# Brightmoor Education — Claude Cowork for Finance & Accounting

**Session:** 3.5 hours | **Audience:** ~25 finance, accounting, payroll, and bursar team members  
**Tool:** Claude Cowork only (no terminal, no desktop app — just the browser-based agent)  
**Data:** Synthetic NetSuite data in Snowflake-shaped exports (Excel files)  
**Theme:** Practical, workflow-first, zero theory — get stuff done

---

## The Core Framework (Talk → Jam → Make → Tune)

Every build follows the same repeatable loop. We'll reinforce it across all three demos:

1. **Talk** — Describe what you want in plain English. Be specific about inputs and outputs.
2. **Jam** — Go back and forth with Claude Cowork on the approach. Let it ask clarifying questions.
3. **Make** — Step back and let Claude build. Watch the files appear.
4. **Tune** — Small, specific requests to polish. "Make the header blue" or "Add a pivot table."

---

## Session Flow

### 1. Why We're Here (0:00–0:10) — 10 min

**Opening frame.** No vision statements, no AI hype. Just a simple message:

> "Your friends at other companies told you Claude can do some of your actual work. Today we're going to prove it — on real finance tasks, with data that looks like your data."

**Quick check-in:** What's one task this week that felt like it should've taken half the time? (Chat answer, not spoken.)

---

### 2. Claude Cowork vs. ChatGPT (0:10–0:20) — 10 min

**Why not just use ChatGPT?** Two things Claude Cowork can do that ChatGPT can't:

| Can do | Why it matters for your work |
|---|---|
| **Reads your files** | Point it at an Excel file on your computer. It sees every row, every sheet |
| **Writes code that runs** | It can run Python to do analysis, build HTML dashboards, update Excel, create PowerPoints — all from your machine |

**Demo:** Open Claude Cowork. Drag in an Excel file. Ask: *"How many rows and columns does this have? Give me a summary of each column."*

**~Hands-on (3 min):** Everyone opens Claude Cowork and uploads an Excel file. Ask a simple question about it.

---

### 3. Build 1: Financial Dashboard (0:20–1:00) — 40 min

**Use case:** Creating a visual dashboard from a NetSuite GL export.

**The scenario:** Brightmoor has monthly GL data in an Excel export from Snowflake. You want a single-page HTML dashboard that shows the key financial picture to your executive — no BI tool required.

**Synthetic data:** `brightmoor_gl_summary.xlsx` — 12 months of revenue, expense, and headcount data by department and entity. 4 sheets (P&L by month, Revenue breakdown, Expense detail, Headcount).

**Step-by-step:**

1. **(Talk — 5 min)** Ask Claude Cowork about the data: *"Open the GL summary file and tell me what's in each sheet. What are the key things I'd want to track monthly?"*

2. **(Jam — 5 min)** In Plan Mode: *"I want to build a single HTML dashboard showing revenue vs expenses by month, department breakdown, and a quick summary table. What data do you need from me?"* Let Claude ask questions and converge on a plan.

3. **(Make — 15 min)** Let Claude build. It will:
   - Read the Excel file
   - Write Python to aggregate and reshape
   - Generate a self-contained HTML file with charts (Chart.js)
   - Open it in the browser

4. **(Tune — 10 min)** Individual customization:
   - *"Change the colors to match Brightmoor's palette (dark navy, gold, white)."*
   - *"Add a YTD variance column to the summary table."*
   - *"Show me the month where expenses were highest."*

5. **Quick share (5 min)** — Turn to a neighbor. Show your dashboard. One thing you changed.

---

### 4. Build 2: Financial Analysis in Excel (1:00–1:50) — 50 min

**Use case:** Using Python (via Claude Cowork) to do reconciliation and trend analysis, outputting results back into Excel.

**The scenario:** Month-end reconciliation. You have a NetSuite trial balance export and a bank statements export. You need to flag differences, identify trends, and produce a clean output file.

**Synthetic data:** 
- `brightmoor_trial_balance.xlsx` — GL account balances by entity
- `brightmoor_bank_transactions.xlsx` — bank statement lines
- About 2,000 rows each — messy enough to feel real

**Step-by-step:**

1. **(Talk — 5 min)** Upload both files. *"I have trial balance data and bank transaction data. I need to find accounts where the GL balance doesn't match the bank balance. Can you show me the differences?"*

2. **(Jam — 5 min)** *"I want a new Excel file with three sheets: 1) accounts with differences over $1,000, 2) a trend analysis of expenses this quarter vs last quarter, 3) a summary dashboard sheet with key metrics. Walk me through what you'd build."*

3. **(Make — 20 min)** Claude Cowork:
   - Loads both Excel files
   - Performs the reconciliation (merge, compare, flag differences)
   - Computes quarter-over-quarter trends
   - Writes a new Excel file (`brightmoor_reconciliation_output.xlsx`) with formatted sheets and conditional formatting
   - Creates a summary dashboard sheet with sparklines and conditional formatting

4. **(Tune — 15 min)** Customization:
   - *"Add a pivot table on the differences sheet grouped by account category."* 
   - *"Show me any accounts where the variance has grown for 3+ consecutive months."*
   - *"Highlight cells in red where variance > 5%."*
   - *"Add a cover sheet with a summary paragraph describing what you found."*

5. **Quick share (5 min)** — Open the output Excel file. Show one insight you found.

---

### ☕ Break (1:50–2:10) — 20 min

---

### 5. Prince of PowerPoint (2:10–2:55) — 45 min

**Use case:** Taking existing Brightmoor data and automatically generating a monthly presentation — charts, tables, formatted slides — via Claude Cowork.

**The scenario:** Every month you need to produce a 6-slide finance review deck for leadership. Same template. Same data sources. You want Claude to do the heavy lifting.

**Synthetic data:** `brightmoor_monthly_data.xlsx` — same NetSuite-shaped data but framed as "this month's numbers."

**Step-by-step:**

1. **(Talk — 5 min)** *"I have this month's financial data in Excel. I need a 6-slide PowerPoint deck: cover slide, executive summary, revenue by category, expenses vs budget, headcount vs plan, and a key metrics slide. I'll give you a template if you need one — but try your best without it first."*

2. **(Jam — 5 min)** Claude may ask: *"Do you have a template .pptx file I can use as a base? What font do you use? What colors?"* If no template, it will create from scratch with reasonable defaults.

3. **(Make — 20 min)** Claude Cowork:
   - Reads the Excel data
   - Uses `python-pptx` to build the deck
   - Creates charts (bar charts, line charts) directly on slides
   - Applies consistent formatting
   - Saves `brightmoor_monthly_review.pptx`

4. **(Tune — 10 min)** Refinement:
   - *"Change the color scheme to navy and gold."*
   - *"Add a footnote on every slide: 'Confidential — Brightmoor Education'."*
   - *"Add a new slide: Top 5 expense variances this month, with explanations."*
   - *"Make the revenue chart a stacked bar showing each department."*

5. **Show (5 min)** — Open the .pptx. Flip through the slides. Compare before/after tuning.

---

### 6. Open Build — Your Own Workflow (2:55–3:20) — 25 min

**Free build time.** Pick one of the three patterns and apply it to something real you worked on this week.

**Prompt starters they can paste:**

| If you want to… | Start with this prompt |
|---|---|
| **Dashboard from data** | *"I have a spreadsheet of [describe data]. Build me an HTML dashboard that shows [key metrics]."* |
| **Analyze in Excel** | *"Take this file. Clean it, analyze trends, and output a new Excel file with a summary sheet and a detail sheet. Flag anything unusual."* |
| **Make a slide deck** | *"Take this data and create a 5-slide PowerPoint. Each slide should show one key insight with a chart."* |

**TAs/helpers circulating to assist.** If someone gets stuck, ASK Ladder:

- *"Ask Claude to fix it."* — Paste the error, say "fix this."
- *"Simplify."* — *"Explain what you're doing in simple terms, then try a different approach."*
- *"Kill and restart."* — Start a fresh session. Paste the prompt again.

---

### 7. Wrap-Up & What's Next (3:20–3:30) — 10 min

**Recap the three patterns:**

| Pattern | When to use | Prompt template |
|---|---|---|
| **Dashboard** | Data needs to be seen, not just read | *"Build an HTML dashboard from this file showing [X], [Y], [Z]."* |
| **Analysis** | You need to clean, compare, or find trends | *"Analyze this data and output a formatted Excel file with findings."* |
| **Slides** | You need to report data to someone else | *"Create a PowerPoint from this data with [X] slides."* |

**One action item:** Before next week, pick one repetitive task and try the relevant pattern. Use the prompt starters above.

**Materials they walk away with:**
- The three synthetic data files + their own built dashboards, analysis files, and slide deck
- A one-page cheat sheet with the prompt starters

---

## Prep Notes for the Facilitator

### Synthetic data to generate

1. **`brightmoor_gl_summary.xlsx`** — 4 sheets:
   - P&L by month (12 months × revenue/expense lines × 3 departments: Academics, Operations, Admin)
   - Revenue breakdown (tuition, grants, other — monthly)
   - Expense detail (payroll, facilities, software, professional services — monthly)
   - Headcount by department (monthly, FTE and contractor split)

2. **`brightmoor_trial_balance.xlsx`** — Account, entity, GL balance, period — ~1,000 rows
3. **`brightmoor_bank_transactions.xlsx`** — Date, description, amount, entity — ~1,500 rows with intentional mismatches
4. **`brightmoor_monthly_data.xlsx`** — Clean, presentation-ready summary — P&L, budget vs actual, headcount

### Formatting tips

- Keep synthetic data "NetSuite-shaped": account codes like `4000-Revenue-Tuition`, entities like `Brightmoor-Academy` and `Brightmoor-Online`
- Include realistic education-specific line items: tuition revenue, grant revenue, facilities expense, software subscriptions, payroll by department
- Add some intentional variance (accounts where GL ≠ bank) so the reconciliation exercise works
- Put budget vs. actual columns in the monthly data so the PowerPoint can show variance

### Key talking points (keep punchy)

- **Claude Cowork reads and writes files** — that's the whole superpower vs. a chatbot
- **You don't need to know code** — you describe what you want in English, Claude writes the code
- **Excel stays the source of truth** — Claude works with what you already have
- **The Talk → Jam → Make → Tune loop** is the repeatable process they take home

### Timing guardrails

- If build 1 runs long, skip the tune step and move to build 2 — 15 min of tune is optional
- If build 2 runs long, the Excel output alone is the win — the trend analysis and dashboard sheet are bonus
- PowerPoint build is the most impressive to execs — protect that time even if it means cutting open build short
- Keep the "why we're here" and "Claude vs ChatGPT" segments tight — these people want to build, not listen

### What to avoid

- No terminal, no CLI, no VS Code — Claude Cowork is the entire tool
- No GitHub, no deployment — irrelevant for this audience
- No theory about how LLMs work — they don't need it
- No "Claude Code" terminology — use "Claude Cowork" consistently
- No ASCII art or "fun" icebreakers — the icebreaker is "what's a task that took too long this week"