# FAST-04 Projects: AI Productivity with Microsoft Copilot and Google Gemini Provider: 7formations | Course: FAST-04 | Instructor: Wadood Chaudhary | One capstone and two mini projects Projects turn the four suite workflows (inbox, meetings, documents and slides, spreadsheets) into habits with measured results. Scoring uses **Not yet / Meets / Strong**. Completion requires Meets on every capstone criterion; mini projects are formative unless the employer asks for scoring. One resubmission allowed. Sandbox or sanitized material only; never real customer, patient, or employee data (G.L. c. 93H, 201 CMR 17.00). Follow employer policy on recording and transcription. --- ## Capstone: My 30-Day Productivity Playbook **Brief.** Extend your W5 playbook into a 30-day trial. Document five recurring tasks across at least three of the four areas (mail, meetings, documents/slides, spreadsheets), run each at least three times, log time including checking and errors caught, run a personal permissions check, and present results and decisions to your manager. **Deliverables** 1. Playbook (five tasks): steps, prompt, check, data rules, owner. 2. Run log: 15+ runs (3 per task) with time by hand (baseline), time with AI, checking time, errors caught, errors found later. 3. Permissions self-check: list of files or folders you own that are shared org-wide or by link, and what you changed (no file contents). 4. Recording/transcription note: which of your meetings may be transcribed under policy and which may not. 5. Results page (one page): time saved per task after checking, error patterns, keep / adjust / stop decisions. 6. Two live demos (session 4) and a 5-minute share-back to your manager or team. **Step plan** | Step | When | What you do | | --- | --- | --- | | 1 | Session 1-3 | Pick tasks; record baselines by hand | | 2 | Session 4 | Finalize playbook; demo two tasks | | 3 | Days 1-14 | Run each task at least twice; log | | 4 | Days 15-25 | Adjust prompts and checks; third run; permissions self-check | | 5 | By day 30 | Results page; share-back; submit | **Rubric** | Criterion | Not yet | Meets | Strong | | --- | --- | --- | --- | | Accuracy checking | Checks missing from runs | Every run logged with its check and errors | Error trends drive prompt or process changes | | Permissions awareness | No self-check | Self-check done; over-shared items fixed or reported | Team habit proposed (e.g., quarterly sharing review) | | Applied playbook | Fewer than five tasks or areas | Five tasks across three areas, complete fields | Tasks reusable by a coworker without help | | Measured value | No baseline or checking time omitted | Baseline and checking time for every run | Includes at least one honest "stop" or redesign | | Communication | No share-back | Clear one-page results and share-back | Manager decision recorded (adopt, expand, stop) | **Strong vs weak example (one task)** | | Weak | Strong | | --- | --- | --- | | Task | "Meetings" | "Weekly ops stand-up recap (internal, transcribed with notice per policy)" | | Prompt | "Summarize" | Decisions / actions (owner, date, or 'not stated') / open questions; 12 lines max | | Check | none | Line-by-line vs transcript; owners confirmed; relative dates converted | | Time | "Saves an hour" | Baseline 35 min; AI 2 min + check 12 min = 14 min; saved 21 min; 2 owner errors caught in 3 runs | | Decision | none | Keep; added "owner not stated" rule after run 2 | --- ## Mini project 1: Recap Error Catalogue **Brief.** Using sandbox or policy-approved internal meetings only, collect three AI recaps and mark every error by type. Build a short catalogue of error types and the prompt or check changes that reduce them. **Deliverables.** Three marked recaps (sanitized; no names beyond roles), error counts by type, two prompt changes tested, a one-paragraph team guideline. **Step plan.** (1) Confirm policy and consent; (2) generate recaps; (3) mark errors; (4) change prompts; (5) retest; (6) write guideline. **Rubric** | Criterion | Not yet | Meets | Strong | | --- | --- | --- | --- | | Error marking | Errors missed | Every error marked by type | Errors linked to transcript lines | | Improvement | No changes tested | Two prompt changes tested | Error count reduction shown | | Policy | Consent not addressed | Consent and policy confirmed for each meeting | Guideline names meetings never to transcribe | **Strong vs weak example.** Weak: "The AI got some owners wrong." Strong: "Across three recaps, 4 of 9 errors were owners assigned to the person who raised an item rather than the one who took it. Adding 'owner = person who agreed to do it; if unclear, write owner not stated' cut owner errors to 1." --- ## Mini project 2: Spreadsheet Answer Audit **Brief.** Take a non-sensitive work spreadsheet (or the sandbox workbook) and ask the assistant five plain-language questions. Verify every answer manually and document where and why they differ. **Deliverables.** Five questions; assistant answers with formula or pivot shown; manual results; difference analysis; a "how to prepare a sheet for AI questions" checklist (clean table, one header row, no merged cells, no totals in data). **Step plan.** (1) Clean a copy of the sheet; (2) ask; (3) verify each by formula or pivot; (4) explain differences; (5) write the checklist. **Rubric** | Criterion | Not yet | Meets | Strong | | --- | --- | --- | --- | | Verification | Answers not checked | All five checked manually | Checks reproducible by a coworker | | Diagnosis | Differences unexplained | Each difference explained (filter, blanks, duplicates, method) | Data fixed and questions re-asked | | Data handling | Sensitive data used or original altered | Non-sensitive copy; original untouched | Preparation checklist adopted by the team | **Strong vs weak example.** Weak: "AI said North Shore, I said Metro West, so AI is wrong." Strong: "The AI respected the leftover LTL filter (North Shore −$6,300). With filters cleared, Metro West dropped most (−$6,500, −11.6%), driven by FTL. The tool was right for the data it saw; the sheet was wrong. Added 'clear filters' to the checklist."