How to Create a Production Dashboard in Looker with AI

A production dashboard is your command center, giving you a real-time view into the health of your systems and applications. While powerful platforms like Looker can certainly build these, the process is often slow and heavily reliant on data specialists. This guide will walk you through the traditional steps of creating a production dashboard in Looker and then introduce a modern, AI-powered approach that gets you the same results in a fraction of the time.

What is a Production Dashboard and Why Do You Need One?

A production dashboard is a collection of charts, metrics, and logs that monitor the performance and stability of a live software application or system. Think of it as the EKG for your technology stack. Instead of guessing how your app is performing, you have concrete data telling you the full story.

These dashboards aren't just for show, they are critical tools for:

Engineers and DevOps Teams: To spot and diagnose issues like service outages, high error rates, or slow performance before they impact a large number of users.
Product Managers: To understand how system performance affects user experience and to correlate feature rollouts with performance metrics.
Business Leaders: To get a high-level view of system reliability, which directly impacts customer satisfaction and revenue.

Typically, a production dashboard tracks Key Performance Indicators (KPIs) like:

System Uptime: What percentage of the time are our services available?
Error Rate: How many errors are occurring, and where are they coming from?
Latency / Response Time: How long does it take for our application to respond to a user request?
Resource Utilization: How are our servers handling the load (CPU, memory, disk usage)?
Request Throughput: How many requests are our systems processing per minute or second?

Having this information centralized and visualized helps teams move from a reactive "firefighting" mode to a proactive state of monitoring and continuous improvement.

Building a Production Dashboard in Looker: The Manual Method

Looker is a powerful business intelligence platform, but it requires a structured, multi-step process to get from raw data to a finished dashboard. If you're tackling this yourself, here’s what the typical workflow looks like.

Step 1: Define Your Key Production Metrics

Before you even open Looker, you need a plan. A dashboard with a dozen random charts is just noise. Sit down with your team and decide what questions you need to answer. Start simple.

For example, a solid starter dashboard might answer:

What is our overall API error rate over the last 24 hours?
What is the average response time for our key endpoints?
Is our server CPU usage within a healthy range?

Having clear goals will guide every subsequent step and prevent you from building something that looks impressive but provides no real value.

Step 2: Connect Looker to Your Data Sources

Your production data lives somewhere - often in databases like PostgreSQL, Google BigQuery, or Amazon Redshift, or in logging platforms like Datadog or Splunk. Looker needs to be able to access this data.

In the Looker Admin settings, you’ll navigate to the Database section and create a new connection. This usually involves providing credentials like the host, port, database name, username, and password. Once the connection is tested and successful, Looker can begin querying your production data directly.

Step 3: Model Your Data with LookML

This is where the real technical work begins. Looker doesn’t just query raw tables, it relies on a modeling layer called LookML to define the business logic. LookML is a proprietary language that tells Looker how your database tables are related, how metrics should be calculated, and what fields are available for analysis.

Creating a LookML model involves:

Creating a Project: You’ll start a new LookML project and connect it to a Git repository for version control.
Generating Views: Looker can generate baseline "view" files from your database tables. A view file defines the dimensions (like timestamp, service_name, error_type) and measures (like count_of_errors, average_latency) for a specific table.
Customizing and Refining Views: You will need to manually write LookML code to create custom dimensions and measures. For instance, to calculate an error rate, you might create a measure that is SUM(case when error_code is not null then 1 else 0 end) / COUNT(*).
Building an Explore: An Explore defines how one or more views are joined together for analysis. For example, you might join a logs view with a services view to analyze log data specific to each microservice.

This stage is the most time-consuming and requires specialized knowledge. A clean, well-structured LookML model is the backbone of any good dashboard in Looker.

Step 4: Create "Looks" for Each KPI

Once your model is ready, you can start building individual visualizations, which Looker calls "Looks." Using the Explore you created, you can now start exploring your data with a point-and-click interface.

For each metric you defined in Step 1, you will:

Navigate to the appropriate Explore.
Select the dimensions and measures you need (e.g., Timestamp and Average Latency).
Apply necessary filters (e.g., filter for the last 24 hours).
Choose a visualization type (e.g., a line chart to show the trend over time).
Save the result as a Look.

You’ll repeat this process for every single chart you want to see on your dashboard - one Look for the error rate trend, another for current CPU usage, and so on.

Step 5: Assemble Your Looks into a Dashboard

With all your individual Looks saved, you can finally put them all together. You’ll create a brand new dashboard (which starts as a blank canvas) and add your saved Looks as "tiles."

You can drag and drop these tiles to arrange them, resize them, and add titles. You can also add powerful dashboard filters, like a date selector or a filter for a specific service, that apply to all the tiles at once, allowing for interactive exploration.

The Problem with the Traditional Path

While the process above works, it comes with significant challenges that many teams face:

High Technical Barrier: The biggest hurdle is LookML. It's powerful, but it's essentially its own programming language. Your marketing, product, and junior engineering team members can’t just jump in and create their own dashboards, they are entirely dependent on a short-staffed data team or a dedicated Looker developer.
It's Slow: From modeling the data to creating each individual Look and carefully arranging them on a dashboard, the process can take many hours, if not days. By the time the dashboard is ready, the immediate need for information may have passed.
It’s Rigid: When a new question comes up (e.g., "What was the error rate specifically for the user login flow last Tuesday?"), you can't just ask the dashboard. Answering that question might require creating a brand-new Look or even modifying the underlying LookML model, sending you right back into the development cycle.

A Faster Path: Using AI to Build Production Dashboards

What if you could skip most of the manual steps and get straight to the insights? That's the promise of modern AI-powered analytics platforms. Instead of requiring you to learn a complex tool, they allow you to simply describe what you want to see using plain English.

The workflow looks completely different:

Connect Your Data: This step is similar, involving a simple one-click authentication to connect your databases (like BigQuery or Snowflake) or platforms (like Datadog).
Ask for Your Dashboard: Instead of building everything piece by piece, you use a conversational prompt. You can simply ask: "Create a production dashboard showing me a timeline of our overall error rate, a bar chart of errors by service, and a number showing current CPU usage for the last 7 days."
Get Your Dashboard Instantly: The AI interprets your request, writes the necessary queries in the background, selects the best visualizations for each metric, and generates a fully interactive, real-time dashboard in seconds.

This approach transforms the entire experience. New ideas aren't met with days of development time but are answered immediately. You can continually refine your analysis with follow-up questions like, "Okay, now change the error rate chart to show the rate per hour," and watch as the dashboard updates instantly.

The AI handles the heavy lifting of data modeling and visualization, freeing your team to focus on interpreting the data and taking action.

Final Thoughts

Looker provides a robust, governable way to build production dashboards, but its power comes at the cost of speed and complexity, requiring deep technical knowledge and significant time investment. For teams that need to move fast and empower everyone to answer their own questions, the traditional BI workflow can feel like a major bottleneck.

Here at Graphed, we're building the tool we always wished we had. We made it possible to connect your sources and create real-time dashboards with simple, natural language prompts. It's built for engineers, product managers, and leaders who need immediate answers about their systems from their production databases but don't have the time to become LookML experts. Forget spending hours building reports, just ask your question and get an interactive dashboard back in seconds.