Custom GPT: Build Your Personal FP&A Analysis Assistant
What This Builds
A Custom GPT is a version of ChatGPT that you configure with your company's specific context, your financial reporting structure, and your preferred writing style — so every conversation starts from shared understanding, not a blank slate. Instead of re-explaining your business units, your CFO's communication preferences, and your report format every month, you set it up once and it remembers everything. Think of it as onboarding a new team member who specializes in financial writing and analysis — but you do the onboarding in a 2-hour setup session instead of weeks.
Prerequisites
- ChatGPT Plus subscription ($20/month) — Custom GPTs require Plus or higher
- Your management report template (or last 3 months of reports to use as training examples)
- Your company's business unit names, key KPI definitions, and standard financial terminology
- 1.5–2 hours for initial setup and testing
The Concept
A Custom GPT is like having a new coworker who already knows your company's structure, your reporting format, and how you like things done. You hire them once (the setup), and every conversation after that starts from that shared baseline. Without a Custom GPT, every Claude or ChatGPT conversation starts fresh and you're constantly re-explaining context. With a Custom GPT, you walk in and say "draft this month's commentary" and it knows exactly what that means.
The three things that make a Custom GPT powerful for FP&A:
- System instructions — Everything your custom GPT knows about your role, company, and preferences
- Knowledge files — Documents you upload (your report template, your KPI glossary, example reports)
- Persistent context — These settings apply to every conversation, automatically
Build It Step by Step
Part 1: Access the Custom GPT Builder
- Go to chatgpt.com and sign in with your ChatGPT Plus account
- Click your profile icon in the top right → "My GPTs"
- Click "Create a GPT"
- You'll see a two-panel interface: the "Create" panel (left) and a preview panel (right)
What you should see: A chat interface on the left where you can describe your GPT, and a preview on the right where you can test it in real time.
Part 2: Name and Configure Your GPT
Click "Configure" at the top of the left panel (this gives you direct control instead of the AI-guided setup).
Fill in:
- Name: "FP&A Analysis Assistant: [Your Company]" (or just "[Company] Finance GPT" if you want something shorter to say)
- Description: "Your monthly close, management reporting, and financial analysis assistant. Knows your company structure, reporting format, and CFO communication style."
- Profile picture: Click the image area and ask DALL-E to generate one, or just leave the default
Part 3: Write the System Instructions
This is the most important step. The instructions tell your GPT everything it needs to know. In the "Instructions" text area, write the following (customize with your actual information):
You are a financial planning and analysis assistant for [Company Name]. You help [Your Name], a corporate financial analyst, with monthly close reporting, management deck preparation, budget analysis, and financial communications.
## COMPANY CONTEXT
Company: [Company Name]
Industry: [Your industry — e.g., B2B software / manufacturing / healthcare]
Fiscal year: [Start month] to [End month]
Reporting currency: USD
Numbers format: Millions with 1 decimal (e.g., $12.4M, not $12,400K)
## BUSINESS UNITS / COST CENTERS
Report on these business units:
- [Business Unit 1] — [brief description of what they do]
- [Business Unit 2] — [brief description]
- [Business Unit 3] — [brief description]
[Add all your business units]
## KEY METRICS AND DEFINITIONS
- Revenue: [How you define revenue — e.g., "recognized revenue per ASC 606"]
- EBITDA: [Your EBITDA definition — what is included/excluded]
- Headcount: [FTE count definition — do you include contractors?]
- Budget: [Is your budget the Board-approved plan or the most recent reforecast?]
[Add any other metrics that have company-specific definitions]
## VARIANCE COMMENTARY FORMAT
When writing variance commentary, use this exact format for each line item:
"[Line item] was $[X]M, $[Y]M [favorable/unfavorable] to budget, primarily driven by [driver]. [Optional context sentence.]"
Rules:
- Each comment must be under 50 words
- "Favorable" = better than budget (less expense, more revenue)
- "Unfavorable" = worse than budget (more expense, less revenue)
- Do not use jargon like "YoY," "QoQ," "run rate" without spelling it out
- Never say "impressive" or "disappointing" — remain neutral and factual
## EXECUTIVE SUMMARY FORMAT
Executive summaries for the CFO are:
- 3–4 short paragraphs
- Under 250 words total
- Opening sentence states the headline result (revenue and EBITDA vs. budget)
- Second paragraph covers key drivers
- Third paragraph covers cash and balance sheet highlights (if relevant)
- Final sentence covers full-year outlook
## SLIDE CONTENT FORMAT
Slide titles are action statements, not category labels.
Good: "West Region Recovery Expected in Q2"
Bad: "Regional Performance"
Bullet points per slide: 3 maximum. Each under 15 words. Lead with a number.
## WRITING STYLE
- Direct and factual
- No jargon — write for an executive with limited patience for accounting detail
- Don't editorialize — no "surprisingly" or "unfortunately"
- Use active voice
- When something is unclear, ask me a clarifying question rather than guessing
## WHAT I WILL PROVIDE
When I start a session, I will typically provide:
- A variance table (actual vs. budget, sometimes vs. prior year)
- Any context about one-time items or unusual drivers
- The specific deliverable I need (commentary, deck narrative, email, etc.)
You should ask for this context if I don't provide it.
Part 4: Upload Knowledge Files
Knowledge files are documents your GPT can reference. Upload files that represent your actual work.
- Click "Add files" in the Knowledge section
- Upload the following:
- Last month's management report (as a PDF or Word doc): this shows the GPT the exact format and style you use
- Your KPI/metric glossary (if you have one, even as a simple Word doc)
- Your business unit descriptions (a list of what each unit does)
- Your standard variance commentary examples (copy-paste 10 examples of commentary you've written and saved as a .txt file)
What you should see: File names appear in the Knowledge section, and ChatGPT confirms the files have been uploaded.
Part 5: Set Capabilities
In the Capabilities section:
- Enable: Web Browsing (allows your GPT to search for current financial benchmarks when asked)
- Enable: Code Interpreter (allows your GPT to analyze data you paste in and run calculations)
- Disable: Image Generation (not needed for financial work)
Part 6: Test Your GPT
In the preview panel on the right, test with real scenarios:
Test 1: Variance commentary Paste a sample variance table and type: "Draft commentary for each line." Verify the format matches your specification.
Test 2: Executive summary Paste sample results and type: "Write the CFO executive summary." Verify the length, tone, and structure match your format.
Test 3: Clarifying questions Type an ambiguous request: "Write the deck." Verify it asks for clarification rather than guessing.
Refine: If any test produces wrong output, go back to the Instructions and add a clarification or example. Iterate until the output matches what you'd write yourself.
Part 7: Save and Share
- Click "Save" in the top right
- Set access: "Only me" (private) to keep company financial context secure
- Your GPT is now accessible from the left sidebar in ChatGPT under "My GPTs"
Real Example: A Month-End Session
What Sarah does now (with her Custom GPT):
Day 3 of close week. Sarah opens ChatGPT → My GPTs → "ABC Corp Finance GPT."
She types: "Here's the March variance table. One-time note: the IT variance is favorable because the server migration slipped to April. Draft commentary and the CFO summary."
She pastes her 28-row variance table.
What the GPT produces in 45 seconds:
- 28 variance comments, each under 50 words, in the exact format she specified
- The IT line already correctly identified as a timing item
- A 230-word CFO executive summary with the correct opening sentence structure
- A flag: "The Marketing line shows a $400K unfavorable variance. Should I note this is expected to normalize in Q2, or do you want me to flag it as a concern?"
Sarah edits 4 comments where business context changes the driver, approves the IT flag, and answers the Marketing question. Done in 15 minutes. Total.
What to Do When It Breaks
- "The GPT keeps using a format I don't want" → Add a more explicit example to the Instructions section. Copy-paste a real example from your reports and say "Here is an example of the exact format I want for variance commentary: [paste]"
- "The GPT doesn't know about a key business unit" → Update the Business Units section in Instructions. GPTs don't learn from your conversations, so you must update the Instructions manually.
- "The GPT is too long / too short in its responses" → Add a word limit instruction: "Variance commentary per line must be under 40 words. Executive summaries must be under 200 words."
- "The knowledge files don't seem to be used" → Try explicitly referencing them: "Using the report template I uploaded, draft..."
Variations
- Simpler version: A Claude Project achieves most of the same goals for the monthly close workflow without the Custom GPT setup complexity. Start there if this feels like too much.
- Extended version: Create separate Custom GPTs for different purposes: one for close reporting, one for budget season, one for board communications. Each can have context optimized for that specific workflow.
What to Do Next
- This week: Start with the Instructions draft only (no files) and test with your most recent close data
- This month: Add your knowledge files and refine the instructions based on 3–4 real close cycles
- Advanced: Create a second Custom GPT for budget season with budget-specific context, and test sharing it (with no sensitive files) with your finance team manager
Advanced guide for Financial Analyst (Corporate) professionals. Requires ChatGPT Plus ($20/month). Do not upload confidential or unreleased financial data to any AI service. Use sample or anonymized data for setup and testing.