Cloud concept · Updated

Analytics - Analytics is the systematic examination of data to answer questions, identify patterns, and support decisions.

In short

Analytics turns raw records, events, and measurements into useful information through querying, aggregation, statistical analysis, and visualization. In the cloud, managed storage and processing services let engineers analyze changing data volumes without building every infrastructure component themselves.

Cloud Concepts Hub →

6 key aspects of Analytics

Effective analytics combines clear questions, reliable data preparation, suitable processing, cost-aware design, and governance to produce trustworthy answers.

01
Question-driven analysis

Start with a decision or measurable question, then identify the required datasets and calculations. Define metrics such as revenue or active users consistently so different reports do not produce conflicting answers.

02
Data preparation and discovery

Clean, transform, and catalog data before relying on results. Validate schemas, handle duplicates, and document dataset ownership and meaning.

03
Workload-appropriate processing

Choose batch processing for scheduled summaries and streaming processing when fresh results justify the additional complexity. Query engines, warehouses, and distributed processing frameworks address different workload patterns.

04
Separation of storage and analysis

Many cloud architectures retain datasets in durable storage while using separate compute services to process them. This allows multiple analysis tools to reuse data, but requires attention to formats, access permissions, and compatibility.

05
Performance and cost management

Use appropriate file formats, partitioning, aggregation, and compute sizing to control scan volume and execution time. Measure actual query patterns rather than assuming managed or serverless services are automatically inexpensive.

06
Governance and trustworthy results

Apply least-privilege access, encryption, retention rules, and data-quality checks throughout the pipeline. Track data origins and transformations so engineers can investigate unexpected results.

Cloud services for Analytics

An analytics architecture can store source data in Amazon S3, use AWS Glue to catalog and prepare it, and use Amazon Athena for exploratory SQL queries. Amazon Redshift can support a warehouse-centered approach for recurring analytical workloads, while Amazon EMR can run distributed processing jobs when Spark or similar frameworks fit the requirements. Amazon Kinesis Data Streams supplies continuously arriving events to processing applications, and Amazon QuickSight presents prepared results through dashboards. These services are not all required in every solution: choose components according to data volume, freshness requirements, query patterns, governance needs, and operating cost.

Amazon AthenaAmazon Athena

Amazon Athena provides serverless, interactive SQL querying, commonly over data stored in Amazon S3. It suits ad hoc analysis without provisioning a query cluster; file layout, compression, and partitioning influence performance and scan-based query costs.

Amazon RedshiftAmazon Redshift

Amazon Redshift is a managed data warehouse for analytical SQL workloads, including aggregations and joins across large datasets. Provisioned and serverless options support workloads such as recurring reporting and business intelligence.

AWS GlueAWS Glue

AWS Glue provides data integration capabilities for discovering, preparing, and transforming datasets. Its Data Catalog stores metadata that analytical services can use to locate and interpret tables.

Amazon EMRAmazon EMR

Amazon EMR runs big-data processing frameworks such as Apache Spark for large-scale transformations and analysis. It is relevant when engineers need distributed processing frameworks and control over application code and execution settings.

Amazon Kinesis Data StreamsAmazon Kinesis Data Streams

Amazon Kinesis Data Streams collects and retains streaming records for applications that process continuously arriving events. Consumer applications perform the analysis; the stream itself does not create business metrics or dashboards.

Amazon QuickSightAmazon QuickSight

Amazon QuickSight provides business intelligence capabilities for creating visualizations and interactive dashboards from connected datasets. It presents analytical results to users rather than replacing data ingestion, transformation, or storage.

Analytics exam questions

Real practice exam questions on Analytics, with the correct answer and its explanation. Answer them with every option in the exam simulator.

DEA-C01

A support company stores tickets in an Amazon Redshift table named tickets(ticket_id, region, agent_id, priority, opened_at). Each ticket has a unique, non-null ticket_id and a priority of LOW, MEDIUM, or HIGH. Each region has multiple agents, and each agent handles many tickets.

A data engineer must create a report for January 2026 with exactly one row per region that had tickets during the month. The report must contain separate LOW, MEDIUM, and HIGH columns showing the number of tickets at each priority, regardless of which agent handled them.

Which query meets these requirements?

Show answerCORRECT ANSWER

SELECT * FROM ( SELECT region, priority, ticket_id FROM tickets WHERE opened_at >= '2026-01-01' AND opened_at < '2026-02-01' ) AS src PIVOT ( COUNT(ticket_id) FOR priority IN ('LOW', 'MEDIUM', 'HIGH') );

The query filters tickets to January and pivots the priority values into columns. COUNT(ticket_id) counts every ticket because ticket_id is non-null. Region is the only input column outside the pivot column and aggregate argument, so the result has exactly one row per region.

AIP-C01

A company stores evaluation results for two foundation model configurations in Amazon S3. An inference table contains exactly one row per evaluated case, including the configuration, inference cost, and completion timestamp. A separate review table contains scores for three required criteria per case. Reviewers can revise scores, and each revision is appended with a unique revision timestamp. Some cases are still awaiting reviews.

The company uses Amazon Athena and Amazon QuickSight to produce a weekly model-comparison dashboard and an emailed management summary. The reports must show total inference spend for all cases completed that week, the pass rate among fully reviewed cases, and the number of cases awaiting complete review. A case passes only if its latest score for every required criterion is a pass.

An initial implementation joins the tables on case ID, but reported spend increases when reviewers add or revise scores, even though no new inference requests occurred.

Which reporting design meets the requirements?

Show answerCORRECT ANSWER

Use Athena to retain the latest review row for each case and criterion, then summarize those rows into one record per case with review-completeness and pass indicators. Left join that summary to the inference table and use the resulting case-level dataset for QuickSight visualizations and scheduled email reports.

Selecting the latest revision per case and criterion preserves the current evaluation outcome. Summarizing before joining prevents review rows from multiplying inference costs. A left join retains cases with no reviews, allowing all inference spend to be counted while incomplete cases are reported separately and excluded from the pass-rate denominator.

DOP-C02

A company sends application request logs from hundreds of microservices to Amazon Kinesis Data Streams. Each record contains a service ID, an event timestamp, and a response status. Records for the same service can arrive on different shards and can arrive up to 90 seconds out of order.

A DevOps engineer must continuously calculate each service's error rate over a 5-minute window that advances every minute. Windows must use the event timestamps, include the out-of-order records, and produce results within 2 minutes after each window ends during normal operation. If the processing application fails, it must recover its aggregation state and stream positions without requiring the team to implement an external state-management system.

Which solution meets these requirements?

Show answerCORRECT ANSWER

Use Amazon Managed Service for Apache Flink to consume the stream. Assign event timestamps and a 90-second bounded-out-of-orderness watermark strategy. Key records by service ID, apply 5-minute sliding event-time windows with a 1-minute slide, and enable checkpointing for recovery.

Event-time windows group records by when requests occurred, and the watermark strategy accommodates the specified out-of-order arrival interval. Keying by service ID combines records across input shards. Sliding windows provide the required duration and evaluation frequency. Checkpoints preserve aggregation state and source positions for consistent recovery without a custom external state-management system.

Practice Analytics in a full exam

Questions on Analytics in each practice exam, with every answer option and an explanation for each:

Start a free practice exam

What is Analytics

2 min read

Analytics helps organizations understand what happened, investigate why it happened, and make better-informed decisions. Its inputs can include application logs, sales transactions, website events, sensor readings, and operational metrics. Its outputs might be a SQL query result, a dashboard, a report, or a signal that triggers further investigation. For example, a retailer could combine order records with website events to measure conversion rates and identify where customers abandon purchases.

A cloud analytics workflow usually involves collecting data, storing it, preparing it, analyzing it, and presenting the results. Preparation includes correcting data types, removing duplicates, handling missing values, and joining records from different sources. Extract, transform, load, or ETL, transforms data before loading it into a destination. Extract, load, transform, or ELT, loads data first and performs transformations in the destination. A data catalog records metadata such as table definitions and locations so tools and users can discover and interpret datasets.

Storage and query choices depend on the workload. A data lake commonly stores varied datasets in object storage, often retaining original data alongside curated versions. A data warehouse provides a managed analytical environment optimized for queries such as aggregations and joins across large datasets. A lake is not automatically clean or well governed, and a warehouse is not simply an application database with more capacity. Analytical processing typically scans and summarizes many records, while transaction processing emphasizes short reads and writes that support application operations.

Batch analytics processes accumulated data on a schedule, such as calculating yesterday's revenue each morning. Streaming analytics processes continuously arriving events to provide fresher results, such as detecting a sudden increase in failed payments. Streaming introduces concerns including late events, duplicates, processing delays, and recovery after failures. Choose a freshness target that the business actually needs: a daily financial report does not necessarily justify the complexity and expense of a continuously running pipeline.

Cloud services implement different parts of this workflow. In AWS exam scenarios, Amazon Athena is associated with serverless SQL queries over data in Amazon S3, while Amazon Redshift is associated with data warehousing. AWS Glue supports data integration and cataloging, Amazon EMR supports frameworks such as Apache Spark, Amazon Kinesis Data Streams collects streaming records for consumers, and Amazon QuickSight provides business intelligence dashboards. These roles are complementary; collecting a stream is not the same as analyzing it, and displaying a chart does not prepare its underlying data.

Important trade-offs include cost, query latency, data freshness, concurrency, and operational effort. Columnar formats such as Parquet and suitable partitioning can reduce the data scanned by analytical queries. Serverless execution reduces infrastructure management but does not guarantee low cost or fast queries: inefficient scans, repeated transformations, and excessive dashboard refreshes still consume resources. Engineers should also validate completeness and accuracy, define metrics consistently, restrict access to sensitive records, and apply retention policies. Analytics does not inherently require machine learning, and a correlation in a dashboard does not establish causation.

Learn Analytics in lessons

3 short lessons on Analytics in 3 certification courses.

Open the lessons