Role of data in decision making
1. Definition
Data-driven decision making (DDDM) means choosing between alternatives on the basis of evidence — measurements, analysis and models — rather than intuition, seniority or habit alone. Data does not make the decision; it reduces uncertainty about which option is likely to work, and it makes the outcome measurable afterwards. Intuition is not discarded — experience suggests which question to ask and how to interpret what the numbers show — but the argument must be settled by evidence.
Data improves a decision in four ways: it reveals what is actually happening, explains why, predicts what is likely to happen next, and prescribes what to do about it.
2. From Data to Wisdom (the DIKW chain)
DATA → INFORMATION → KNOWLEDGE → WISDOM / ACTION
raw facts processed, patterns, meaning, judgement, decision
(₹450, 11:40, summarised: causal understanding: "shift the 11:30–1:00
counter 2) "peak hour sales "weekday lunch rush staff roster to lunch
fall from 1 pm" drives 40% of sales" hours; add a combo offer"
A dashboard of numbers is only data; a decision that changes staffing is wisdom. Most organisations get stuck at the information stage — plenty of reports, few decisions.
3. The Four Types of Analytics (favourite exam table)
| Type | Question | Technique | Example |
|---|---|---|---|
| Descriptive | What happened? | Aggregation, KPIs, dashboards, reports | Last quarter's sales by region; app's daily active users; attendance percentage |
| Diagnostic | Why did it happen? | Drill-down, correlation, root-cause and segment analysis | Why did sales drop in one store — stock-out? competitor? festival shift? |
| Predictive | What will happen? | Regression, time series, classification, ML models | Forecast next month's demand, predict which students will fail a subject, churn risk |
| Prescriptive | What should we do? | Optimisation, simulation, A/B testing, reinforcement learning, recommendation | Best price and stock allocation, which topic to practise next, optimal bus routes |
The value rises as you move down the list, and so does the difficulty — descriptive analytics needs clean data, prescriptive analytics needs trust in the model and a way to act on it.
4. Where Data Changes Decisions
| Domain | Data used | Decision improved |
|---|---|---|
| Business & retail | Sales, footfall, clickstream, inventory | Stock levels, pricing, promotions, store layout, staffing |
| Banking & finance | Transactions, credit history, repayment behaviour | Loan approval and limits, fraud blocks, risk pricing, collections |
| Healthcare | Symptoms, lab results, images, history, drug data | Diagnosis, treatment choice, triage priority, bed and OT scheduling |
| Education | Attendance, assessment marks, topic-level attempts, engagement | Remedial classes, study plans, question recommendation, resource allocation |
| Agriculture | Soil, weather, satellite, yield history | Sowing window, irrigation and fertiliser quantity, crop choice, insurance claims |
| Government | Census, tax, health, traffic, grievance data | Scheme targeting, budget allocation, traffic signals, epidemic response |
| Manufacturing | Machine sensors, defects, downtime logs | Maintenance scheduling, quality inspection thresholds, production planning |
| Transport & logistics | GPS, travel times, fuel, load data | Routes, fleet utilisation, delivery ETAs, driver shifts |
Everyday Indian examples: UPI and card fraud alerts that block a suspicious transaction; credit scoring that extends small loans using transaction history instead of collateral; traffic-signal timing and FASTag data used for congestion management; Kisan advisories built from weather and soil data; learning apps that pick the next practice question from past performance and inform school interventions for weak topics.
5. Benefits of Deciding with Data
| Benefit | Why it matters |
|---|---|
| Objectivity | Decisions are argued with evidence, not hierarchy or ego |
| Speed and consistency | Model- or rule-driven decisions are instant and uniform across branches/agents |
| Accuracy and personalisation | Better targeting, fewer losses, offers relevant to the individual |
| Measurability and learning | Experiments (A/B tests) tell you what worked, so the next decision is better |
| Accountability and auditability | The basis of a decision is recorded and can be reviewed |
| Risk management | Fraud, credit, safety and health risks are caught earlier |
| Resource optimisation | Money, staff, water, fuel and power go where they matter most |
| Competitive advantage | Faster feedback loops compound over time |
6. Challenges, Risks and How to Mitigate Them
| Risk | Mitigation |
|---|---|
| Bad data quality (GIGO — garbage in, garbage out) | Validation at source, data governance, quality KPIs, audits |
| Bias (unrepresentative or historically discriminatory data) | Diverse data, fairness testing, human review, documented limitations |
| Correlation mistaken for causation | Controlled experiments, A/B tests, causal inference, domain review |
| Analysis paralysis / dashboards nobody acts on | Tie metrics to decisions and owners; keep KPIs few and meaningful |
| Privacy and legal exposure | Consent, minimisation, purpose limitation, DPDP Act/GDPR compliance, encryption |
| Over-reliance on the model ("the algorithm says so") | Human-in-the-loop for high-stakes calls, explainability, confidence reporting |
| Data silos and inconsistent definitions | Single source of truth, common metric definitions, shared warehouse |
| Cost, skills and change resistance | Start with high-value, low-complexity use cases; train staff; show early wins |
| Security | Access control, encryption, monitoring, least privilege on the data platform |
7. A Practical Decision Process with Data
- Define the decision and who owns it (not "explore data", but "how many buses on route 12 next term?").
- Choose the metric/KPI that will judge success, plus guardrail metrics that must not suffer.
- Collect and validate the needed data — and be honest about what is missing.
- Analyse descriptively and diagnostically first; model only if prediction is genuinely needed.
- Generate options and evaluate them with the numbers (simulation, pilot, A/B test, cost-benefit).
- Decide, record the rationale, implement — with a named owner and a date.
- Monitor the outcome against the KPI, and feed the result back into the next cycle (PDCA/OODA loop).
Two habits separate organisations that benefit from data from those that merely collect it: they experiment instead of arguing, and they close the loop by measuring whether the decision actually worked.
8. Exam Points
- Explain the role of data in decision making with examples — 5-7 marks, almost certain.
- Differentiate descriptive, diagnostic, predictive and prescriptive analytics with one example each — 5-7 marks, very frequent.
- Explain the DIKW chain (data → information → knowledge → wisdom) — 5 marks.
- How does data help in business/healthcare/education decisions? — 5 marks.
- List any four benefits and four challenges of data-driven decision making — 5 marks.
- What is data quality's effect on decisions? What is GIGO? — 3-5 marks.
- Why is correlation not causation? How do A/B tests help? — 3-5 marks.
- Outline the steps of a data-driven decision process — 5 marks.