• Start Simple, Then Refine: Don't try to ask a complex, multi-layered question in one go. Start with a foundational request (e.g., "users by country"). Once you have that, add layers with follow-up prompts ("now filter for only mobile users" or "change the chart to a map"). This iterative process mirrors how a human analyst explores data.
• Use Your Model's Language (When You Can): The AI will work better if you occasionally use the actual names of dimensions or measures from your LookML model. For example, if your revenue field is called "Total Order Value," using that term in your prompt can help Looker find the exact match faster.
• Check Its Work: The AI translates your prompt into selected fields and filters. Always give the left-hand field selector a quick glance after the AI runs a query to confirm it grabbed the data you expected. Trust but verify.

Final Thoughts

Ultimately, incorporating AI into Looker makes data available to a wider audience and makes expert users tremendously more efficient. What was once a tedious process of clicking, dragging, and searching through hundreds of fields is now becoming a simple conversation, allowing everyone to focus more on drawing insights from the data rather than on the mechanics of building a report.

While Looker is making great strides in adding AI, it works best when you have a dedicated data team to set up and maintain a high-quality LookML model. For teams without those resources, getting all your data from different sources like Google Analytics, Shopify, and Salesforce ready for analysis can still be a heavy lift. We built Graphed because we believe asking questions about your business shouldn't require such a complicated setup. It lets you connect all your marketing and sales accounts in just a few clicks and immediately start asking questions to create live dashboards - all without ever having to write a line of code or build a data model.