How SaaS Builders Should Price AI Features After Calculating API Cost
When I first added AI features to my SaaS, I had no framework for passing API costs to customers. The bills kept coming, and I had no idea if I was pricing correctly.
So I built a calculator. Then I learned the harder lesson: knowing your cost is only half the battle. You still need to choose the right pricing model.
The Three AI Pricing Models for SaaS
Once you know your API cost per user, there are three ways to structure AI pricing:
Model 1: Per-Message Pricing
The customer pays per AI interaction.
Example: $0.01 per chat message
| Scenario | Calculation | Monthly Cost |
|---|---|---|
| 1,000 users × 50 msgs/month | 50,000 × $0.01 | $500 |
| API cost (10k tokens/user/mo) | 10M tokens × $2.50/1M | $25 |
| Gross margin | $500 - $25 | $475 (95%) |
Model 2: Token Credit Packs
Customer buys credits upfront, deducts as they use.
Example: $20 for 100,000 tokens
| Scenario | Tokens | Pack Cost | API Cost | Margin |
|---|---|---|---|---|
| Light user | 50k/mo | $20 | $0.125 | $19.88 |
| Heavy user | 500k/mo | $20 | $1.25 | $18.75 |
Model 3: Inclusive + Overage
Base plan includes X tokens; pay overage for excess.
Example: $29/mo base includes 5,000 tokens; $0.002/token overage
| Scenario | Base Plan | Token Usage | Overage | Total |
|---|---|---|---|---|
| Light user | $29 | 3,000 | $0 | $29 |
| Heavy user | $29 | 15,000 | $20 | $49 |
A Realistic AI SaaS Pricing Scenario
(Example scenario — numbers are illustrative)
Imagine you're building a writing assistant SaaS:
- 1,500 active users, 40% use AI features
- Average usage: 80,000 tokens/user/month
- Your API cost: $2.50 per 1M tokens
If you choose Model 1 (per-message at $0.005/msg, avg 20 msgs/user/day):
- Monthly revenue: 600 users × 600 msgs × $0.005 = $1,800
- API cost: 600 × 24,000,000 tokens × $2.50/1M = $36
- Margin: 98%
If you choose Model 3 (inclusive+overage, $19/500k tokens):
- 40% heavy users (600) × $49 overage = $294
- 60% light users (900) × $29 base = $261
- Total revenue: $555
- API cost: 15M tokens × $2.50/1M = $37.50
- Margin: 93%
The per-message model gives 3.2× more revenue on roughly the same cost base. That's the difference between a pricing model that funds your AI features and one that bleeds margin.
Which Model Fits Your SaaS?
Choose per-message if:
- Usage is highly variable (chat apps, content generation)
- You want customers to feel the cost directly
- Low friction for occasional use
Choose token credits if:
- You want predictable revenue
- Customers prefer "pay what you use" transparency
- Usage is lumpy (bursty AI tasks)
Choose inclusive+overage if:
- You want simple, predictable plans
- Most users are light; few are heavy
- You're building a freemium tier
The Decision Framework
Before choosing, ask yourself:
- What's my average tokens per user per month? (Use the calculator: https://dev.to/cheapuno/freelance-pricing-calculator)
- What's my target margin on AI features? (Recommended: 80%+)
- How variable is usage across my user base? (If CV > 1, consider credit packs)
5 AI Pricing Mistakes to Avoid
Before you launch, check yourself:
- [ ] No margin floor: Forgot to calculate API cost per user before setting price
- [ ] Flat per-message pricing: Same price for simple and complex queries (heavy users destroy margin)
- [ ] No overage protection: Inclusive plans with no caps invite margin disaster
- [ ] Ignoring usage distribution: A few heavy users can offset many light ones
- [ ] No pricing review clause: API costs change; your pricing should too
My $10 AI Pricing Model Review
If you're building AI features and still unsure which model fits your product, I offer:
$10 AI Pricing Model Review (15 minutes)
I'll review:
- Your current API usage patterns (or help estimate them)
- Which pricing model fits your user behavior
- Your break-even point and target margin
- 3 concrete recommendations to protect your margin
Result: A specific recommendation — per-message, credit packs, or inclusive+overage — with a worked example for your numbers.
PayPal: https://paypal.me/cheapuno
This takes 15 minutes and usually saves more than $10/month in pricing mistakes.
FAQ
Q: What if my users don't use AI features consistently?
A: Per-message or credit packs work best. Inclusive+overage creates billing surprises for users who occasionally use AI.
Q: What if my API costs change?
A: Build a pricing review clause into your terms. Most SaaS tools reserve the right to adjust AI pricing with 30 days notice.
Q: Can I offer multiple models?
A: Yes—but keep it simple. One free/trial model + one paid model is ideal. Three options creates analysis paralysis.
Q: I have 100 users. Is this still relevant?
A: Yes. The model choice matters more than the scale. A wrong model at 100 users becomes a compounding problem at 1,000.
Q: What if I'm already losing money on AI features?
A: That's the most common signal. Stop guessing and get a specific diagnosis. The $10 review gives you a clear number: whether to raise prices, switch model, or cap usage.
The calculator is free. The review is $10. If this helped you think through your AI pricing, consider supporting: https://paypal.me/cheapuno
Disclosure: This post is my own work and may include AI-assisted editing/research where applicable.
If this helped you, you can support my work at: https://paypal.me/cheapuno
No tracking. No affiliate links.
Top comments (0)