Does this feature make money?
Lesson 11 of 24, level 01Index
The AI assistant costs about nine rupees per user per month in model calls. The subscription is one hundred and ninety nine. Finance wants a number, not an opinion.
Four numbers, and everything else is decoration
The customer must repay their cost
- Revenue: ARPU × paid months
- Service: COGS × active months
- Acquisition: CAC paid up front
CAC is what it costs to get a customer. ARPU is what they pay you. COGS is what serving them costs. Churn is how long they stay. Lifetime value is roughly ARPU minus COGS, divided by churn.
The one executives ask about first is payback period: how many months until a customer has repaid what you spent acquiring them. Under twelve is healthy for most subscription businesses.
AI features break the old assumption. Classic software had near zero marginal cost. A feature that delights users and loses forty rupees a month per user kills the company slowly.
Unit economics gives you a small, inspectable model of how serving one customer works. In this exercise the numbers are fictional; the aim is to make assumptions visible, not to value a business.
Use your chosen course project throughout. The additional examples below are fictional practice cases; transfer the method to your own evidence.
Choose a consistent unit
Use one customer per month, one order, or one completed job. Keep revenue and variable costs on the same basis. Do not compare annual revenue with monthly support cost. Record what is excluded, especially fixed salaries and overhead, so contribution is not mistaken for company profit.
Calculate before narrating
Contribution per unit is revenue minus the variable costs included in your model. A simple acquisition payback estimate divides acquisition cost by positive monthly contribution. It assumes that contribution continues and ignores timing complications; it is not a retention model. If contribution is zero or negative, this calculation does not produce a meaningful payback period.
Test the fragile assumption
Change one input at a time and see which changes the decision. An AI feature may have variable inference costs that grow with usage. Include support, retries, and failure handling when those are material. A single average can hide a small group of very expensive users, so inspect a heavy-usage scenario too.
A small model with visible assumptions
- Revenue: Per customer per month
- Variable cost: Same unit and period
- Contribution: Revenue less cost
- Payback: CAC / contribution
What does one customer contribute?
Interactive teaching example · All numbers and tickets are fictional.
Payback = $90 acquisition cost ÷ monthly contribution, only when contribution is positive. Fixed costs, churn, taxes, and cash timing are excluded from this teaching model.
How Dropbox handled it
Dropbox
Dropbox was paying roughly two to three hundred dollars per customer through paid search, for a product that cost ninety nine dollars a year.
They replaced most of that spend with a referral programme that gave both sides free storage, a cost measured in cents of marginal infrastructure rather than dollars of ad spend. Signups rose 60 percent permanently.
The insight was not that referrals are good. It was that their marginal cost of delivery was near zero, so paying in product was almost free while paying in cash was ruinous. The same tactic is brilliant at one company and fatal at another.
Doing it with AI, and where it breaks
Build the model in a spreadsheet, then have a model stress test it: which assumption here, if wrong by 20 percent, changes the conclusion?
Never let a chat window do the arithmetic. Make it write the formulas and compute them yourself in the sheet.
Your AI workbench
Start with your own notes or clearly labelled practice data. Remove private details before sharing. Replace the placeholders, run the prompt in your chosen AI tool, and keep the output beside its source.
Create a transparent spreadsheet model with revenue $30/month, variable cost [value]/month, and acquisition cost $90. Show formulas, units, exclusions, and the zero/negative contribution case. Add a sensitivity table, not an invented forecast. Explain which input needs real evidence before a product decision.
Before you use the output
- All inputs use the same period
- Zero or negative contribution is handled
- Excluded costs and retention assumptions are stated
Stuck? Try this next
Ask AI for formulas rather than a screenshot of a spreadsheet. Change an input yourself and verify that every dependent cell updates as expected.
Keep a brief AI log: input used, useful output, what you checked, and what you rejected. The decision remains yours.
Build it
A unit economics sheet with a go or no go recommendation.
Google Sheets
Checkpoint
If you did the build, these take two minutes. If you cannot answer one of them, that is the part to go back to.
Moving on marks this lesson complete. Finish the build first, it is the part that counts.