Bài 1: Business Analytics Foundations & Types of Analytics
Series: Business Analytics Cheatsheet Series
Topics:
business-analyticsdescriptivediagnosticpredictiveprescriptiveSource: Chapter 1 — Introduction to Business Analytics. Conceptual grounding: what BA is, how it differs from adjacent fields, how it evolved, and the four analytics types. Read before file 02.
1. What Business Analytics is
| Term | Type | Definition / Usage |
|---|---|---|
| Business Analytics (BA) | Discipline | The process of understanding a business’s data-driven activities to draw inferences and make decisions with higher certainty. Operationally: exploring, experimenting, simulating, and summarizing data to extract information that managers act on. |
| Iterative investigation | Principle | BA is not a one-shot report. It is repeated exploration of organizational data with emphasis on statistical analysis, feeding decisions and then re-measuring outcomes. |
| Fact-based decision-making | Principle | BA extends accountability into decisions: the reasoning is written down and can be audited. This is the real organizational value, beyond any single insight. |
| Constituent disciplines | Scope | Statistics, computer science, business domain knowledge. Methods: data mining, predictive modeling, optimization, visualization, ML/AI, big data frameworks. |
| Key words in every BA definition | Mnemonic | Data · Business/Organization · Analytics tool · Process · Data-driven insights · Fact-based decision-making · Leader/Manager · DS/ML/AI. If a proposal is missing one of these, the framing is incomplete. |
Goals of BA
| Goal | Type | Definition / Usage |
|---|---|---|
| Real-time actionable information | Goal | Deliver information at the speed of the decision, not the speed of the warehouse refresh. |
| Tools at all organizational levels | Goal | Decision support around customer goals and profitability, with performance comparison. Analytics restricted to a central team does not change behavior. |
| Objective forecasting | Goal | Increase the accuracy and objectivity of forward-looking statements. |
| Feedback loop → learning organization | Goal | Insight must return as measured outcomes; without feedback, BA degenerates into reporting. |
Unique characteristics of BA
| Characteristic | Type | Definition / Usage |
|---|---|---|
| Purposive | Quality criterion | Analysis must align with a business function (finance, marketing, sales) and a management objective (performance, growth, compliance, risk, profitability). Analysis without a purpose is cost. |
| Intuitive | Quality criterion | Must uncover previously hidden patterns. Analytics that merely confirms the status quo or restates conventional wisdom offers no insight — this is a usable review standard. |
| Expedient | Quality criterion | The output must be actionable: a manager can act on the recommendation with the resources they have. |
2. BA vs Data Analytics vs Data Science
| Term | Type | Definition / Usage |
|---|---|---|
| Two axes of the landscape | Framework | Axis 1: past vs future orientation. Axis 2: experience-driven vs data-driven. All four quadrants are legitimate business activities; only some are data-driven. |
| Business case studies | Activity | Real-world accounts of company success/failure. Past-oriented, experience-driven. Value: pattern recognition and mistake avoidance; no dataset required. |
| Qualitative analytics | Activity | Using intuition and domain knowledge for future planning. Future-oriented, experience-driven. |
| Reporting / dashboards | Activity | Reflection of past data. Past-oriented, data-driven. Sits on the left of the timeline axis. |
| Forecasting | Activity | Future-oriented and data-driven, but still inside the business sphere — the intersection of business analytics and data. |
| Business Analytics | Field | Business-context-bound; optimizes decisions inside a known organizational objective. |
| Data Analytics | Field | Broader analysis of data of any kind; may or may not sit inside a business objective. |
| Data Science | Field | Adds ML/AI methods and engineering scale; concerned with building predictive systems, not only informing a single decision. |
Practical takeaway: the fields overlap on methods and differ on what constrains the question. In BA the constraint is a business objective; in DS it is often the data or the model.
3. Evolution of BA
| Era | Type | Definition / Usage |
|---|---|---|
| Foundations (pre-1950s–1950s) | History | Statistics formalized (1749) supplies collection/analysis/forecasting methods. Industrial Revolution applies statistics to standardization and cost optimization. Operations research emerges 1937 (military), adapts to business in the late 1940s as management science. Time-and-motion studies in the 1950s mark structured data use. |
| BI & Decision Support (1960s–1970s) | History | Centralized inventory control and automatic data processing (1960s); rule-based enterprise systems and materials planning (1970s). Hans Peter Luhn introduces “Business Intelligence” at IBM in 1958. DSS emerge in the late 1960s, combining BI with OR/MS for unstructured managerial problems. |
| Integration & Data Warehousing (1980s–1990s) | History | ERP systems (1980s) for collection/storage → enterprise data warehouses for unified real-time analysis. EIS (1990s) add graphical decision support. IS umbrella covers TPS, MIS, DSS, EIS. Focus: descriptive analytics; predictive techniques begin entering. |
| Big Data & Predictive (2000s–2010s) | History | Transition from BI (historical) to BA (predictive/prescriptive). Web 2.0 and social media drive volume/velocity/variety → NoSQL and cloud (AWS). Netflix-style ML recommendations; NLG-automated reporting. Labelled “Analytics 2.0”. |
| AI Integration & Prescriptive (2020s–) | History | “Analytics 3.0/4.0”: AI/ML as a core pillar alongside visualization and statistical modeling. Sentiment analytics, autonomous AI, self-service tools (Tableau, Power BI, SAP Analytics Cloud). Symbiotic AI↔BA relationship. Rising constraints: GDPR-style privacy, blockchain-mediated data sharing, scalable infrastructure demand. |
Why this matters practically: each era added a layer without removing the previous one. A modern BA function still runs transaction processing, warehousing, descriptive reporting, and predictive modeling simultaneously — most organizational friction comes from these layers having inconsistent definitions.
4. The 6-step BA process
Compare against CRISP-DM (file 02) — same spine, lighter formalism.
| Step | Type | Definition / Usage |
|---|---|---|
| 1. Identifying the problem | Process step | Crisis, unmet business need, or process optimization. Clarify the expected outcome, then decompose into smaller goals. Stakeholders decide the required data. Key questions: what data is available? is it sufficient? |
| 2. Exploring data | Process step | Gather and — more importantly — cleanse. Handle missing data, remove outliers, derive new variables from combinations of existing ones. Cleaning is also how you build a feel for the dataset. |
| 3. Analysis | Process step | Apply statistical methods (hypothesis testing, correlation). Pivot the analysis around the target variable and the factors affecting it. Slice, compare, state assumptions. Output: actionable insight. |
| 4. Prediction and optimization | Process step | Model with prediction techniques (neural nets, decision trees). Run multiple models in parallel, select on accuracy, answer “what if…?” via parameter conditions. |
| 5. Decide and evaluate outcome | Process step | Turn model insight into a viable action plan, execute, wait, then measure the realized outcome. The waiting period is part of the method, not a delay. |
| 6. Optimizing and updating | Process step | Refine if the plan can be improved; otherwise register outcomes into an ever-improving database. Evaluate ROI here — this step is what makes future analytics cheaper. |
Mapping to CRISP-DM: 1 → Business Understanding · 2 → Data Understanding + Preparation · 3 → Modeling (diagnostic) · 4 → Modeling (predictive/prescriptive) · 5 → Evaluation · 6 → Deployment + Monitoring.
5. Types of Analytics
The four-layer ladder. Each layer answers a different question and requires different evidence.
| Type | Question | Definition / Usage |
|---|---|---|
| Descriptive | What happened? | Aggregate, organize, and present historical performance so trends and patterns are visible. Output: reports, dashboards, KPIs. |
| Diagnostic | Why did it happen? | Identify reasons behind past outcomes via patterns and correlations. Goes beyond reporting the numbers into underlying causes and relationships. |
| Predictive | What is likely to happen? | Use historical data, statistics, and ML to forecast the next state. |
| Prescriptive | What should we do? | Recommend actionable strategies to optimize the decision. |
Descriptive Analytics — detail
| Item | Type | Definition / Usage |
|---|---|---|
| Applications | Usage | Reports, dashboards, KPIs summarizing sales trends and customer behavior. Domains: finance, inventory management, marketing, HR, sales. |
| Benefits | Value | Clear business visibility · smarter decision-making · early issue detection · better communication and alignment · foundation for deeper analysis. |
| Process | Process | 1) Define key questions. 2) Collect relevant data. 3) Clean and prepare. 4) Aggregate and analyze. 5) Visualize and share. 6) Monitor and iterate. |
| Challenges | Anti-pattern | Data silos · data quality issues · over-reliance on averages · static reports · lack of context. Averages hide the distribution — always pair a mean with a spread and a segment cut. |
Diagnostic Analytics — detail
| Item | Type | Definition / Usage |
|---|---|---|
| Applications | Usage | Root cause analysis, data mining for performance drivers (e.g. why sales dropped in a specific region). |
| Techniques | Method set | Hypothesis testing · anomaly detection · root cause analysis · correlation analysis · diagnostic regression analysis. |
| Process | Process | 1) Define problem. 2) Collect data. 3) Process data. 4) Choose the analytic technique. 5) Apply statistical methods. 6) Visualize. 7) Interpret. |
| Benefits | Value | Clearer visibility · smarter decisions · process improvements · better customer experience · risk reduction. |
| Challenges | Anti-pattern | Data quality · silos · confusing correlation with causation · missing the bigger picture · skills gaps · information overload. |
Predictive Analytics — detail
| Item | Type | Definition / Usage |
|---|---|---|
| Definition | Method | Historical data + statistical techniques + ML algorithms to forecast what is likely to happen next. |
| Applications | Usage | Demand forecasting · experience personalization · customer churn prediction · risk assessment. |
| Prerequisite | Constraint | Requires a stable, well-defined target variable and leakage-safe features (file 03). |
Prescriptive Analytics — detail
| Item | Type | Definition / Usage |
|---|---|---|
| Definition | Method | Recommend actionable strategies from data insight to optimize decision-making. Usually optimization or decision-policy on top of a predictive model. |
| Applications | Usage | Supply chain optimization · pricing strategy · resource allocation · finance. |
| Benefits | Value | Real-time actionable insight · simplifies complex data · full-picture view · faster decisions · reduces bias and guesswork · aligns decisions with business goals. |
Choosing the right layer
| Stakeholder question sounds like… | Layer | Typical method set |
|---|---|---|
| “How many / how much / what’s the trend?” | Descriptive | Aggregation, cohorting, dashboards |
| “Why did X change?” | Diagnostic | Segmentation, hypothesis tests, regression, anomaly detection |
| “Will this customer churn / how much will we sell?” | Predictive | Classification, regression, forecasting |
| “What should we do about it?” | Prescriptive | Optimization, uplift modeling, decision policy + experiment |
Misdiagnosis is the most expensive error here: answering a diagnostic question with a predictive model produces a technically valid artifact that no one can act on.
Definition of Done
- The request is classified into one of the four analytics layers before any method is chosen.
- The analysis passes all three BA quality criteria: purposive, intuitive, expedient.
- Averages are never reported without a spread and at least one segment cut.
- Correlation and causation are labelled distinctly in every diagnostic readout.
- Step 6 (ROI evaluation and outcome registration) is scheduled, not optional.
Next: CRISP-DM, KPI Tree & Problem Framing for the formal project process.
🔗 Cùng series
Mở đầu: Business Analytics Cheatsheet Series
Bài 2: CRISP-DM, KPI Tree & Problem Framing
Nếu bài này hữu ích, hãy chia sẻ cho người đang học Data Science / Business Analytics. Mọi góp ý về lỗi kỹ thuật rất được hoan nghênh.