Operational vs BI&A Systems and Analytics

Differences between Operational and BI Systems

Operational Systems (OLTP) support daily transactions, focusing on efficient, reliable processing of business events [1, 2]. BI&A Systems support decision-making by analyzing historical, aggregated data to reveal insights and trends [2, 3].

Orientation

Operational: Transaction-oriented, optimized for capturing and processing individual events like sales or sign-ups [1, 2].
BI&A: Analytics-oriented, optimized for exploring and interpreting data to provide insights and reports [2, 4].

Level of Voluntariness

Operational: Use is obligatory for daily tasks (e.g., exam sign-up) [3, 6].
BI&A: Use is voluntary; managers choose to trust system insights or rely on intuition [6, 7].

Data Analysis Process & Exploratory Data Analysis

The Data Analysis Process involves understanding the problem, acquiring and preparing data, analyzing it, interpreting results, and acting on them [8–12]. It is iterative, not strictly linear [9, 13].

Exploratory Data Analysis (EDA)

EDA is examining basic statistics and visualizations to uncover patterns, anomalies, and initial hypotheses [13, 14].
Purpose: Identify data quality issues and guide preparation [13–15].
Importance: Crucial for correct interpretation and preparation, and for shaping further analysis and data collection [9, 13].

Operational vs Analytical Data Processing

• Data Nature: Operational uses current, dynamic data; analytical uses static, archived, derived, aggregated data [16, 18–20].
• Usage: Operational follows predefined processes (e.g., payments); analytical explores, drills down, slices and dices data [18, 21].
• Response Time: Operational demands sub-second responses; analytical tolerates seconds or minutes for complex queries [17, 22].

Real-time integration (e.g., HANA) blurs this divide but remains rare due to cost and complexity; “right time” is preferred over “real-time” [18, 23–29].

Business Performance Management & BI&A

BPM defines strategic goals, KPIs, and performance measures [28–30].
BI&A provides the technologies to collect, analyze, and present data to support BPM [33, 42, 43].

• BPM Defines “What”: Sets targets and metrics that BI&A must calculate and report [31, 34–36].
• BI&A Provides “How”: Consolidates data into a single source of truth, computes KPIs, and analyzes deviations [35, 37, 44–46].
• Mutual Dependence: BPM guides BI&A priorities; BI&A supplies the data that makes BPM effective [47].
• Coordination: Joint BPM and BI&A initiatives yield greater performance improvements than siloed efforts [35, 48, 49].

Types of Analytics

• Descriptive: Summarize what happened via reports and dashboards [50–52].
  Example: Total monthly sales, page views [52–54].
• Diagnostic: Explain why it happened using drill-down and correlation [53–56].
  Example: Why did sales drop 5% last month? [53, 56].
• Predictive: Forecast future outcomes probabilistically with statistics and ML [57–59].
  Example: Sentiment analysis [57–59].
• Prescriptive: Recommend actions and show likely outcomes, the most complex form [58, 60–62].
  Example: Pricing simulations, recommendation engines.

Analytical Technologies

They transform raw data into actionable insights [42, 63–65].

1. Human-Driven: Interactive, user-led analysis [66].
   Examples: Standard reports, ad-hoc queries, OLAP, associative analytics, in-memory, visualization, dashboards [62, 66–76].
2. Data-Driven: Algorithmic pattern discovery [76].
   Examples: Data mining, text mining, sentiment analysis [77–81].
3. Model-Driven: Mathematical/statistical simulations and optimization [81].
   Examples: What-if modeling, optimization, complex forecasting [82–84].

Systemic vs Ad-hoc Analytics Approaches

1. Data Preparation:
   • Systemic: Predefine and structure data in advance [85, 86].
   • Ad-hoc: Prepare data only when needed for a new problem.
2. User Involvement:
   • Self-Service: Users analyze data themselves [85, 87].
   • On-Demand: IT experts handle data and analysis [45, 86, 88].

In practice, a hybrid environment combines systemic data prep with self-service at the final analysis stage [89].

Self-Service BI: Benefits & Challenges

Benefits

• Increased productivity and faster decisions [84, 87–91].
• Flexibility to explore role-specific questions [87, 88, 91].
• Scalability and cost-efficiency for smaller teams [87, 88, 91, 92].
• Real-time what-if analyses foster an informed workforce [87, 88, 92, 93].

Challenges

• Data quality risks and incorrect analyses [93–95].
• Fragmentation and siloed “versions of truth” [94, 96].
• Lack of standardization across KPI methods [96].
• Information overload without proper training [96].
• Security and governance concerns [93, 96].
• Training needs remain significant [94, 97].

When to Use Self-Service

• Repeated reporting or analysis tasks [94].
• Prototyping systemic solutions (proof of concept) [94].
• Spreadsheets OK for final analysis, not for primary data storage [94].

Data Preparation & Integration

Real-world data is often dirty and complex. Preparation includes consolidation, cleansing, transformation, and reduction [14, 95–99].

• Consolidation: Merge sources to create one truth [100–102].
• Cle cleansing: Remove or correct errors [100, 103–105].
• Transformation: Restructure, derive attributes, change granularity [100, 106–111].
• Reduction: Drop or combine data for efficiency [102, 112, 113].

Levels of Integration Problems

1. Same IDs, Different Values: Consistent IDs but mismatched attributes; solve by trusting the most recent or a referential source [114–116].
2. Different IDs, Same Entity: Map disparate IDs using matching tables or unique attributes and confirm manually [114, 117–120].
3. Semantic Mismatch: Varying definitions of concepts (e.g., “customer”) across systems; requires organizational agreement on definitions before integration [119, 121–123].

Master Data Management Styles

Master Data (e.g., customers, products) is stable, contextual data [117, 123–126]. MDM ensures a single authoritative view across systems [124, 127–130].

1. Centralization: A dedicated MDM system creates and distributes the golden record centrally [132, 133].
2. Coexistence: Source systems master data locally, then sync with MDM for redistribution [134].
3. Registry: Systems store data independently; MDM provides links on demand [133, 134].
4. Consolidation: Independent systems; MDM merges data via ETL at consolidation time [133, 135].