Data Engineering with AWS Chapter 12: Visualizing Data with Amazon QuickSight

This is post 18 in my Data Engineering with AWS retelling series.

We have spent eleven chapters ingesting data, transforming data, cataloging data, querying data. But here is a simple truth: nobody wants to stare at 10,000 rows in a spreadsheet. Our brains are not built for that. We process pictures way faster than text. A well-designed chart can tell you in two seconds what would take twenty minutes to figure out from raw numbers.

Chapter 12 is about that final step – taking all the data we have been building pipelines for and turning it into something a human can actually look at and understand quickly. The tool for the job is Amazon QuickSight.

Why Data Visualization Matters

Eagar starts with a claim you have probably seen on the internet: humans process images 60,000 times faster than text. He then immediately calls it out – there is no real evidence behind that specific number. But the general idea is solid. Think about how Instagram, Pinterest, emojis, and GIFs took over communication. Visuals just hit different.

For data, this means: if you need to spot a trend, find an outlier, or understand how your business is doing this quarter, a chart will always beat a table of numbers. You cannot look at a spreadsheet with thousands of rows and say “oh, sales dipped in March.” But put that data on a line chart and the dip jumps out at you instantly.

Common Chart Types

The chapter walks through three visualization types.

Line charts track things over time. The book uses AWS S3 costs over nine months – you can immediately see that API request costs dropped in February (the team started writing fewer, larger files) and storage costs dropped in March (lifecycle rules cleaned up old data). Two insights in one glance.

Geospatial maps overlay data on a map. Plot city populations as circles – bigger circle means bigger city. Instantly you see that Asia has the most mega-cities. Same approach works for disease outbreaks, product popularity by region, or temperature differences.

Heat maps show the relationship between two things using color intensity. The book plots product sales by category and month. Darker squares mean higher sales. Ice cream in summer, coffee in winter – obvious. But heat maps really shine when the patterns are not obvious and the visual reveals something your gut missed.

What Is Amazon QuickSight?

QuickSight is the AWS answer to tools like Tableau, Power BI, and Qlik. It is a fully managed, serverless BI service. No servers to provision, no infrastructure to manage. You pay per user on a subscription basis.

There are two editions: Standard and Enterprise.

Standard costs about $9 per user per month. Every user gets full author capabilities. Good for small teams getting started.

Enterprise adds the serious stuff: Active Directory integration, the ability to embed dashboards in your own apps, scheduled email reports, fine-grained access control, ML-powered insights, and encryption at rest. It also splits users into two roles:

• Authors ($24/month) – create and edit dashboards
• Readers ($0.30 per session, capped at $5/month) – view and interact with dashboards but cannot create new ones

That reader pricing is clever. Five hundred people who glance at a dashboard once a week cost a few dollars each, not $24. For large organizations with many casual dashboard consumers, this matters.

SPICE: The Speed Engine

SPICE stands for Super-fast, Parallel, In-memory, Calculation Engine. It is QuickSight’s built-in data storage layer.

When you create a dataset, you have two choices. Option one: query the source directly every time someone opens the dashboard. Freshest data, but more latency and a two-minute timeout. Option two: import into SPICE. Much faster reads, with scheduled refreshes (daily, weekly, monthly, or every 15 minutes on Enterprise). You can also trigger refreshes via API.

Every author user gets 10 GB of SPICE storage. Ten authors means 100 GB shared across them. Need more? Buy extra at around $0.38 per GB on Enterprise.

Bonus: if your source is Athena or Redshift Spectrum (which charge per query), SPICE means you only pay for the query at load and refresh time, not every time someone opens a dashboard.

Data Sources and Preparation

QuickSight connects to a lot of things: RDS, Redshift, Athena, S3, Aurora, Snowflake, Teradata, Oracle, SQL Server, Postgres, MySQL (on-premises or EC2), and SaaS platforms like Jira, Salesforce, GitHub, and Twitter. You can also upload Excel, CSV, and JSON files directly. For unsupported sources, ingest into S3 first and use Athena as the bridge.

QuickSight includes basic data preparation – joining datasets, filtering rows, dropping columns, renaming fields, changing data types, calculated fields. But should you do transformations inside QuickSight or outside it?

Eagar’s guidance: if the dataset will be used by other tools, do the ETL outside QuickSight using Glue or DataBrew. If the transforms are simple and the data is QuickSight-only, doing it inside is fine. The real tension is between letting business users move fast and maintaining governance for datasets that drive important decisions.

Analyses, Dashboards, and Visual Types

You create an analysis (your workspace), build visuals inside it (up to 50 datasets, 20 sheets, 30 visuals per sheet), then publish it as a dashboard for readers. When publishing, you control what readers can do – apply filters, download CSVs, drill down, sort.

QuickSight supports a long list of visual types:

• AutoGraph – let QuickSight pick the best chart type based on your data
• Line charts, bar charts, geospatial maps, heat maps – the classics
• KPIs – show progress against goals with a percentage and progress bar
• Tables – raw data display, best for small datasets
• Pie charts, box plots, gauge charts, histograms, pivot tables, Sankey diagrams, tree maps, word clouds – and more

You can also embed custom content like images, videos, web pages, and forms directly into a dashboard. Though these custom visuals will not show up in emailed copies of the dashboard.

ML Insights: Machine Learning Without the Machine Learning

This is an Enterprise-only feature and it is genuinely useful. ML Insights uses machine learning under the hood, but the person creating the dashboard does not need to know anything about ML algorithms.

Autonarratives generate plain-language summaries of your data. Instead of just a chart, you get a sentence like: “Year-to-date revenue decreased by 4.6% from $906,123 to $864,441 compared to the same period last year.” That is something you can paste into a report or email directly.

Anomaly detection scans across millions of metrics and flags things that look unusual. You configure how often it runs (hourly to monthly) and it highlights outliers you might never spot manually.

ML-powered forecasting takes historical data and predicts future values. The model automatically handles outliers (ignoring that one-time sales spike from a promotion) and fills in missing data points. You need a minimum amount of historical data – at least 38 daily data points or 35 quarterly ones.

Embedded Dashboards

For organizations that do not want users logging into the AWS console, QuickSight supports embedding dashboards directly into your apps or website.

For registered users, you integrate with existing authentication (AD, SAML 2.0, Okta, Auth0). Users log into your app and see QuickSight dashboards as a native part of the experience. You can customize branding, though a “Powered by QuickSight” label appears.

For anonymous users, you purchase reader session capacity. Perfect for public-facing dashboards – like a health department sharing outbreak data. No login required, full interactivity. An annual plan even lets you remove the QuickSight branding.

The Hands-On Exercise

The practical section creates a geospatial visualization from scratch. You download a world cities dataset from simplemaps.com (free, Creative Commons licensed), sign up for QuickSight Standard (30-day free trial), and upload the CSV – about 41,000 rows that get imported into SPICE automatically.

You create a “Points on Map” visual: latitude and longitude for location, population for circle size, city name for color. Add a filter for population >= 3,000,000 and you get a clean map of every mega-city in the world. You can zoom, filter, and interact with it in real time.

Quick reminder from Eagar: unsubscribe after the free trial if you do not plan to keep using QuickSight. Those subscription costs add up quietly.

Key Takeaway

QuickSight is the last mile of your data pipeline. You spent chapters building ingestion, transformation, cataloging, and querying. Now you need to put that data in front of the people who make decisions.

The tool is solid – serverless, pay-per-user, connects to practically everything, SPICE keeps dashboards fast, ML Insights brings machine learning to non-technical users, and embedded dashboards extend reach beyond AWS.

But the bigger lesson is not about QuickSight specifically. All the engineering work only matters if someone can look at the output and make a better decision because of it. A beautiful pipeline that nobody can understand the results of is just expensive plumbing.

Book: Data Engineering with AWS by Gareth Eagar | ISBN: 978-1-80056-041-3

Previous: Chapter 11: Ad Hoc Queries with Amazon Athena Next: Chapter 13: Enabling AI and Machine Learning