Claude Cowork for Finance: Brightmoor Education

Recommended run of show

Format: 3.5-hour, hands-on beginner workshop
Audience: 20–30 people across accounting, payroll, bursar, and strategic finance
Tools: Claude Cowork and Claude in Excel only
Working scenario: A monthly operating review built from a synthetic, NetSuite-shaped export from Snowflake
End products: A checked analysis workbook, an executive HTML dashboard, and an updated PowerPoint deck

The session should follow one continuous workflow rather than three unrelated demos:

NetSuite-shaped Snowflake export → Excel checks and analysis → executive dashboard → monthly PowerPoint

This mirrors Brightmoor's actual work and keeps the session focused on getting a recurring finance deliverable done. It also makes the central lesson easy to remember: Claude can do the file work, but the finance owner still verifies the numbers.

What participants should be able to do by the end

1. Give Claude Cowork a folder of finance files and ask it to inspect, analyze, and create new deliverables without using a terminal or writing code.
2. Use Claude in Excel to understand a workbook, add calculations, investigate variances, and build visible checks.
3. Turn validated finance data into a clean, filterable HTML dashboard for leadership.
4. Use an existing PowerPoint template and chart as a style reference, then ask Cowork to update the deck with the new month's numbers and commentary.
5. Re-run the workflow next month while preserving source files and checking every output against control totals.

Detailed facilitator run sheet

Times below are elapsed so the schedule works regardless of the event's start time.

Time	Segment	What happens	Participant output
0:00–0:10	Start with the job	Set the goal: reduce the manual work between a NetSuite/Snowflake export and a leadership-ready monthly report. Show the three finished outputs. No AI trends, icebreaker, or speculative use cases.	Clear view of the finish line
0:10–0:25	First useful result in Excel	Open the synthetic monthly workbook in Excel. Use Claude in Excel to explain the workbook, calculate an actual-versus-budget variance, and flag the five largest exceptions. Replace the usual ASCII warm-up with a real finance task.	A useful result in the first 15 minutes
0:25–0:35	Cowork in plain English	Explain only what is needed: unlike a typical ChatGPT-style conversation that mainly returns an answer in the chat, Cowork can work with files in an approved folder, run code behind the scenes, and create Excel, HTML, and PowerPoint files. Demonstrate selecting the workshop folder and requiring new versions rather than overwriting source files.	A safe working folder and a mental model
0:35–1:20	Build 1: financial analysis and reconciliation	In Cowork, inspect the export, map accounts, calculate revenue and EBITDA, compare actuals with budget, split payroll and non-payroll costs, identify campus trends, and reconcile totals. Save the result as a new Excel workbook with a Checks sheet.	Brightmoor_Monthly_Analysis.xlsx
1:20–1:30	Verify and debrief	Tie source totals to the analysis, inspect unmapped accounts and duplicates, and review the largest variances. Ask participants to fix one issue through a plain-language follow-up prompt.	A checked workbook and exception log
1:30–1:40	Break	Ten-minute break.	—
1:40–2:25	Build 2: executive dashboard	Give Cowork the checked workbook and ask for a standalone HTML dashboard. Add leadership-friendly KPI cards, campus and period filters, actual-versus-budget charts, an EBITDA bridge, and short callouts. Participants make two visible changes, such as changing a chart to horizontal bars and adding a filter.	Brightmoor_Executive_Dashboard.html
2:25–3:05	Build 3: refresh the monthly PowerPoint	Give Cowork the checked workbook, a six-slide monthly-review template, and one completed chart as a style reference. Update charts, tables, titles, and draft commentary. Demonstrate the high-value shortcut Brightmoor requested: mirror one approved chart style across the next five slides.	Brightmoor_Monthly_Review_Updated.pptx
3:05–3:20	Make it repeatable	Ask Cowork to write a one-page monthly refresh checklist: which files to replace, what prompt to reuse, what must tie out, and what requires human review. Demonstrate a new-month rerun using a second synthetic export.	Monthly_Refresh_Checklist.docx or .md
3:20–3:30	Close on the outputs	Show the three artifacts side by side. Recap the working pattern below and take only workflow-specific questions. Give participants the prompt card and checklist.	A reusable process, not just a demo

The agenda provides about two hours of hands-on work. The facilitator should type the first prompt with the group, then pause after each major action so beginners do not fall behind.

The working pattern to repeat in every exercise

Use a finance-specific version of the existing build loop:

1. Brief: State the business outcome, audience, files, and required output.
2. Inspect: Ask Claude to describe the data, assumptions, and proposed checks before changing anything.
3. Make: Let Claude create a new version of the deliverable.
4. Verify: Tie totals to the source, review exceptions, and make targeted corrections.
5. Tune: Change presentation, commentary, filters, or chart types only after the numbers pass checks.

The order matters: verify before beautifying.

Demo prompts

These are deliberately written in normal workplace language. Participants should edit the month, audience, and thresholds rather than learn prompt syntax.

1. Claude in Excel: first practical prompt

Review this workbook and explain what each sheet contains in plain English. Do not change anything yet. Identify the fields we can use to compare actuals with budget by month and campus, and tell me about any missing values, duplicate rows, unmapped accounts, or inconsistent signs that could affect the analysis.

Follow with:

Add a new sheet called Variance Review. Show actual, budget, dollar variance, and percentage variance by campus and reporting line. Clearly flag the five largest unfavorable variances. Add a total row and visible checks that tie actuals and budget back to the source sheets. Preserve the original sheets.

2. Cowork: analysis and reconciliation

Work only in the Brightmoor workshop folder. Use Brightmoor_Monthly_Finance.xlsx as the source and do not overwrite it. First inspect the workbook and give me a short plan, including the control totals you will use.

Then create Brightmoor_Monthly_Analysis.xlsx. Include revenue, operating expenses, and EBITDA for the current month and year to date; actual versus budget in dollars and percentages; payroll versus non-payroll costs; and trends by campus. Create separate sheets called Executive Summary, Variance Analysis, Campus Trends, Reconciliation, Checks, and Assumptions.

Flag duplicate records, unmapped accounts, missing campus values, and reconciliation differences above $1,000. Put every exception in the Checks sheet. Do not invent missing values. If a definition is unclear, record the assumption instead of silently choosing one. At the end, report whether source totals tie to the output and list anything a finance owner must review.

Useful follow-ups:

Explain the three largest unfavorable variances in plain English. Separate facts supported by the workbook from possible explanations that need an owner to confirm.

Reconcile the GL payroll accounts to the payroll detail. Show the difference by month and campus, and flag anything over $1,000.

Create a repeatable compliance report showing the source, transformation, control total, exception, reviewer, and review status for each check.

3. Cowork: executive HTML dashboard

Use the checked numbers in Brightmoor_Monthly_Analysis.xlsx to create a polished standalone HTML dashboard for Brightmoor's senior leadership. Do not recalculate or alter the financial data.

Include KPI cards for revenue, EBITDA, EBITDA margin, and total operating expense; actual-versus-budget comparisons; a monthly trend; an EBITDA bridge; campus performance; and a short Key Takeaways section. Add filters for month and campus. Use a restrained executive style with accessible colors, clear labels, dollar formatting, and unfavorable variances shown consistently. Include the reporting period, data source, and last refresh date.

Save it as Brightmoor_Executive_Dashboard.html. Before finishing, compare every displayed total with the Checks sheet and list any differences.

Tuning prompts:

Change the campus comparison to horizontal bars, sort from largest unfavorable variance to largest favorable variance, and keep the values visible on the chart.

Add a campus filter and a Reset Filters button. Do not change any totals or other parts of the dashboard.

Make the page easier to scan in a leadership meeting: reduce visual clutter, increase label size, and put the three most decision-relevant insights above the fold.

For the workshop, “publish” should mean producing a self-contained HTML file and showing how it opens in a browser. Sharing it beyond finance should happen only through a Brightmoor-approved internal location. Public hosting, authentication, and live Snowflake connections are separate IT and security decisions, not beginner workshop tasks.

4. Cowork: PowerPoint refresh

Update Brightmoor_Monthly_Review_Template.pptx using only the checked figures in Brightmoor_Monthly_Analysis.xlsx. Preserve the slide master, fonts, colors, footers, and existing layout. Do not overwrite the template; save the result as Brightmoor_Monthly_Review_Updated.pptx.

Update the reporting period and create six slides: executive summary, revenue versus budget, EBITDA bridge, campus performance, payroll versus non-payroll, and reconciliation/open items. Use the approved chart on slide 2 as the style reference for charts on the next five slides. Keep chart scales and labels honest and consistent.

Draft no more than three commentary bullets per slide. Label statements based directly on the data as findings. Put possible business explanations in speaker notes and prefix them with “To confirm:” rather than presenting them as facts. Add a small source note to each slide. Then compare the figures on every slide with the workbook's Checks sheet and provide a slide-by-slide QA summary.

Tuning prompts:

Slide 4 is too crowded. Keep the same content, but simplify it for a two-minute executive discussion without changing any numbers.

Mirror the approved chart treatment from slide 2 across slides 3–6, including font sizes, colors, labels, and spacing. Leave the underlying values unchanged.

Check the deck for clipped text, overlapping objects, inconsistent number formats, stale month labels, and unsupported commentary. Fix presentation issues and list every change you made.

Synthetic workshop pack

Use synthetic data so Snowflake access is optional and no attendee is blocked by permissions. The data should resemble Brightmoor's workflow without pretending to reproduce its exact production schema.

Finance workbook

Brightmoor_Monthly_Finance.xlsx should contain 12 months of data and these sheets:

• GL_Actuals: period, transaction date, subsidiary, campus, department, class/program, account number, account name, account type, document number, vendor, memo, debit, credit, net amount, currency, and source system.
• Budget: period, campus, department, account number, and budget amount.
• Account_Mapping: account number, management-reporting line, payroll/non-payroll classification, EBITDA inclusion, and display sign.
• Payroll_Detail: synthetic employee ID, period, campus, department, gross pay, employer tax, benefits, and total payroll cost.
• Reconciliation: period, campus, GL balance, subledger balance, difference, owner, and status.
• Instructions: fictional-data notice, business definitions, sign conventions, and expected control totals.

Include a small number of intentional issues so the checking exercise is real: one duplicate journal, two unmapped accounts, a missing campus, an inconsistent expense sign, and two payroll reconciliation differences. Provide a clean answer key to the facilitator only.

Dashboard and slide inputs

• A clean Brightmoor-style guide with approved colors, typography, logo placeholder, and number formats.
• Brightmoor_Monthly_Review_Template.pptx, with six slide layouts and one fully styled example chart.
• A prior-month completed deck for visual reference.
• A second month's synthetic workbook for the repeatability demo.
• A one-page participant prompt card containing the four core prompts above.

Verification standard used throughout

Every output should make these checks visible:

• Source row count and source total before transformation.
• Output row count and output total after transformation.
• Actual, budget, and payroll totals tied back to their source sheets.
• Duplicate, missing-value, unmapped-account, and sign-convention exceptions.
• Reconciliation threshold and unresolved differences.
• Assumptions separated from facts.
• Reporting period, source file, refresh time, and output version.
• Named human reviewer and review status before distribution.

The facilitator should say this plainly: Claude accelerates preparation and presentation; it does not approve a close, certify a reconciliation, or own the forecast.

Pre-work and logistics

Send this checklist at least two business days before the session:

• Claude access is provisioned for every attendee; do not spend workshop time creating accounts.
• Claude Cowork opens successfully in the desktop app.
• Claude in Excel is installed, signed in, and visible in Excel.
• Excel, PowerPoint, and a modern browser are available.
• The complete synthetic workshop folder has been downloaded locally and unzipped.
• Each attendee can open the workbook, PowerPoint template, and HTML sample.
• Attendees know not to use production data during the workshop.
• Snowflake access is explicitly optional. The session must work without it.

Facilitator backups should include completed versions of every artifact, screenshots of each major step, a spare copy of the source folder, and a version of the dashboard and deck ready to show if file creation takes longer than expected.

What to leave out

• Claude Code, the terminal, GitHub, repositories, or coding concepts.
• ASCII art or location-based warm-ups.
• General AI history, model comparisons, or “future of work” framing.
• Open-ended build time and speculative use-case brainstorming.
• Public deployment or live system integration during the session.
• Claims that the full 15-slide monthly process will be perfect after one prompt.

The realistic promise is more useful: participants will leave knowing how to produce and iterate on a checked workbook, a leadership dashboard, and a six-slide monthly review from familiar finance-shaped data. Turning that into Brightmoor's production 15-slide workflow will require the approved template, agreed definitions, repeatable control checks, and several monthly iterations.

Assumptions

• The working session is 3.5 hours inside the four-hour block mentioned in the preparation call.
• The attendee count is 20–30 and most participants are beginners.
• Brightmoor will approve the security and sharing rules for Cowork before anyone uses production data.
• The workshop uses synthetic data modeled on a generic NetSuite-to-Snowflake finance export; the exact field names and reporting definitions will be adjusted if Brightmoor supplies sanitized examples.
• Internal dashboard publishing and direct Snowflake connectivity are follow-on implementation topics because they depend on Brightmoor's access controls and IT environment.