How Should AI Applications Charge? Per-User Pricing Is the Biggest Trap in the AI Era
Deep thoughts on AI and aspirations —— ByteDance Deep Thinking Circle
How to price AI products is where most startup teams take the easy way out. They see competitors charging monthly fees per user and copy the same model. This “copying” often leads straight into a trap—because per-user pricing assumes “the more people, the more you earn.” But the value of AI products is precisely to need fewer and fewer people.
First, understand one thing: Pricing models aren’t just pricing strategies—they determine what you’re actually selling.
Three Pricing Models, Three Businesses
There are essentially three types of pricing models, each corresponding to completely different business logic.
Per user (per seat), you’re selling “seats.” The more people, the more you charge—assuming more people use it. This is the traditional SaaS approach.
Per usage, you’re selling “workload.” Charge based on consumption—by token, by call, by resource usage. This is the infrastructure layer approach.
Per outcome, you’re selling “results.” Charge only when successful, nothing if it fails. This is the most extreme and powerful model in business.
No model is absolutely better or worse—the key is fit. But the AI era has a major problem: the most comfortable path, per-user pricing, is being blocked by AI itself.
Why Per-User Pricing Is a Trap in the AI Era
Per-user pricing logic is built on “the product is for people to use.” But the selling point of AI products is precisely “replacing people.”
If an AI customer service agent truly replaces three human agents, charging per user means you only collect one fee while delivering the value of three people. The more customers save on headcount, the less you earn—the greater the value your product creates, the less money you collect. This is a direct conflict between pricing logic and product logic.
This is why a wave of forward-thinking AI companies would rather take risks with usage-based or outcome-based pricing: they know that per-user pricing caps their own value.
Outcome-Based Pricing Is the Strongest but Hardest
The most extreme model is outcome-based: charge only when your product solves the problem, nothing if it doesn’t.
Customers love this model—no risk, tied to results. Suppliers like it too—once successful, both deal size and trust are extremely high.
But it’s the hardest, for three reasons: you must define “success” (what counts as solved?); you must measure “success” (where does the data come from?); you must attribute “success” (is it your contribution, or customer luck?). Without all three, outcome-based pricing won’t work.
A counterexample illustrates the standard: some companies handle credit card chargeback disputes, counting only successful recoveries, with bank settlements providing the most objective “success” evidence—so they dare to charge entirely based on outcomes. If your “success” relies only on customer subjective judgment, start with a hybrid model.
Apply This to Your Own Product
When pricing your product, don’t default to per-user pricing. Think through this sequence:
First ask: does my product “augment people” or “replace people”? Augmentation can use per-user pricing; replacement should use per-usage or per-outcome—not per-head.
Then ask: can my “success” be defined, measured, and attributed? If yes, move toward outcome-based pricing; if not, transition with usage-based pricing first, then gradually approach outcomes.
Finally ask: do I dare tie pricing to customer success? If yes, your product will appear more valuable; if not, it means you’re not confident enough in your product yet.
Pricing isn’t filling out a price sheet—it’s redefining your product’s value. Answering these three questions is far more useful than copying a competitor’s price list.
Key points: Pricing models determine what you’re selling (seats/workload/outcomes); per-user pricing gets undermined by AI’s logic of replacing labor; outcome-based is strongest but hardest (define/measure/attribute success); first ask augment or replace, then decide pricing model.