For a long time, I thought pricing was something I could figure out after launch.
Build the product, get people using it, then decide what to charge. HairOver changed that. The moment I looked closely at what each generated image cost me, pricing stopped being a later problem. It was deciding what the product could even be.
What this means
A business model answers four questions:
- Who pays? Not always the user. Sometimes it's an advertiser, an employer, or a parent.
- What are they paying for? The outcome they get, not your feature list.
- How often? Once, every month, or every time they use it.
- What's left over? Revenue minus the cost of delivering the product. That's gross margin.
If you can answer those four simply, you have a business model. If you can't, you have a product people may or may not pay for.
The main options, simply
One-time purchase. Pay once, own it. Simple to understand. Hard to grow, because every sale is a new sale.
Subscription. Pay every week, month, or year. Good when people get ongoing value. Every customer you keep adds to next month's revenue, which is what MRR (monthly recurring revenue) measures.
Free / freemium. Free to start, pay for more. Works if enough free users become paying users and free users don't cost much to serve.
Credits / pay-as-you-go. Buy a bundle, spend it as you use the product. Fits products where each use has a real cost, which describes many AI products.
None of these is "best." The right one matches how people actually use your product and what it costs you to serve them.
Real example: HairOver's pricing, as an experiment
HairOver is a good case study because it has a real cost every time someone uses it. When I did the analysis, each generated image cost about NPR 15 (roughly $0.10).
The first model was credits only: 5 💎 per generation, a $7 pack for 25 generations, a $55 pack for 200, and no subscription.
When I worked through the math, I found two problems. I'd left out the app store commission, which dropped the main pack's margin from 64% to about 58%. And the $55 pack was only about 2% cheaper per generation, which is no real deal at all.
The rebuilt plan mixes subscriptions (weekly, monthly, yearly) with credit packs for heavy users. It also has a rule I now use everywhere: every tier must stay profitable even if the customer uses 100% of what they paid for.
I want to be clear: this isn't the "correct" model. It's the current experiment. The real answer comes from what users actually do. The full breakdown is in HairOver Pricing: What NPR 15 Per Image Does to Unit Economics.
Gross margin: the number people skip
Gross margin is what's left from each sale after the direct cost of delivering it.
For most software, that cost is tiny, so margins are high. For AI products, it isn't. Every generation, every model call, every image costs money. If you price an AI product like normal software, heavy users can cost you more than they pay.
That's why HairOver's yearly plan is capped at about 25 generations a month. It isn't stinginess. At NPR 15 per image, that's roughly the most a ~$60/year plan can support while keeping a healthy margin.
Revenue vs downloads
This deserves its own line: downloads aren't revenue.
A free app can have lots of installs and no business. Downloads show people were curious. Revenue shows they got something they value enough to pay for. They're different stages, and the gap between them is where most consumer apps struggle. Chapter 10, Metrics That Matter, goes deeper on this.
CAC and LTV, without the jargon
CAC (customer acquisition cost): what you spend, on average, to get one paying customer.
LTV (lifetime value): how much gross profit that customer brings you over the whole time they stay.
The rule is simple: LTV needs to be comfortably bigger than CAC. If it costs $20 to get a customer who brings you $15 before leaving, every new customer loses you money. (Those numbers are only an example.)
Early on, you often won't know either number. That's fine. What matters is knowing they exist, so you don't scale something that loses money on every customer.
What to do
- Answer the four questions (who, what, how often, what's left) in one sentence each.
- Work out your cost per use. Include platform fees, API costs, and payment fees.
- Check your worst case. What happens if a customer uses everything they paid for?
- Pick a model that matches usage. Occasional use suits one-time or credits. Ongoing use suits subscription.
- Treat your first price as an experiment, with a date to review it.
Mistakes to avoid
Pricing later. It shapes the product, especially when each use costs you money.
Forgetting the platform's cut. I did this. App stores, payment processors, and marketplaces all take a share.
Big packs that aren't better deals. If the bigger option isn't clearly cheaper per unit, nobody sensible buys it.
Copying another app's pricing. Their costs and users aren't yours.
Calling your current price "correct." It's your current guess.
My current thinking
The biggest lesson from HairOver's pricing wasn't about price at all. It was about cost. Reducing the cost per image would help more than any pricing change, because it would let every plan be more generous without charging more.
I don't know yet which plan people will choose, or how much they'll really use. I'll find out from real behavior and share it in the HairOver guide.
Key takeaway
A business model is who pays, for what, how often, and what's left after costs. Match the model to how people actually use your product, check your worst-case margin, and treat your first price as an experiment. For AI products especially, cost per use shapes everything.
Next: Finding Your First Customers.