Analytics

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.