AI for Business Intelligence Capstone Exercise — AI for Business Intelligence

What you'll learn

Apply skills from across this course to design a complete AI-augmented dashboard and insight workflow for a realistic business scenario
Produce a concrete, role-relevant deliverable that specifies both the AI-assisted steps and the verification applied to each one
Self-assess your design against the proportional validation standard and workflow embedding principles from this course

This course has taken you from natural language querying through AI-powered dashboards, the specific copilot capabilities inside Power BI and Tableau, anomaly detection, insight validation, and full team workflow design. The throughline across all of it is the same: AI compresses the mechanical work of BI, and the value of that compression depends entirely on whether verification is built in from the start rather than bolted on after something goes wrong. This capstone asks you to design a workflow under realistic constraints and defend every AI-assisted step with a specific check.

Capstone Exercise

Designing an AI-Augmented Sales Operations Dashboard Under a Two-Week Deadline

Context

You are a BI developer at a mid-size B2B software company. The VP of Sales Operations needs a new dashboard covering pipeline health, win rate by segment, and quota attainment across 40 sales reps in six regions, to be reviewed weekly by regional sales directors and monthly by the executive team. You have two weeks to build it. The underlying CRM data has known issues: deal stage naming is inconsistent across two legacy sales teams that were merged eighteen months ago, some closed-lost deals are missing a loss reason, and a handful of large enterprise deals have irregular close-date patterns that could distort any automated trend or anomaly detection applied naively. You plan to use AI tools throughout the build to meet the deadline.

Your Task

Design the full workflow for this dashboard in five parts. Part one: specify the semantic model documentation work you would do first, and why, before enabling any AI querying or copilot feature on this data. Part two: draft the natural language prompt you would use to get an AI-assisted design proposal for the dashboard layout, specifying the business decision, data, and audience as covered in this course. Part three: identify one calculated measure this dashboard needs (for example, win rate by segment) and write out how you would generate it with an AI copilot and the specific row-level test you would run before trusting it. Part four: decide whether and how you would enable automated anomaly detection on any of these metrics, including what you would do about the irregular enterprise deal close-date pattern before trusting any alert. Part five: assign a consequence tier (self-review, peer check, or full owner sign-off) to the weekly regional commentary and the monthly executive commentary, and justify the difference.

Your notes (optional)

Deliverable

A five-part workflow design document covering semantic model preparation, the dashboard design prompt, one fully specified AI-generated measure with its verification test, an anomaly detection decision with justification, and a consequence-tiered sign-off plan for both commentary audiences.

Quick check

Part one asks for semantic model documentation before any copilot or natural language query feature is switched on for this data. What goes wrong if that order is reversed?

Select one answer.

Key takeaways

A trustworthy AI-augmented dashboard starts with semantic model preparation, not with the AI feature — inconsistent naming and undocumented business logic will produce unreliable AI output regardless of which platform or copilot is used.
Every AI-generated measure needs a row-level verification test before it is trusted, and every automated alert needs its training window checked for contamination before it is escalated.
Consequence-tiered review is what makes a validation standard survive a real deadline — the same underlying data can warrant different levels of scrutiny depending on who receives the output and what they will do with it.
Designing the workflow before building the dashboard is what separates an AI-augmented BI practice from a team that is simply hoping the AI output happens to be right.