Analytics: Descriptive, Predictive & Prescriptive

From reporting to recommending

Analytics is the systematic analysis and interpretation of data — using mathematical, statistical, and computational tools — to improve understanding of a domain. It evolved from Decision Support Systems (1960s) through Business Intelligence (late 1980s) to the broad umbrella term analytics (mid-2000s on). Data warehousing (Module 9) is its infrastructure foundation. Analytics is conventionally split into three types, in increasing sophistication and business value:

• Descriptive analytics — "What happened?" Describes the past status of the domain using reporting, dashboards, scorecards, data visualization, and OLAP. This is the most mature capability in most organizations.
• Predictive analytics — "What might happen?" Applies statistical/computational models to past and current data to forecast future events. Backward-looking in its data (needs history to train), forward-looking in its output (probabilistic predictions). Data mining lives here. Tools: R, Python (scikit-learn), SAS Enterprise Miner.
• Prescriptive analytics — "How can we make it happen?" Uses predictive results together with optimization and simulation to recommend specific actions that achieve a desired outcome. The most sophisticated type; it often includes a feedback loop that learns from the actions taken.

Worked example: classify three goals at a hospital

Goal                                         Type           Question        Technique
----------------------------------------------------------------------------------------
Report patient outcomes by department        descriptive    What happened?  OLAP / dashboard
Flag patients at high risk of 30-day         predictive     What might      data mining /
  readmission                                                happen?         ML model
Automatically route high-risk patients to    prescriptive   How to make     optimization /
  the right intervention program                            it happen?      simulation + rules

The ladder is cumulative: you cannot predict without first describing the past accurately, and you cannot prescribe without a working prediction. And all three rest on the same foundation — a high-quality, integrated data warehouse. The clean separation also maps to the three generations of BI&A: 1.0 (internal structured data → traditional warehousing + OLAP), 2.0 (web/text data → text mining), 3.0 (mobile/IoT sensor data → real-time streaming).