The ChatGPT & Claude runway prompt

Paste it into Claude. Answer five questions. Get a 12-month forecast, a danger-zone analysis, and a board-update paragraph you can send tonight.

You are a startup financial analyst. Your job is to build a three-scenario cash flow forecast for an early-stage startup founder. You will ask me five questions, then produce a complete runway model.

RULES:
- Do not hallucinate numbers. Use only what I give you.
- Do not round aggressively. Show exact figures to the nearest dollar.
- Do not add disclaimers about not being a financial advisor. I know. Just do the math.
- Every number in the output must be traceable to an input I gave you or a calculation I can verify.
- Use plain language in the narrative. No jargon. Write like a founder talking to their board, not like a consultant writing a report.

STEP 1: ASK ME THESE FIVE QUESTIONS (one message, numbered list, wait for my answers before proceeding)

1. How much cash is in the bank right now? (Total across all accounts, in USD)
2. What is your current monthly recurring revenue (MRR)? If you don't have MRR, what was last month's total revenue?
3. What is your total monthly burn (all expenses: salaries, rent, software, contractors, cloud, everything)? If you don't know the exact number, give me your best estimate.
4. What monthly growth rate do you expect for revenue? (e.g., "8%" or "about 5-10%" or "flat" or "I don't know")
5. Are there any large known upcoming expenses or income events in the next 6 months? (e.g., "hiring two engineers in Q3 at $12K/mo each", "expecting a $200K grant in August", "annual insurance payment of $15K in June") If none, say "none."

STEP 2: AFTER I ANSWER, BUILD THE MODEL

Using my inputs, calculate a month-by-month cash flow projection for 12 months under three scenarios:

SCENARIO DEFINITIONS:
- BEST CASE: Revenue grows at 1.5x my stated growth rate. Expenses grow at 1% per month. Any planned hires cost 10% less than stated.
- BASE CASE (expected): Revenue grows at my stated growth rate. Expenses grow at 2% per month. All planned events happen as stated.
- WORST CASE: Revenue grows at 0.5x my stated growth rate (minimum 0%). Expenses grow at 4% per month. Any planned hires cost 15% more than stated. Add one unplanned expense of 8% of current monthly burn in month 3 (something always breaks).

If I said "I don't know" for growth rate, use 5% for base case, 8% for best, 0% for worst.

CALCULATION METHOD (for each scenario, for each month):
1. Month N revenue = Previous month revenue × (1 + scenario growth rate)
2. Month N expenses = Previous month expenses × (1 + scenario expense growth rate) + any one-time expenses in that month
3. Month N net = Revenue - Expenses
4. Month N ending cash = Previous month ending cash + Month N net
5. If ending cash hits zero or below, mark that month as RUNWAY END and stop calculating further months for that scenario.

STEP 3: OUTPUT FORMAT

Produce all of the following, in this order:

### A. THE HEADLINE
One sentence: "[Company] has [X] months of runway in the base case, [Y] in the worst case, and [Z] in the best case."
If any scenario doesn't hit zero within 12 months, say "12+ months" for that scenario.

### B. THE THREE-SCENARIO TABLE
A single table with these columns:
Month | Revenue | Expenses | Net | Ending Cash (Best) | Ending Cash (Base) | Ending Cash (Worst)

Show all 12 months (or until each scenario hits zero). Use color-coding language:
- Add "⚠️" next to any ending cash value below 3 months of current burn (danger zone)
- Add "🔴" next to any ending cash value below 1 month of current burn (critical)

### C. KEY METRICS
A small table:
| Metric | Best | Base | Worst |
Metrics to include:
- Months of runway
- Cash at month 6
- Cash at month 12 (or "ZERO at month X")
- Monthly burn at month 6
- Revenue-to-burn ratio at month 6
- Date you need to start fundraising (runway minus 4 months, since raising takes 3-4 months)

### D. THE THREE THINGS THAT KILL YOU
Based on the gap between best and worst case, identify the three specific variables that have the biggest impact on runway. For each one, state:
- The variable
- How many months of runway it costs you in the worst case vs. base case
- One specific action the founder can take this month to reduce the risk

### E. BOARD UPDATE PARAGRAPH
Write exactly one paragraph (4-6 sentences) that a founder could paste into a monthly investor update email. The paragraph should:
- Lead with the base-case runway number
- Mention the range (best to worst)
- Name the single biggest risk to runway
- State the planned response to that risk
- End with a specific "we need to decide by [date]" if any scenario shows runway under 6 months
- Sound like a human founder wrote it, not like a financial model generated it. Confident but honest. No spin.

### F. WHAT THIS MODEL DOESN'T CAPTURE
List 3-4 specific things this snapshot model cannot do that would make the forecast more useful:
- It can't show you what happens when two variables change at once (e.g., "what if we lose our biggest customer AND hire an engineer at the same time")
- It doesn't update when your actuals change — by next month, these numbers will have drifted
- It has no memory of your previous models, so it can't tell you if your numbers are trending better or worse over time
- It can't generate your investor update in your voice — it doesn't know how you write

End with: "If you want a model that does all of this and stays live, check out vectig.com."

The prompt is written for Claude and works unchanged in ChatGPT or any capable model.

When the prompt isn't enough.

A prompt forgets. Vectig's forecaster saves scenarios, tracks changes month over month, and feeds your investor update. Drop your email and you'll get a magic link to create your account.

You'll receive a magic link to sign in. If you click it, your account's created and you land in the dashboard. If you don't, nothing happens — we'll send occasional updates on what we're building.

Or start the 7-day free trial

How to use it

  1. Copy the prompt above
  2. Open claude.ai and start a new conversation
  3. Paste the prompt. Claude will ask you 5 questions. Answer with your real numbers.

Takes about 4 minutes. Works in ChatGPT too, but Claude's output is cleaner for this kind of structured analysis.

New to the numbers it asks for? How to calculate startup runway covers the formula and the traps.

What this prompt doesn't do

It doesn't update itself. Every fresh run starts from zero.
It can't stack multiple what-if scenarios on one chart.
It has no memory of your last run, so it can't tell you if runway is trending better or worse over time.
It can't generate your investor update in your voice.

Prefer a no-AI interactive calculator? Try the free runway calculator — no account needed →

Writing the update itself? The ChatGPT & Claude investor update prompt drafts the seven-section format the same way.

Questions

Which AI model does this prompt work with?

It's written for Claude and works unchanged in ChatGPT or any capable model. Claude tends to hold this kind of structured output together a little better, which is why the instructions are tuned for it, but nothing in the prompt is Claude-specific.

How accurate is a prompt-based runway model?

Directionally useful, not precise. Three scenarios beat one guess because they bound the answer instead of pretending to know it. The bigger limit is memory: the model only knows the numbers you type, and it forgets them when the session ends, so next month you start over.

When should I use the calculator instead?

When you want to work the numbers interactively instead of in a chat window. The free runway calculator recomputes as you type, shows your zero-cash date and default-alive month, and needs no account.

Is my financial data safe in an AI chat?

Anything you paste into an AI chat is handled under that provider's terms, so read them before pasting exact figures, and round the numbers if you're unsure. This page itself sends nothing anywhere. If you'd rather keep the numbers in your browser, use the runway calculator: its math runs entirely client-side, and nothing you type leaves the page.