Week 6 | Session 2: Types of Analytics & SCM Applications

Course: Supply Chain Digitization — Module 3: Analytics in SCM

Session Agenda

Four Types of Analytics — Overview

#	Type	Core Question	Time Orientation	Typical Methods / Tools
1	Descriptive	What has happened?	Past data only	Bar charts, dashboards (Power BI, Tableau), mean, std. deviation
2	Diagnostic	Why did it happen?	Past data — root cause investigation	Root cause analysis, drill-down analysis, correlation analysis
3	Predictive	What will happen?	Past data → model → future forecast	Statistical forecasting, ML models (regression, neural nets)
4	Prescriptive	What should be done?	Future-focused — action recommendation	Optimisation models (LP, MIP), simulation, decision support

Key mental model: Descriptive → Diagnostic → Predictive → Prescriptive = a chain. Each level builds on the previous one. Prescriptive = Optimisation: Minimise cost | Maximise profit | Minimise risk — subject to real-world constraints

1. Descriptive Analytics — “What Has Happened?”

Definition: Analyses past data to summarise what occurred — average, median, standard deviation, trends, patterns
Output: Charts, dashboards, summary statistics
Tools: Power BI | Tableau | Excel dashboards | Bar charts | Pie charts

SCM Examples

Demand analysis: What was avg. demand per month for each SKU over the past 2–3 years?
FMCG CEO dashboard: How many units in each DC? Which warehouse has dead stock?
Quality monitoring: Which products (X, Y, Z) have quality failures? Over what time period?
Worker performance: Which pickers/packers in an e-commerce sorting facility met their timeline targets?

2. Diagnostic Analytics — “Why Did It Happen?”

Definition: Digs into past data to identify root causes of events — why performance deviated from expectations
Output: Root cause identified | Corrective action triggered
Analogy: Microscope — zoom into the data to find the underlying problem

SCM Examples

Product demand down: Descriptive shows Product A demand fell. Diagnostic asks: Why? → Competitor entered market → corrective action: reposition, price adjustment
Product demand up: Descriptive shows Product B demand rose. Diagnostic identifies WHY it rose → replicate the cause to sustain growth
Quality failure: Product Z has quality issues. Diagnostic traces root cause: Component C1 is defective → Component C1 supplied by Supplier 1 → Supplier 1 notified

3. Predictive Analytics — “What Will Happen?”

Definition: Uses past data to build models that forecast future outcomes — demand, machine failure, delivery ETA, worker performance
Output: Forecast values with associated probability or confidence range
Tools: Statistical models (ARIMA, exponential smoothing) | ML models (regression, neural networks, time series)

SCM Examples

Demand forecasting: Predict demand for each SKU for next 15 days, 2 months, 6 months. ML is needed for high variability driven by social media.
Predictive maintenance: Model predicts: “This machine will break down in 2–3 days.” → Maintenance scheduled proactively → avoids production stoppage
Worker efficiency prediction: Use past performance data to predict each worker’s efficiency score for the next 6 months
Overseas shipment ETA: Predict actual arrival date of a shipment (e.g., 5-day delay or 4-day early) → plan production schedule accordingly

4. Prescriptive Analytics — “What Should Be Done?”

Definition: Uses predictions and optimisation to recommend the best decision or action — minimise cost, maximise profit, minimise risk, subject to operational constraints
Core technique: Optimisation (Linear Programming, Mixed Integer Programming, Simulation)
Output: A recommended action or decision — not just a forecast

SCM Examples

Capacity expansion: Predictive says demand will exceed capacity. Prescriptive asks: Should I build a new plant or outsource? If new plant, WHERE? What CAPACITY?
Supplier selection: Supplier 1 disqualified. Prescriptive optimises selection based on cost, lead time, responsiveness, and quality → decides which supplier to choose AND how much to order.
Worker allocation: Use predicted efficiency scores to assign workers to jobs during Diwali surge: Best performers → fastest, error-sensitive roles.
Slotting optimisation: In a warehouse, which product goes in which slot? Fast-moving near dispatch, items ordered together in same zone (clustering).

Three End-to-End Worked Examples (All 4 Analytics Types)

Example 1 — Demand & Capacity Planning

Step	What Is Done
Descriptive	Bar/pie chart of past demand per product (A and B). Product B demand is clearly rising; Product A is falling.
Diagnostic	Product A fell → competitor entered market. Product B rose → new use case discovered, positive social media buzz.
Predictive	Forecast demand for Product B for next 2–3 years. Demand will exceed current plant capacity within 18 months.
Prescriptive	Decide: new plant vs. outsource. If new plant: optimise location + capacity. If outsource: select 3PL manufacturer.

Example 2 — Quality Failure & Supplier Management

Step	What Is Done
Descriptive	Track quality failures across Products X, Y, Z. Product Z has repeated quality failures for last 4–5 months.
Diagnostic	Root cause analysis on failed Product Z units → defective Component C1 → C1 was supplied by Supplier 1 → Supplier 1 notified.
Predictive	Predict likelihood of continued quality failure if Supplier 1 remains. Also predict alternative suppliers’ future quality, cost.
Prescriptive	Optimise supplier selection from Suppliers 2–5. Criteria: cost + delivery time + responsiveness + quality record.

Example 3 — Worker Allocation in E-Commerce Sorting Facility

Step	What Is Done
Descriptive	Analyse past performance of each picker/packer — who met timelines, who did not, who had the fewest sorting errors.
Diagnostic	Why did certain workers underperform? Fatigue? Unclear process? Wrong zone assignment?
Predictive	Predict each worker’s efficiency score for next 6 months. Rank workers into high / medium / low efficiency bands.
Prescriptive	Allocate workers to tasks based on predicted efficiency. During festival peak: assign highest-rated workers to fastest-moving jobs.

Analytics Across the SCM Value Chain — Master Reference

SCM Domain	Descriptive (What happened?)	Diagnostic (Why?)	Predictive (What will happen?)	Prescriptive (What to do?)
Procurement	Analyse past supplier cost, quality, delivery time	Why did supplier X underperform last quarter?	Predict future supplier cost, lead time, reliability	Select optimal supplier(s) + decide order quantity from each
Manufacturing & Quality	Identify which product/component has quality failures	Root cause: Which component C1 is defective? Which supplier?	Predict machine failure before it occurs (predictive maintenance)	Select replacement supplier. Optimise maintenance schedule.
Warehousing	Inventory count report — compare ERP vs. physical.	Why are certain SKUs always showing discrepancy?	Predict worker efficiency scores. Predict fast-moving items.	Slotting optimisation. Zoning. Work allocation to pickers.
Logistics & Transport	GPS track delivery history. Driver eyeball data.	Why did truck delay on Route X last week?	Predict ETA of overseas shipments. Predict driver drowsiness.	Route optimisation. Fleet assignment. Driver scheduling.
Demand Planning	Past demand by SKU, zone, month. Sales trends.	Why did demand for Product A drop? Competitor entry?	Forecast demand for next 15 days, 2 months, 6 months per SKU.	Capacity expansion planning. Plant location. Inventory pre-positioning.

Domain Highlights — Key SCM Analytics Applications

Procurement

Supplier selection: Multiple suppliers available → prescriptive optimisation selects best based on cost, lead time, quality, responsiveness
Order quantity decision: Once supplier is selected, how much to order from each?

Manufacturing — Predictive Maintenance

Definition: Model analyses machine sensor data → predicts failure 2–3 days before it occurs → maintenance team scheduled proactively
Data used: Vibration, temperature, operating hours, error codes — sensor-based real-time data

Warehousing — Slotting Optimisation & Zoning

Slotting: Decide which product occupies which slot in the warehouse → fast-moving near dispatch
Zoning: Clustering algorithm groups products frequently ordered together → kept in the same zone → lower picking time + fewer errors

Logistics — Overseas Shipment Delay Prediction

Analytics fix: ML model predicts actual arrival date → production schedule adjusted in advance → avoids holding costs (early arrival) or production stoppage (late arrival)

Demand Planning — The Hardest Forecasting Problem

Why difficult: Consumer behaviour is driven by social media, influencers, and fast-changing preferences.
Demand sensing: Real-time tracking of demand signals (social media, search trends) to update forecast continuously
ML models: Better than traditional models for high-variability demand because they capture non-linear patterns and external signals.

Session Summary

4 analytics types: Descriptive (past) → Diagnostic (why) → Predictive (future) → Prescriptive (action)
Prescriptive = optimisation: Minimise cost / maximise profit / minimise risk, subject to constraints
3 full worked examples: (1) Demand + capacity planning | (2) Quality failure + supplier selection | (3) Worker allocation in sorting facility
Procurement analytics: Supplier selection + order quantity optimisation + predictive cost/lead time
Manufacturing: Image analytics for quality | Predictive maintenance → avoid unplanned downtime
Warehousing: Slotting optimisation | Zoning (clustering) | Video analytics drone inventory count
Logistics: Overseas shipment ETA prediction | Driver drowsiness detection (image analytics)
Demand planning: ML > traditional models for high-variability demand | Demand sensing | Multiple forecast horizons
Next: AI/ML for forecasting + Python hands-on