AI-Powered Business Intelligence Fundamentals

Add AI to business intelligence: judge data readiness, model data for BI, build KPIs and forecasts, let people ask questions in plain language, and find the drivers and anomalies behind the numbers.

Intermediate · Data Science · about 17 hours · 5 modules · certificate on completion

Taught by Valentinà Lopèz, Thandiwè Dlaminì, AI Learning Companions.

About this course

Business intelligence tells an organisation what happened; AI can help it understand why, see what comes next and let anyone ask. This course shows how to add AI to BI without losing what makes BI trustworthy. You'll work through a year in the life of Northwind Living, a fictional homeware retailer whose West region has quietly fallen behind: computing KPIs and comparing them with targets, judging whether the data is ready (the 4Vs, data quality and a star schema that reconciles with the source), and learning the machine learning a BI analyst needs, from churn prediction judged by precision and recall to forecasts tested on months they never saw.

Then you'll build a small natural language query engine over a semantic layer, see how synonyms and naming make or break it, and put guardrails around SQL written by a large language model. Finally you'll diagnose the decline: decompose the shortfall, rank the key influencers of churn (and catch a confounder), segment customers with clustering and flag anomalous days. Labs use free tools, Python with DuckDB and a spreadsheet, on generated data that's identical for everyone; the capstone is a diagnostic sales dashboard with natural-language questions, in Power BI, Tableau, Looker Studio, a spreadsheet or Python.

What you'll learn

  • Apply the 4V framework and data-quality checks to judge whether data is ready for AI
  • Evaluate the feasibility and business value of an AI-enhanced BI project before building it
  • Model data as a star schema and define KPIs that everyone computes the same way
  • Implement natural language querying over a semantic layer, with guardrails for LLM-written SQL
  • Use diagnostic AI techniques to identify the drivers, segments and anomalies behind trends
  • Design dashboards that communicate AI-driven insights, forecasts and their uncertainty

Skills you'll gain

Business intelligence, KPI design, Data readiness (4Vs), Data quality, Star schemas, SQL with DuckDB, Forecasting, Natural language querying, Key influencer analysis, Anomaly detection

Tools

Python 3, DuckDB, pandas, scikit-learn, Excel, Google Sheets or LibreOffice Calc, Power BI Desktop or Tableau Public (optional)

Before you start

SQL for Data Analysis with DataGrip, or equivalent SQL (GROUP BY, joins, CTEs), and some Python with pandas (Data Analysis with Python and pandas). No machine learning needed. Install Python 3.10 or later, then pip install duckdb pandas scikit-learn, and have a spreadsheet (Excel, Google Sheets or LibreOffice Calc). The capstone uses any dashboard tool, such as Power BI Desktop, Tableau Public, Looker Studio or Python.

Modules

  1. Introduction to AI in BI

    Trace the move from reports to AI-powered business intelligence, place questions on the analytics spectrum from descriptive to prescriptive, and define KPIs that lead to decisions rather than just charts.

    • From reports to AI-powered BI (12 min)
    • The analytics spectrum: from descriptive to prescriptive (12 min)
    • KPIs and decision intelligence (12 min)
    • Hands-on lab: Compute KPIs and ask why with DuckDB

  2. Feasibility and Data Readiness

    Decide whether an AI-enhanced BI project is worth doing and whether the data can support it: judge business impact first, assess the 4Vs, profile and fix data quality, and model the data as a star schema.

    • Business impact before technical feasibility (12 min)
    • The 4Vs: is the data ready? (12 min)
    • Data quality checks and transformations (14 min)
    • Data modelling for BI: the star schema (12 min)
    • Hands-on lab: Audit a sales extract and build a star schema

  3. ML Fundamentals for BI Analysts

    Learn the machine learning a BI analyst needs: supervised and unsupervised learning for business questions, evaluating classifiers with accuracy, precision and recall, and forecasting honestly without overfitting history.

    • Supervised and unsupervised learning for BI questions (12 min)
    • Evaluating models: accuracy, precision and recall (12 min)
    • Forecasting without overfitting history (12 min)
    • Hands-on lab: Predict churn and forecast revenue

  4. Descriptive Analytics with NLP

    Let people ask questions of the data in everyday language: how natural language querying works, why it depends on a well-designed semantic model, and how to use SQL written by a large language model safely.

    • How natural language querying works (12 min)
    • Data modelling for NLQ: semantic layers, naming and synonyms (12 min)
    • LLM-powered NLQ and AI assistants (14 min)
    • Hands-on lab: Build a small NLQ engine with guardrails

  5. Diagnostic Analytics and Anomaly Detection

    Explain why metrics move: decompose a change into its drivers, rank key influencers while watching for confounders, segment customers with clustering, flag anomalies for review, and present the findings on a dashboard.

    • Key influencers: what drives a metric (12 min)
    • Segmentation with clustering (12 min)
    • Anomaly detection for metrics (12 min)
    • Dashboards that communicate AI-driven insights (12 min)
    • Hands-on lab: Diagnose the decline

Project: Diagnostic sales dashboard with NLP

Northwind Living's leadership want to understand 2025: which regions grew, which missed their targets, and why. Build a diagnostic sales dashboard from the course's data (python make_bi_data.py) in the tool of your choice: Power BI Desktop, Tableau Public, Looker Studio, a spreadsheet, or Python with a charting library. It should show the headline KPIs against target and last year, regional trends, the drivers of the Q4 shortfall, the key influencers of customer churn, flagged anomalies with their explanations, and a simple forecast with its uncertainty. Add natural-language questions: your tool's own Q&A feature, set up with synonyms, or the course's NLQ engine extended with at least five new synonyms and four new test questions. Finish with a one-page brief for leadership: the three most important findings, the evidence for each, a recommended action, and your data-readiness assessment with its caveats.

Part of

The products this course names are trademarks of their owners, who haven't made or endorsed it.

Start the course free