FAST-11 Projects: AI-Assisted Analytics and Business Intelligence
Projects use fictional data from the course folder or a sanitized version of a report from your own job (no customer personal information, PHI, or confidential financials; your manager approves the dataset). Scoring: Not yet / Meets / Strong; completion needs Meets on every capstone criterion (one resubmission allowed).
Capstone: AI-Assisted Dashboard and Data Story for a Real Decision
What you produce: a dashboard or report page (4+ visuals) plus a one-page data story, with W1 (transformation log), W2 (question-and-verify log), W4 (fact-check) and W5 (sharing checklist) attached. Presented in 5 minutes in session 5.
Required contents
| Part | Content |
|---|---|
| Decision | Who decides what, by when |
| Data | Source, rows, period, sensitivity classification; W1 attached |
| Questions | Three to five reader questions |
| Verification | At least three AI answers checked by an independent method; mismatches explained (W2) |
| Dashboard | 4+ visuals, each matched to a question, takeaway titles, labeled axes and units, source note |
| Data story | Answer and action first; three points; uncertainty stated; decision needed; W4 fact-check attached |
| Sharing | Label, audience, permissions, RLS if needed, refresh (W5) |
Rubric
| Criterion | Not yet | Meets | Strong |
|---|---|---|---|
| Clean data | Changes not logged or values guessed | Every transformation logged with rows and reason; reproducible | Partner reproduced totals; refreshable query steps |
| Verified answers | Fewer than three checks, or mismatches unexplained | Three or more independent checks; every mismatch explained | Checks cover a share, a trend, and a filtered count |
| Visuals | Misleading axes, wrong types, or clutter | 4+ visuals fit their questions; titles state takeaways; labels and units | Tested with a reader and revised |
| Narrative | Any unsupported or overstated claim | Every claim traced to a number or chart; answer first | Uncertainty quantified; method in an appendix |
| Sharing | Wrong audience, edit for readers, or public link | Label, least-privilege access, RLS where needed | RLS tested with "View as"; Copilot answers tested as a viewer |
Weak vs. strong
| Weak | Strong |
|---|---|
| "Scrap is way up and VF-4 is the problem." | "Recommend a VF-4 bearing inspection this week: VF-4 produced 41% of Q3 scrap cost ($6,602 of $16,102), up 15% from July to September. Verified with SUMIFS and a pivot (W2)." |
| Chart titled "Chart 1" with a y-axis starting at $1,000 | Sorted bar chart from zero, titled "VF-4 produced 41% of Q3 scrap cost," source note under it |
| "Shared with the team." | "Confidential label; agents have Viewer access through the app with an RLS role by agent, tested as two agents; refresh nightly at 2 am, last refresh shown." |
Mini project 1: The Mismatch Casebook
Task: Ask 15 plain-language questions of a fictional course dataset (inn bookings, dental schedule, or renewals). Verify at least eight. Write a one-page casebook of every mismatch: question, AI answer, check, cause (wrong column, hidden filter, average of averages, date range, wording), and the prompt or setting that fixed it.
| Criterion | Not yet | Meets | Strong |
|---|---|---|---|
| Coverage | Fewer than 8 checks | 8+ independent checks | Checks across 3+ question types |
| Diagnosis | Mismatches noted, not explained | Each cause identified | Fix tested and re-verified |
| Lesson | None | Three "check first" rules for your team | Rules turned into a one-page team checklist |
Example entry: "Q7: 'Average weekend rate in August.' AI: $312. Check (AVERAGEIFS on weekend nights): $289. Cause: AI included Friday as a weekend night. Fix: define weekend as Saturday and Sunday nights in the prompt. Re-verified: $289."
Mini project 2: Chart Makeover Portfolio
Task: Collect three charts (AI-drafted in class, or public charts with no confidential data). For each: the original, the problem (axis, type, clutter, labels, unsupported title), the fixed version, and a one-sentence takeaway title. Add a text alternative for each fixed chart.
| Criterion | Not yet | Meets | Strong |
|---|---|---|---|
| Diagnosis | Generic "looks bad" | Specific problems named | References the chart guidance checklist |
| Fix | Cosmetic only | Type, axis, labels, and title fixed; numbers checked | Highlight color used to focus on the message |
| Accessibility | None | Text alternative for each chart | Color contrast checked |
Example: "Original: 3D pie of 12 months of room-nights. Problem: wrong type for a trend, 3D distortion. Fix: line chart, July peak highlighted. Title: 'Room-nights peaked in July at 1,240.' Text alternative: 'Line chart of monthly room-nights, May to October 2026, rising from 610 in May to a peak of 1,240 in July and falling to 520 in October.'"