exploring-data-for-looker

Onboarding Interview & Data Exploration

This skill guides you through the initial phase of onboarding a Looker user: interviewing them to understand their goals, and exploring their BigQuery data to identify a good candidate for a dashboard.

Prerequisites

• The user must have a fresh Looker Core instance.
• The user must have gcloud installed and authenticated to GCP.

Instructions

1. Interview the User (Blocking Step)

You MUST start by interviewing the user to understand their goals and get the target GCP project and dataset. DO NOT run any gcloud or bq commands before asking these questions and receiving the user's explicit response.

Propose a set of questions like:

"Welcome to Looker! To help you get started, I'll guide you through connecting your data, setting up a project, and building your first dashboard.

To make sure we build something highly relevant to you, could you tell me: 1. What is the main goal of your Looker exploration? (e.g., Sales analysis, User behavior, Inventory tracking, Support metrics) 2. Which BigQuery project and dataset contains the data you want to use? 3. Are there specific metrics (e.g., total revenue, active users) or dimensions (e.g., product category, region) you are most interested in?"

STOP AND WAIT for the user to reply. Do not call any tools in this turn.

2. Explore BigQuery Data

Only after the user has responded to your interview questions with the target project/dataset, proceed to explore their BigQuery data using the bq CLI.

[!IMPORTANT] Do NOT assume any table naming conventions (such as dim_ or fct_ prefixes). While some datasets use them, many do not. You must identify dimension and fact tables by analyzing their schema (columns, types) and content, not just their names.

1. List Datasets in the project:

   bq ls --project_id={gcp_project_id}
   

   Look for datasets that match the user's business area (e.g., ecommerce, analytics).

2. List Tables in the selected dataset:

   bq ls {gcp_project_id}:{dataset_name}
   

   Identify candidate tables:

   • Dimension tables contain reference or lookup data (e.g., users, products, categories, locations). Look for tables that describe entities.
   • Fact tables contain transactional or event data (e.g., orders, order_items, events, sessions, payments). Look for tables that contain numeric metrics and timestamps.

3. Examine Table Schemas for key candidate tables:

   bq show --format=prettyjson {gcp_project_id}:{dataset_name}.{table_name}
   

   Look for timestamps, numeric fields for aggregations, and foreign keys that link tables.

3. Propose the Onboarding Goal

Based on the user's input and your data exploration, propose a specific, actionable dashboard goal.

Example proposal:

"I've explored your dataset {dataset_name} and found some great tables: - {table_1} (contains transactional data with revenue and timestamps) - {table_2} (contains product dimensions like category and brand)

I propose we target a Sales Overview Dashboard as our onboarding goal. It will include: 1. Total Revenue over time (daily/weekly). 2. Top Product Categories by sales volume. 3. Order Status Distribution (e.g., completed, pending, cancelled).

If this sounds good, I will proceed with setting up the Looker CLI and connecting this data!"