# Claude Cowork for Beginners — Brightmoor Education Finance Team
## Run Sheet / Run of Show

**Audience:** 20–30 finance staff (accounting, payroll, bursar). Mostly non-technical, beginner-to-rookie with AI. A small handful already use Claude in Excel.

**Company:** Brightmoor Education (workplace education company)

**Data stack:** NetSuite → Snowflake (central source of truth). The session uses **synthetic NetSuite-shaped Excel exports** so no Snowflake/IT provisioning is needed on the day.

**Platform:** Claude Cowork (web) only + Claude in Excel add-in. **No Claude Code desktop app, no terminal.**

**Length:** 3.5 hours (4-hour block with buffer)

**Goal:** Three practical builds, each mapped to a real weekly/monthly finance task:
1. Build an HTML dashboard from Excel data (make the numbers look visibly better).
2. Run financial analysis in Python and output to Excel (reconciliation / trend).
3. Update a PowerPoint template automatically (weekly/monthly executive deck).

---

## How to run this session

- **One shared build at a time.** The facilitator drives Claude Cowork on a big screen; every participant follows along in their own Cowork chat with the same data files.
- **No setup friction.** Everyone starts with a clean slate and the same pre-loaded folder of synthetic files. Do not spend session time on installs or provisioning.
- **Talk → Jam → Make → Tune.** For every build: (Talk) describe the end state, (Jam) go back and forth on details, (Make) let Claude build it, (Tune) make small, specific changes to go from functional to good.
- **Finance guardrail built in.** Every numeric output is reconciled back to the source export before anyone trusts it. See the Trust & Accuracy section.

---

## Principles (keep it tight)

- 100% practical. Zero AI hype, zero "we're excited about AI," zero wider theory.
- Every minute maps to a task the team already does (monthly close, campus reporting, executive deck).
- If a task takes more than one session to perfect, teach the *process* for finishing it after the workshop, not just a one-shot result.
- Replace the usual ASCII-art icebreaker with an Excel-based starter (see Session Start).

---

## Pre-work / prerequisites (complete before the day)

| Item | Owner | Notes |
|------|-------|-------|
| Claude Cowork access provisioned for all 20–30 attendees | IT / Brightmoor | Confirm before the session |
| Claude in Excel add-in available | IT / Brightmoor | Used in Build 2 path |
| Synthetic data folder pre-loaded on each machine | Facilitator | Four CSV files (see Synthetic Data) |
| One PowerPoint template (monthly finance deck) | Brightmoor | Used in Build 3; bring a real template if available, otherwise synthetic |
| Facilitator screen-share of Claude Cowork | Facilitator | One shared build projected for the room |

---

## Synthetic data (NetSuite-shaped)

All files live in `session_assets/synthetic_data/`. Shape mirrors a NetSuite export so the team recognizes the columns.

### 1. `ns_gl_transactions.csv` — General Ledger transaction export
`Transaction Date, Transaction Type, Document Number, Account, Account Name, Account Type, Department, Class, Location (Campus), Entity, Entity Name, Debit Amount, Credit Amount, Memo`

Sample rows:
- Tuition revenue invoices across Downtown / North / West / Online campuses
- Instructor pay bills, facilities rent, software subscriptions, marketing, utilities, payroll journals
- Two months (Jan–Feb 2026) so trend analysis is possible

### 2. `ns_chart_of_accounts.csv` — Chart of accounts
`Account, Account Name, Account Type, Subsidiary, Currency`

Income (Tuition Revenue, Program Fees), Expense (Instructor Pay, Facilities Rent, Software & Subscriptions, Marketing, Utilities, Payroll Expense), Asset (Cash, AR), Liability (AP), Equity (Retained Earnings).

### 3. `ns_campus_budget.csv` — Campus budget vs actual
`Campus, Program, Budget Category, FY26 Budget, YTD Actual, Variance $, Variance %`

Four campuses × programs, 50% through the year.

### 4. `ns_payroll_export.csv` — Payroll export
`Pay Period, Employee ID, Employee Name, Campus, Department, Role, Gross Pay, Employer Benefits, Payroll Taxes, Total Cost`

Jan–Feb 2026, five staff across campuses.

---

## Minute-by-minute agenda (3.5 hours)

| Time | Min | Segment | What happens |
|------|-----|---------|--------------|
| 0:00–0:10 | 10 | **Welcome & why we're here** | Practical framing: three builds, each solves a real monthly finance task. No theory. |
| 0:10–0:25 | 15 | **Claude Cowork vs ChatGPT + setup** | Key difference: Cowork can read your files and run code; ChatGPT only chats. Open Cowork, upload the data folder, confirm everyone can see the files. |
| 0:25–0:35 | 10 | **The data walkthrough** | Open the four CSVs in Excel. Point out NetSuite-shaped columns (campus, account, debit/credit, period). This is the shape of their real exports. |
| 0:35–1:20 | 45 | **Build 1: HTML dashboard from Excel** | Take GL transactions + budget, produce a styled HTML dashboard (campus revenue, expense by category, variance). Make it look visibly better than raw Excel. |
| 1:20–1:30 | 10 | **Break** | — |
| 1:30–2:15 | 45 | **Build 2: Financial analysis (Python → Excel)** | Reconcile GL to budget, identify trends/variances, output results back to Excel. Show both Claude Cowork (Python) and Claude in Excel paths. |
| 2:15–2:25 | 10 | **Break** | — |
| 2:25–3:10 | 45 | **Build 3: Automated PowerPoint deck** | Feed the analysis into a PowerPoint template; auto-generate charts/slides for a monthly executive deck. |
| 3:10–3:30 | 20 | **Trust, accuracy & wrap-up** | How to check AI numbers (reconcile to source), where each build breaks, next steps, Q&A. |

---

## Session start (replaces the ASCII-art icebreaker)

Instead of an ASCII image, open `ns_gl_transactions.csv` in Excel and ask Claude Cowork:

> "I've uploaded a NetSuite general ledger export for Brightmoor Education. Tell me: (1) how many rows, (2) which campuses appear, (3) which account types, and (4) the date range. Confirm you can read the file."

This proves file-system access in 60 seconds with something the team actually recognizes.

---

## The intro: Claude Cowork vs ChatGPT (keep this)

Say this, almost word-for-word:

> "The difference between Claude Cowork and ChatGPT is access. ChatGPT is a chatbot — it answers questions. Claude Cowork can see your file system, read your Excel and CSV files, run code (Python), and write files back out. For finance work that means it can actually process your NetSuite exports, not just talk about them."

Then demonstrate: upload the data folder, run the session-start prompt above, show the answer.

---

## Build 1 — HTML dashboard from Excel (45 min)

**The task:** Turn monthly GL exports into a clean, campus-level HTML dashboard that looks visibly better than raw Excel — for senior leadership.

**Talk → Jam → Make → Tune**

- **Talk:** "Build a single-file HTML dashboard from the GL transactions and budget CSVs. Show revenue, expense, and variance by campus, with a clean modern design. No build step — one HTML file I can open in a browser."
- **Jam:** Facilitator and room agree on layout (KPI cards at top, campus table below, expense-by-category bar chart, variance highlights). Decide color palette (neutral, finance-appropriate).
- **Make:** Claude Cowork reads the CSVs, computes totals, writes `dashboard.html`. Facilitator opens it in a browser to verify.
- **Tune:** Small requests — "make the variance column red when negative," "add a filter for campus," "switch the expense chart to horizontal bars," "add a generated-on date."

**Sample prompt (paste into Cowork):**

```
I've uploaded ns_gl_transactions.csv and ns_campus_budget.csv (NetSuite-shaped exports for Brightmoor Education).
Build a single-file HTML dashboard that:
- Shows total revenue, total expense, and net by campus (KPI cards)
- Shows expense by account category (bar chart)
- Shows budget vs actual variance by campus (table, highlight negative variance in red)
- Uses a clean, professional design suitable for senior leadership
- Is one self-contained HTML file (embed chart via Plotly CDN or inline SVG; no external build step)

Read the CSVs directly, compute the numbers, and write dashboard.html. Then tell me how to open it.
```

**What good looks like:** One HTML file the team can open offline, numbers match the CSV, campus breakdown matches their existing monthly reporting.

**Deliverable:** `dashboard.html` (generated during the session).

---

## Build 2 — Financial analysis, Python → Excel (45 min)

**The task:** Reconcile the GL to the budget, identify trends and variances, and write the results back into Excel. Two paths shown: Claude Cowork (Python, full control) and Claude in Excel (in-sheet, lighter).

**The scenario:** Month-end reconciliation. "Did actual spend match budget by campus? Where are the variances? Any transactions missing a budget line?"

**Path A — Claude Cowork (recommended for this group)**

Prompt:

```
I've uploaded ns_gl_transactions.csv and ns_campus_budget.csv.
Using Python (pandas), do the following and write results to a new Excel file `financial_analysis.xlsx`:
1. Aggregate actuals by campus and account for Jan and Feb 2026.
2. Join to the budget on Campus + Program + Budget Category.
3. Compute variance $ and variance % for each line.
4. Flag any actuals with no matching budget line.
5. Identify the top 3 expense variances by absolute $.
Output one Excel file with sheets: Summary, Variance by Campus, Unmatched Transactions, Top Variances.
Show me the Python code you ran and the key findings.
```

**Path B — Claude in Excel (lighter alternative)**

- With the data open in Excel, use Claude in Excel to write formulas: `SUMIFS` by campus/account, variance columns, conditional formatting on negative variance.
- Good for quick checks; Cowork/Python is better for multi-file joins and repeatable monthly runs.

**Trust check (live):** Reconcile the Python total revenue against a manual `SUMIF` in Excel. If they match, the team trusts the pipeline. If not, debug together — this is the lesson.

**Deliverable:** `financial_analysis.xlsx`.

---

## Build 3 — Automated PowerPoint from template (45 min)

**The task:** Take the analysis from Build 2 and populate a monthly PowerPoint template so the weekly/monthly executive deck rebuilds itself instead of being copied/paste rebuilt for 8 hours.

**The scenario:** Dana's team spends ~8 hours/month building a 15-page deck. The ask: "If I already have a chart, can you mirror it and make six more on the next five pages?" and "turn my five Excel tables into PowerPoint."

**Talk → Jam → Make → Tune**

- **Talk:** "Use the financial_analysis.xlsx output and a PowerPoint template. Generate a deck with one slide per campus plus an executive summary, charts embedded, numbers filled from the analysis."
- **Jam:** Agree on slide order (exec summary → campus slides → variance detail → appendix). Confirm the template structure (title slide, content layout, chart placeholder).
- **Make:** Claude Cowork uses python-pptx to read the Excel and populate the template, writing `monthly_deck.pptx`.
- **Tune:** "Make the variance slide a horizontal bar chart," "add a footer with the month," "bold the negative numbers."

**Sample prompt:**

```
I've uploaded financial_analysis.xlsx and a PowerPoint template (monthly finance deck).
Using python-pptx, populate the template to create monthly_deck.pptx:
- Slide 1: Executive summary (total revenue, total expense, net, top variance)
- One slide per campus: revenue, expense, net, budget vs actual, with a chart
- Final slide: top 3 variances and recommended actions
- Numbers come from the Excel file, not hardcoded
- Keep the template's theme/colors

Write the deck and tell me how to open it.
```

**If no real template is available:** Use a synthetic template with a title slide and one content layout so the mechanics are still demonstrated.

**Deliverable:** `monthly_deck.pptx`.

---

## Trust & accuracy (finance guardrails)

This comes up in every finance session. Set expectations up front:

1. **Reconcile before you trust.** Every AI number gets checked against a manual Excel total (`SUMIF` / pivot) before it goes in a deck or dashboard.
2. **AI is strong at cleaning and formatting, weaker at judgment.** The team should treat Claude's output as a first draft that a human signs off.
3. **Never paste real student, payroll, or supplier data.** Use the synthetic files in-session; apply the same caution with real exports after.
4. **Where it breaks:** PowerPoint formatting often needs 2–3 iterations. Teach the team to iterate after the workshop, not expect perfection on the first try.

---

## Wrap-up & next steps (20 min)

- Recap the three builds and where each maps to their monthly calendar.
- Hand out the synthetic data folder + a one-page prompt cheat sheet (the three prompts above).
- Assign owners: who runs the first real dashboard, who runs the first reconciliation, who owns the monthly deck template.
- Follow-up: 30-minute office-hours session the following week for the team to bring their own real (sanitized) files.

---

## Session assets

Created for this run:
- `session_assets/synthetic_data/ns_gl_transactions.csv`
- `session_assets/synthetic_data/ns_chart_of_accounts.csv`
- `session_assets/synthetic_data/ns_campus_budget.csv`
- `session_assets/synthetic_data/ns_payroll_export.csv`

These are NetSuite-shaped, Brightmoor-branded synthetic files. No real company data is included.
