When to use it
When you need to know whether the business model works: does a customer pay back, or are you pouring money into a hole. The role is a startup financial analyst. Result: CAC, ARPU, LTV, LTV/CAC, payback — step by step, with formulas and a verdict of "works / doesn't".
The prompt (copy and paste)
You are a startup financial analyst. Calculate the unit economics from the inputs and explain every step.
INPUTS: price/average order value = <...>; margin = <...%>; marketing spend for the period = <...>; customers acquired in the period = <...>; average customer lifetime or monthly churn = <...>; repeat purchases/subscription = <...>.
Calculate step by step, showing the formulas: CAC; ARPU; LTV; LTV/CAC; CAC payback period (months).
Compare against the health benchmarks (LTV/CAC >= 3; payback < 12 months) and say whether the economics add up.
Name 2-3 levers that would improve the picture the most, and what data was missing for a precise answer.
If any input is missing, ask for it instead of guessing.
Filled-in example
Inputs: 2,000 ₽/month per customer; 80% margin; 300,000 ₽/month on marketing; 150 customers acquired; 5% monthly churn; monthly subscription.
Expected AI response: CAC = 300000/150 = 2,000 ₽; ARPU = 2000 x 0.8 = 1,600 ₽/month of margin; lifetime = 1/0.05 = 20 months; LTV = 1600 x 20 = 32,000 ₽; LTV/CAC = 16 (>3 — excellent); payback = 2000/1600 ≈ 1.25 months (<12 — excellent). Conclusion: the economics add up with plenty of headroom. Levers: you can push acquisition harder; what was missing — cohort and seasonality data.
Variations
- Scenarios. "Run it at 5/10/15% churn — show me where the economics break."
- Break-even. "How many customers do I need to break even given fixed costs?"
- Lever impact. "If I raise the price by 20%, how do LTV/CAC and payback change?"
Pro tips
- Garbage in, garbage out. The AI will honestly compute whatever numbers you give it, but it won't sanity-check them — the inputs are on you.
- Double-check the arithmetic: LLMs do sometimes get multiplication wrong. Asking for the formulas (CoT) is exactly how you catch that.
- This is about money — don't paste your company's real financials into a public AI; use rounded/dummy figures or a local model.