Summary: Pricing AI products is harder than pricing traditional SaaS because your costs are variable, per-customer, and change every time a model provider updates their rates. This guide walks through a practical framework: start with cost data, build unit economics, choose a pricing model, and iterate with real numbers.
The Problem With Pricing AI Products Today
An automation consultant charges her client $1,500/month for an AI workflow. The workflow's OpenAI API costs run $2,400/month. She's losing $900/month and doesn't know it.
This story, shared recently on LinkedIn, is not unusual. It's the natural result of pricing AI products the same way we price traditional SaaS — pick a number, ship it, forget about it.
Traditional SaaS has near-zero marginal cost per user. Add a customer, the infrastructure cost barely moves. You can price based on value, competitive positioning, or gut feel, and your margins stay roughly the same.
AI products are different. Every API call has a real, variable cost. That cost changes based on the model used, the complexity of the query, the length of the response, and how many retries or tool calls the workflow requires. Two customers on the same plan can have 10-100x cost differences based on how they use the product.
Pricing without cost-to-serve data isn't a strategy. It's a guess.
Step 1: Know Your Costs Before You Set Your Price
This sounds obvious. It isn't.
Most AI companies track aggregate API spend — the monthly OpenAI or Anthropic bill. Fewer track cost per customer. Almost none track cost per feature, per workflow, or per outcome.
The unit of AI cost accounting is not the token. It's the trace — a complete workflow execution from user request to final response. A single customer interaction might involve embedding a query, retrieving context from a vector database, calling a planner model, executing tool calls, generating a response, and running safety checks. That's six cost centers in one interaction. Token math captures maybe two of them.
Before setting any price, answer these questions:
- What does it cost to serve each customer per month? Not on average — per customer, with variance.
- Which features drive the most cost? A chat feature and a document analysis feature have very different cost profiles.
- What does a single workflow execution cost? Not the LLM call alone — the full trace including retrieval, tool calls, and retries.
- How does cost correlate with the value delivered? Some high-cost interactions produce high value. Some don't.
If you can answer these, you can price with confidence. If you can't, any pricing framework is built on assumptions that may be wrong by an order of magnitude.
Step 2: Build Your Unit Economics
With cost data in hand, the core financial model becomes straightforward.
Gross margin per customer = Revenue from customer - Variable cost to serve that customer
For AI products, the variable cost includes:
- LLM API costs (input and output tokens across all models used)
- Embedding and retrieval infrastructure
- Tool call and external API costs
- Compute for any self-hosted models
- A proportional share of infrastructure (vector databases, caching layers)
A healthy AI product targets 60-80% gross margins. Below 50%, pricing or cost structure needs attention. The challenge is that these margins vary dramatically by customer — your lightest user might have 95% margins while your heaviest user is margin-negative.
The ratios that matter:
| Metric | Target | Why |
|---|---|---|
| Gross margin | 60-80% | Covers fixed costs and profit |
| LTV/CAC | 3x minimum | Validates acquisition economics |
| Payback period | Under 12 months | Runway sustainability |
| Cost variance across customers | Know it, even if it's high | Pricing tiers should reflect reality |
A practical sanity check: if a customer pays $100/month, costs $10/month to serve, and costs $300 to acquire, payback is 3.3 months and three-year gross LTV is roughly $3,240. LTV/CAC of 10.8x. That's healthy. But if your heaviest users cost $80/month to serve on the same $100 plan, the math changes entirely.
Step 3: Choose a Pricing Model That Fits Your Product
There are two fundamental approaches to pricing AI products, and the right choice depends on how specific your use case is.
Fuel Pricing: Charge for Inputs
Charge based on what gets consumed — tokens, API calls, compute time, storage.
This works when:
- You're a horizontal platform serving diverse use cases
- You don't control what customers build on top of your product
- Cost scales predictably with usage
- Your customers are technical and understand unit-level pricing
OpenAI and Anthropic use fuel pricing because they serve an entire market. They can't predict what will be built on their APIs, so per-token pricing is the natural model.
The risk: Fuel pricing can create the "pain of paying" problem. When Clay introduced per-action pricing, users during onboarding chose to enrich 10 emails instead of 1,000 — not because 10 was enough, but because they were nervous about spending tokens. The onboarding trained users to be conservative instead of discovering value. If your users are doing math instead of experiencing your product, fuel pricing may be working against adoption.
Solution Pricing: Charge for Outcomes
Charge based on what gets delivered — resolved tickets, completed reports, generated documents, successful automations.
This works when:
- You solve a specific, well-defined problem
- The outcome is measurable and valuable
- You can absorb cost variance between outcomes
- Your customers are non-technical and care about results, not inputs
Intercom moved to charging per resolved conversation rather than per message. The pricing matches what customers actually buy — fewer support tickets, not fewer tokens.
The risk: Cost variance per outcome can be enormous. Some tickets take 500 tokens to resolve. Some take 100,000. Solution pricing means absorbing that variance, which requires deep understanding of your cost distribution.
Hybrid: Tiered Plans With Usage Limits
Most AI products land here. A subscription base with usage-based components.
Three tiers work for most products:
- Starter: Low price, limited usage (e.g., 1,000 AI interactions/month). Low friction entry.
- Growth: Higher limits, more features. This should be the obvious choice for most customers.
- Enterprise: Custom limits, SLAs, dedicated support. Priced on negotiation.
The key insight: hide pricing complexity for 90% of users. Let heavy users buy additional credit packages after hitting their plan limit. Don't make everyone a meticulous tracker with a gas meter.
Design upgrade pinch points that are obvious and value-aligned. Users should hit the limit right as they're getting real value — that's when upgrading feels like a natural next step, not a punishment.
Step 4: Price to Value, Not to Cost
Cost data is the floor. Value is the target.
If your AI replaces a $50,000/year employee, a $10,000 annual price is a strong value proposition regardless of what it costs you to deliver. If you prevent a $100,000 compliance violation, $10,000 is rational risk reduction.
A useful rule of thumb: aim for 5-10x ROI for the customer. At 10x, the purchase decision is obvious. Below 3x, it becomes a negotiation.
Customers don't compare your price to your cost. They compare it to the next best alternative:
- Hiring someone to do the work manually
- Using a competitor's product
- Building it themselves
- Doing nothing (the status quo has a cost too)
Anchor your pricing to these alternatives. If you replace three $30,000/year support agents, $30,000-$50,000/year is credible. If you're priced at $5,000, you're either signaling low quality or leaving significant revenue on the table.
Price signals quality. In B2B, underpricing can be as damaging as overpricing. A $29/month AI product competing against $500/month solutions raises suspicion, not excitement.
Step 5: Handle Free Tiers Carefully
Free trials and freemium reduce the first purchase decision to zero. For AI products, this comes with a specific risk: you're paying real, per-interaction costs for users who haven't committed to paying.
Free Trials
Full access for 14-30 days. Creates urgency. Works best when users experience value quickly — within minutes or hours.
One consideration: users who sign up for a trial and forget to use it don't convert, and they often feel frustrated when the window closes. The trial clock should start when the user takes their first meaningful action, not when they create an account.
Freemium
Limited features, available indefinitely. Works when usage reveals value over time (days or weeks) and cost-to-serve on the free tier stays low.
For AI products, "low cost to serve" is the hard part. Set clear boundaries:
- Limit the number of AI interactions per month
- Restrict which models or features are available
- Cap compute time or output length
- Consider a waitlist to control free-tier size and costs
The Math
Monitor two numbers: cost to serve free users and free-to-paid conversion rate. If you're spending $5/month per free user with a 2% monthly conversion rate, you're spending $250 in free-tier costs per converted customer. That's your effective CAC from the free tier. Is it better or worse than your other acquisition channels?
Step 6: Iterate With Data, Not Intuition
Pricing is not a launch decision. It's a continuous process.
How Often to Revisit
In the first year, revisit pricing quarterly at minimum. Align changes with new capabilities or customer segments. As the product matures, semi-annual reviews are sufficient unless the market shifts significantly.
What to Measure
Track these signals continuously:
Signs the price is too low:
- Customers accept without negotiation
- Sales cycles are unusually fast
- Buyers describe it as "a steal" or "no-brainer"
- High churn from low engagement — users don't value what they didn't invest in
Signs the price is too high:
- Frequent discount requests
- High abandonment at the payment step
- Long sales cycles stuck at procurement approval
- Win rates declining despite strong product-market fit
How to Adjust
- Raise prices for new customers first. Measure conversion impact before applying broadly.
- Grandfather existing customers when raising prices, or give 90+ days notice with clear communication about added value.
- Test packaging changes (what's included in each tier) before testing price changes. Packaging often has a larger impact than the number.
- Ask customers directly how price compares to value received. The 1-10 price pain scale is a simple diagnostic: 1-3 means underpriced, 4-6 is near optimal, 7-10 suggests overpricing.
The Foundation: You Can't Price What You Can't Measure
Every step in this framework requires one thing: actual cost-to-serve data at the customer level.
Without it, your unit economics are estimates. Your tier boundaries are guesses. Your margin calculations are fiction. You can't tell if your free tier is a growth engine or a cash furnace. You can't tell if your biggest customer is your most profitable or your most expensive.
AI pricing is getting more complex, not simpler. Companies are layering subscriptions, credits, per-action charges, and model-specific pricing on top of each other. The companies that navigate this well will be the ones that see their cost structure clearly — per customer, per feature, per interaction — and price accordingly.
The teams that track per-customer costs early have options. The teams that wait have constraints.
Bear Lumen provides per-customer cost-to-serve data for AI products — the foundation for every pricing decision in this guide. See how it works.