Day 1 | Session 1: Power BI Foundations & AI Prompting
Participants define the business questions, become familiar with Power BI, and import the course dataset. They learn to give AI enough context to produce useful and testable suggestions.
- 20 minutes: Course introduction and business case. Define the audience, decisions, and success measures for a sales performance report.
- 25 minutes: Power BI workflow and interface. Explore Desktop, Power Query, the data model, report pages, and the Power BI service.
- 35 minutes: AI tools and effective prompts. Distinguish Claude, Copilot in Power BI, and MCP. Practice giving context, a task, constraints, and an expected output.
- 40 minutes: Connect to Excel or CSV files. Inspect tables, column types, missing values, and the level of detail in each row.
- 40 minutes: Guided exercise. Load sales and reference tables, create a first visual, and use Claude to propose useful business questions and KPIs.
- 20 minutes: Review and knowledge check. Compare suggestions with the available fields, resolve common import issues, and save the working report.
Session 1: Topics, Practice & Outputs
- Reports, dashboard-style report pages, and service dashboards; the difference between the terms.
- Business questions, KPI definitions, and data grain such as one row per sales line.
- Prompting with a table schema and approved sample data; checking assumptions and avoiding invented columns.
- The distinction between an AI chat suggestion and an AI assistant connected to a model through tools.
- Practical exercise: Using the supplied sales case, identify 5 management questions, map each question to available fields, and import the required tables. Create one simple sales visual and record any data quality issues for the next session.
- Example AI prompt: “Using this sales table schema, propose 5 questions for a sales manager and the measures needed to answer them. Use only the listed fields. Ask me to clarify missing definitions before suggesting formulas.”
- Session output: A saved Power BI file containing the imported tables and first visual, a list of business questions, and an initial KPI definition sheet.
Day 1 | Session 2: Data Preparation & Modeling
Participants prepare the data for analysis and build a simple star schema. AI supports explanations and troubleshooting while participants check every transformation and relationship.
- 25 minutes: Review data quality. Identify blanks, duplicates, inconsistent labels, invalid dates, and incorrect data types.
- 45 minutes: Clean data with Power Query. Change types, trim text, handle errors and nulls, and remove duplicate records where the business rules justify it.
- 45 minutes: Build the model. Separate sales facts from product and region reference tables. Set one-to-many relationships and explain filter direction.
- 30 minutes: Create a date table and connect it to sales. Review date coverage, unique keys, and the effect of filters on totals.
- 25 minutes: AI-supported troubleshooting exercise. Ask Claude to explain an M step or diagnose a relationship issue, then apply and test the correction.
- 10 minutes: Validation checkpoint. Reconcile row counts and sales totals before and after transformation.
Session 2: Topics, Practice & Outputs
- Repeatable transformation steps and refresh behavior.
- Power Query M as the language behind transformations; introductory explanation rather than advanced coding.
- Fact tables, dimension tables, unique lookup keys, and unmatched records.
- A date dimension and basic model naming and descriptions.
- Why a clean model improves both reports and AI responses.
- Practical exercise: Clean the sales data, standardize product and region labels, and build the model. Investigate a deliberately introduced data issue. Use Claude to propose a correction from the supplied schema or formula, then confirm that the change preserves the intended records and totals.
- Example AI prompt: “Explain this Power Query step in plain English. Identify how it handles blank and invalid dates, suggest a correction if needed, and give me 3 checks to confirm that no valid sales rows are lost.”
- Session output: A clean model with connected tables, a working date table, and a short validation record showing row counts, totals, and any exclusions.
Day 2 | Session 3: DAX, Reports & Copilot
Participants create business calculations, design useful visuals, and explore how Copilot can support report authoring and interpretation. Copilot activities use the prepared training environment.
- 25 minutes: DAX foundations. Distinguish measures from calculated columns and explain filter context using a region or product slicer.
- 45 minutes: Create measures with AI support. Build Total Sales, Total Cost, Gross Profit, and Gross Margin using SUM, subtraction, and DIVIDE. Check totals manually.
- 35 minutes: Build report visuals. Add KPI cards, a monthly trend, product or region comparisons, and date and region slicers.
- 35 minutes: Copilot guided activity. Ask questions about the model, request a report page or summary where supported, and compare the response with the report.
- 25 minutes: Improve report design. Apply consistent formatting, clear titles, sensible number formats, accessible colors, and useful interactions.
- 15 minutes: Validation checkpoint. Test measures under filters and identify an incorrect or unsupported AI suggestion.
Session 3: Topics, Practice & Outputs
- Business definitions and basic DAX syntax using actual table and column names.
- Ratio of totals for margin; handling division by zero.
- Selecting charts that match the question and avoiding misleading scales.
- AI-generated DAX, visuals, and narratives as suggestions that require verification.
- Copilot in Power BI and GitHub Copilot as separate products with different roles.
- Practical exercise: Build an executive sales performance page with 4 KPI cards, a monthly sales trend, and product and region comparisons. Use Claude to explain one measure and use Copilot to answer a business question. Confirm both answers against filtered report values.
- Example AI prompt: “Create a Gross Margin measure using the existing Gross Profit and Total Sales measures. Return blank when sales are zero. Explain the formula and describe how I should test it for a region filter and at the grand total.”
- Session output: A report page with validated core measures, interactive visuals, and a brief comparison of an AI answer with the underlying data. If attendee Copilot access is unavailable, the activity uses an instructor demonstration and a supplied response for validation.
Day 2 | Session 4: MCP & Final Business Dashboard
Participants observe how an AI assistant can use tools to work with a Power BI model, then complete and present their report. The MCP portion is a guided demonstration suitable for a public beginner course.
- 15 minutes: MCP explained. Identify the AI assistant, MCP client, Power BI server, tools, and model. Compare a chat suggestion with an executed model change.
- 45 minutes: Instructor demonstration. Connect to a copy of the sample model, inspect tables and measures, propose a measure, approve the change, and query results to validate it.
- 20 minutes: Publishing and sharing overview. Demonstrate publishing, workspace access, refresh considerations, and the role of licenses and permissions.
- 65 minutes: Final project workshop. Complete the sales report, test calculations and filters, refine the layout, and write a short management summary.
- 25 minutes: Participant presentations or peer review. Explain the main findings, show validation evidence, and receive structured feedback.
- 10 minutes: Final knowledge check, questions, and an individual action plan for applying the learning at work.
Session 4: MCP Demonstration Scope & Outputs
- The trainer uses the local Power BI Authoring MCP server with a tested compatible client, such as Claude Desktop or GitHub Copilot in VS Code. The demonstration focuses on semantic model operations: inspect the schema, create one measure, and run a validation query. This server does not author report pages or visual layouts.
- The trainer saves a backup, begins with inspection, shows the proposed change before execution, and compares the result with a known value. Participants observe and explain the workflow; independent installation is an optional follow-up activity.
- Example MCP prompt: “Inspect the connected training model and list its tables and measures. Propose a Gross Margin measure using the existing measures. Show the formula and wait for approval before changing the model. After approval, create it and compare its result with the existing report.”
- Session output: A completed Power BI file, a short management summary, and a validation checklist. Participants can explain where MCP fits, what the demonstrated tools can change, and when they would use Claude or Copilot.
- Important boundary: The demonstration covers semantic model operations. Report pages and visual layouts are completed by participants in Power BI Desktop.
Final Project & Assessment
Business case: A sales manager needs a clear view of revenue and profitability across months, products, and regions. Participants use the supplied fictional business dataset to identify performance differences and support management decisions. Any explanation of a cause must be supported by the dataset or presented as a question for further investigation.
- One saved Power BI file with cleaned data, a simple star schema, and a connected date table.
- 4 validated measures: Total Sales, Total Cost, Gross Profit, and Gross Margin.
- One executive report page with KPI cards, a monthly trend, product and region comparisons, and interactive filters.
- A short management summary containing 3 findings and 2 recommended actions or follow-up questions.
- A validation checklist covering totals, relationships, filters, zero sales, and at least one AI suggestion.
- A brief record of 2 useful prompts, the tool used, and how the results were checked.
- Assessment criteria: Data preparation and model 25%; calculations 25%; report design 20%; AI use and validation 15%; business interpretation 15%.
Assessment combines the project, session checkpoints, and the final knowledge check. The rubric supports trainer feedback and participant self-assessment; it is not a vendor certification examination.
Delivery Requirements: AI Tools & Access
- Claude supports the prompt, formula explanation, and troubleshooting exercises. Direct access to a Power BI model requires a separately configured MCP connection; ordinary chat does not provide that connection.
- Copilot in Power BI requires eligible paid Microsoft Fabric or Power BI Premium capacity, supported regional availability, and the relevant administrator settings. The provider should arrange and test a licensed training environment. A Power BI Pro or Premium Per User license alone does not provide the required capacity.
- The MCP demonstration requires the local Power BI Authoring MCP server, a compatible configured AI client, and the sample model open in Power BI Desktop. The trainer should test the chosen Claude or GitHub Copilot setup, permissions, required runtime, and any license terms before class. GitHub Copilot access is separate from Copilot in Power BI.
Teaching Approach
- Short explanations, instructor demonstrations, guided practice, and review checkpoints. Participants build one report across all 4 sessions.
- The teaching-approach page states that breaks sit outside the 720 instructional minutes, while session pages place short breaks within the 9:00 AM–3:00 PM schedule.
- If live AI access is unavailable, prepared responses and the reference model are used so participants can still practice evaluation.
- Advanced DAX, production deployment, enterprise governance, and independent MCP administration are beyond this introductory course.
AI Prompting Toolkit
- Method used throughout: State the business objective, supply the relevant schema or approved sample, name the exact task, define constraints, and specify the desired output. Ask the assistant to identify missing information and describe how to test its answer.
- Claude — Report layout: “Suggest a layout for a sales manager using these 4 KPIs and these dimensions. Explain which decision each visual supports.”
- Copilot — Performance summary: “Summarize monthly sales performance and identify the largest regional differences. State the filters and time period used.”
- Validation — DAX review: “Review this DAX measure and its business definition. Identify assumptions and propose tests for blank values, zero sales, and filtered totals.”
Preparation & References
- Prepare a fictional sales dataset with Sales, Products, Regions, and Date tables. Include Sales Amount and Cost Amount fields with clear definitions.
- Prepare exercise instructions, a data dictionary, KPI definitions, a prompt sheet, validation answers, and a completed reference report.
- Save checkpoint Power BI files after each session so participants can recover and continue.
- Test imports, refresh, Copilot prompts, and the full MCP demonstration before delivery. Keep screenshots or a recorded demonstration available if a live connection fails.
- Use approved training data for all AI activities. Explain how to review AI outputs, control tool permissions, and retain a backup before model changes.
- Technical references: Microsoft Learn — Copilot for Power BI overview; Microsoft Learn — Power BI Authoring MCP server; Microsoft — Power BI Modeling MCP repository. The brochure states that access requirements and capabilities were checked on 23 October 2026 and should be rechecked before each delivery because services and licensing can change.