Data Management for Business Intelligence
Application of computational techniques to support business activities, such as decision making, business understanding, data analysis, business process automation, learning from data, producing and using data-centric business models, ontology-based data access and integration, data quality assessment and cleaning and use of contextual data.
Study notes
Independent, AI-drafted study guide · Public sources linked below · Added November 5, 2026
Original guide aligned to fixed topics in Carleton's public 2026–27 undergraduate calendar. No particular term, instructor, or assessment coverage is claimed; examples and exercises are independently written.
Decisions, measures, and data contracts
Business intelligence connects data processing to a decision. Define the question, intended user, reporting period, and action before building a dashboard. A metric needs a population, numerator, denominator, unit, and time convention. Revenue booked, cash received, and profit are different measures even when a source system labels each as sales. Record the meaning and provenance of fields so that joining data does not silently combine incompatible concepts. A technically correct query can still answer the wrong business question.
References: PostgreSQL official tutorial — queries, joins, and aggregation
Integration, identity, and provenance
Data integration combines sources with differing schemas, identifiers, units, and update frequencies. Matching two records requires a defensible identity rule; identical names are not generally sufficient. Preserve source identifiers and transformation provenance so that a result can be traced back and corrected. Define precedence when sources disagree and distinguish a missing value from a recorded zero. Incremental processing should identify inserts, updates, and deletions consistently, including late arrivals and records that are corrected after the initial load.
References: PostgreSQL official tutorial — queries, joins, and aggregation
Quality assessment and cleaning
Assess data quality along dimensions such as validity, completeness, consistency, uniqueness, and timeliness. A syntactically valid date can still be wrong, so format checks do not establish truth. Profile distributions and relationships, then investigate anomalies against domain knowledge. Cleaning should be reproducible and auditable rather than an undocumented spreadsheet edit. Keep raw evidence or appropriate provenance and distinguish a corrected value from an imputation. Deleting difficult records can create a biased view of the population.
References: PostgreSQL official tutorial — queries, joins, and aggregation
Grain, aggregation, and a worked example
Define the grain of each table before aggregating. Suppose Orders contains one row per order and OrderItems contains three rows for one order totaling 90 dollars. Joining an order-level delivery fee of 5 to all items and summing the fee gives 15, which overcounts it. Aggregate items to order grain first or handle the order measure at its own grain. Average percentages also need denominator weighting: a 90% rate over ten cases and 50% over ninety cases combine to 54%, not 70%.
References: PostgreSQL official tutorial — queries, joins, and aggregation
Ontologies and contextual data
An ontology makes domain concepts and relationships explicit so that data can be understood across sources. RDF represents statements as subject-predicate-object triples, while identifiers connect references to the same resource. A shared vocabulary can support integration without forcing every source into the same physical table layout. Context matters: a sensor value needs units, location, time, and measurement conditions. An inference derived under one context should not automatically be treated as a universally observed fact.
References: W3C RDF 1.1 Concepts and Abstract Syntax
Models, automation, and responsible reporting
A predictive model used in business analysis needs evaluation that respects time, group structure, and what information was available when the decision was made. Correlation can guide investigation without proving causality. Business-process automation requires exception handling and an auditable record of actions, especially when a data error could trigger an incorrect decision. Present uncertainty, freshness, and relevant definitions beside results. Access controls should follow data sensitivity and purpose, including exported reports and cached dashboard results.
References: PostgreSQL official tutorial — queries, joins, and aggregation
Practice & worked answers
Original exercises for this guide.
1. Two branches report conversion rates of 80% from 10 visitors and 20% from 90 visitors. What is the combined rate?
Show worked answer
The branches contribute 8 and 18 conversions, for 26 conversions from 100 visitors: 26%. Averaging the two percentages as 50% would incorrectly give the small branch the same weight as the large one.
2. Why preserve source IDs after creating a unified customer ID?
Show worked answer
They support provenance, debugging, correction, and reconciliation when the identity match changes. Without them, it becomes difficult to explain which original records contributed to a derived customer profile.
Public notes & references
Course sources establish the scope above. Reference materials support the explanations; linked lecture notes and books remain with their authors.
2018
Topics
- Data access through ontologies.
- analytical processing, elements of data min- cleaning.
- management, data quality assessment, data
- cleaning, elements of business process model- Reading Material (mandatory)
- Lecture notes posted after every lecture
- niques rather than specific applications or
- systems/implementations. 2. Relevant survey and research papers
- will be posted for reading.
- Pregnancy obligation
- Religious obligation
Prerequisites & course information
Prerequisite(s): COMP 3005 . Also offered at the graduate level, with different requirements, as COMP 5111 , for which additional credit is precluded. Lectures three hours a week.