Skip to main content

Descriptive, Predictive & Prescriptive Analytics

Module: Module 1 — Management FoundationsCode: UDPPA (PRN)Faculty: Prof. Pritam RanjanSessions: 2Status: ✅ Drafted

Big idea​

Analytics is not a tool — it is a disciplined, cyclical process of turning a real-world question into a defensible answer. Prof. Pritam Ranjan frames it as six pillars: business understanding, data requirements, data collection, data preparation, modelling & evaluation, and deployment. As a senior manager you will rarely write the code, but you must know the process well enough to ask the right questions, recognise when your team is rushing past data quality, and pick between models on accuracy and reliability rather than novelty. Equally important is knowing your data structure (scalar, vector, time-series, functional, image/video, text/audio) because each demands a different family of models — there is no universal "best" model.

Key concepts​

  • The six-pillar analytics lifecycle. Business understanding → data requirements → data collection → data preparation → modelling & evaluation → deployment. It is cyclical, not linear — most projects loop back to data collection at least once before they ship.
  • Four levels of analytics. Descriptive (what happened — dashboards), diagnostic (why — drill-down, correlation), predictive (what will happen — regression, ML), prescriptive (what should we do — optimisation). Each level adds value and complexity in equal measure.
  • Data structures. Scalar, vector, time-series, functional, image, video, text, audio — each unlocks a different model family. Naming the structure correctly is half of model selection.
  • Data types. Continuous, discrete, binary, categorical, ordinal, count — mislabelling them (e.g. treating an ordinal scale as continuous) produces statistically meaningless results that look credible.
  • Accuracy vs reliability. Accuracy is fit on data you have; reliability is performance on data you have not seen. Always split train/validation; a slightly less accurate model that generalises beats a perfect-fit model that overfits.
  • Open-source toolchain. R and Python are the practical defaults — free, huge community, production-grade. SPSS/SAS/JMP remain useful in regulated industries where the audit trail matters more than flexibility.

Self-check​

A team has built a regression model with R² = 0.95 on the training set but R² = 0.42 on a held-out validation set. What is the most likely diagnosis?

  • A. The model is highly accurate and reliable — ship it
  • B. The model has overfit the training data; reliability is poor
  • C. The validation data is wrong; retrain on the combined set
  • D. The data type was misclassified as ordinal
Name the six pillars of the analytics process
1) Business understanding, 2) Data requirements, 3) Data collection, 4) Data preparation/cleaning, 5) Modelling & evaluation, 6) Deployment. Cyclical, not strictly sequential.

Click the card to flip

1 of 4

Continue learning​

📘Book · FreeAn Introduction to Statistical LearningJames, Witten, Hastie & Tibshirani — the modern reference on predictive modelling. Free PDF + R/Python labs.📄HBR · ReadCompeting on AnalyticsTom Davenport's seminal article on how data-driven companies build durable competitive advantage.🎓MIT · CourseThe Analytics EdgeFree MIT OpenCourseWare class teaching analytics through real cases — healthcare, sports, web search, finance.🧰Tool · PractiseKaggle LearnFree hands-on micro-courses in Python, pandas, ML, SQL and data viz — short, project-based, no install needed.
🪞 Apply it — reflection prompts
  1. Pick one recurring decision in your team. Which of the four analytics levels is it at today, and what would the next level look like?
  2. Audit one dashboard you receive weekly. What action does it actually change? If none, what should be removed or replaced?
  3. For your next data project, write down the question, the acceptable error, and the action you will take for each possible answer — before the data is collected.

📝 Going deeper. Davenport & Harris, Competing on Analytics (HBR, 2006) is the executive primer. For hands-on depth, James/Witten/Hastie/Tibshirani, An Introduction to Statistical Learning (free PDF) covers the modelling half of the lifecycle without heavy maths. For data-quality discipline, see the Google Cloud Data Quality Framework.