As AI adoption accelerates, companies in nearly every industry are rushing to build products that use the latest generative AI models. In a recent Stripe survey, 63% of tech industry respondents said they currently offered an AI product or application. Perhaps more striking, 31% of nontech industry respondents said the same thing—and a further 45% planned to offer AI products in the near future. 

But monetizing these products remains a challenge. The unprecedented revenue growth we’ve seen in the top AI companies on Stripe (we power 78% of the Forbes AI 50) belies the difficulty. Costs scale swiftly with additional AI model use, and value delivered can be hard to define. On top of these unique pressures, companies face more familiar challenges: building predictable revenue streams, growing their customer base, and protecting margins.

What do successful monetization strategies have in common? Though AI pricing structures vary considerably, AI leaders tend to have clear answers to the following questions: 

• What kind of value does my product deliver to customers, and can I charge for it in a way that covers incremental costs?
• How can I translate this charge metric into a pricing model that encourages customer adoption and growth, while also offering predictable revenue? 
• What additional risks does this pricing model create and how can I manage them?

Below, we’ll walk through this decision-making framework and the trade-offs companies encounter, and we’ll highlight examples from AI companies on Stripe that have built successful pricing models.

Aligning value and charge metrics  

The charge metric is an incremental unit used to measure and calculate the ultimate cost for a product. It’s not a concept unique to AI; old cell phone plans charged per minute of talk (and per text), and cloud storage companies still charge per gigabyte. 

But setting a good AI charge metric is uniquely challenging, because it needs to align with how customers derive value and how the product incurs costs—both of which can be difficult to define and predict:

The value metric 

Defining the “value metric” is the key first step to a successful charge metric. Through user research, companies need to understand which outcomes matter most to their customers and how their products contribute to these outcomes.

It’s important to keep “value” distinct from “usage.” For example, Intercom’s support agent, Fin, holds conversations with a customer’s end users when they have questions or complaints. But Fin only truly delivers value when it actually solves that end user’s problem and can resolve the resulting support ticket. 

Cost to deliver

Once a company understands which outcomes are important, it can start working out how much those outcomes typically cost to deliver. Generative AI models are expensive, and it isn’t always easy to predict how much computational expense a given task will incur.

It’s particularly difficult if an outcome is more complex. Using the same example above: when Fin resolves a ticket, it might require one long conversation or three short conversations—and vastly different numbers of calls to underlying AI models.

There are three main categories of charge metrics that AI products tend to use, and each strikes a different balance between alignment on value and alignment on cost: 

• Consumption-based (per API call, per LLM token): This type of charge metric is closely tied to underlying infrastructure cost. This means the cost incurred per action is predictable for the company selling the product—but it can be harder to link to concrete business value for the customer. This kind of metric is most commonly used by AI model companies, and it makes the most sense when the customer wants to have granular control over precisely what they consume. 
• Workflow-based (per completed task): This metric introduces more variability on cost, but it’s somewhat easier to tie to value. Companies charge a set price for the completion of tasks that can require a series of actions that might vary somewhat in resource consumption, such as booking a meeting or analyzing a spreadsheet. These kinds of tasks have clear time-saving value, and customers easily understand what they’re buying. 
• Outcome-based (per successful outcome): This metric tends to have the most variability on cost, but it can be most closely linked to real business results. Intercom ultimately chose an outcome-based model: the company charges for each ticket Fin successfully resolves. Leaders decided that achieving this compelling alignment between Fin’s value metric and charge metric was worth the risk incurred on cost to deliver.

Balancing revenue predictability and customer growth

Once a company decides on a charge metric, it needs to work out the preferred way to bill its customers. The company might decide to simply charge for usage as it’s incurred, but that’s not its only option. 

In our recent survey, 56% of AI company leaders said they used a “hybrid” pricing model, including various combinations of subscription and usage-based fees. This was the most common response, with pure usage-based pricing in second place at 38%. That’s because hybrid models provide more levers for companies to use as they balance two forces that are both important for long-term company growth: 

Revenue predictability

Recurring subscription fees help companies more accurately forecast revenue, which can help them make long-term investments in hiring and research. Some customers, particularly larger enterprises, value a predictable monthly spend for the same reasons. But for other customers, subscription fees introduce friction that might deter adoption and long-term growth.

Customer acquisition and growth 

Companies can lower the barrier to adoption by allocating a larger proportion of revenue to usage-based fees. This approach can benefit early-stage companies with aggressive targets for customer acquisition, or any company with a product-led growth motion. Customers can explore a product with low financial commitment, and they only pay more as their usage grows over time.

Browserbase, a company that manages and hosts the “headless” browsers AI agents use to interact with websites, built its hybrid pricing model to offer predictable revenue while meeting the needs of its largely developer audience. The company offers tiered subscriptions, each including a prebundled amount of usage geared at a different customer persona. Its free, “introductory” tier lets potential customers test the product at no cost, and clearly communicated rates for “overage” usage mean customers can grow before committing to a higher subscription tier. A custom plan is available for enterprise customers. 

Managing remaining risk with guardrails

A well-designed charge metric and a balanced pricing model can both help protect against the cost risks associated with AI products. But risk remains whenever customer usage is variable: it tends to be higher in a pure pay-as-you-go model, but it’s still present in the subscription-with-overages model many AI companies use. Deliberate fraud can be an even more significant problem, as costs might increase rapidly before an attack is detected. 

Additional guardrails on usage-based charges can keep risk to manageable levels. The right guardrails depend on which patterns of use are most likely to be problematic: 

• If well-meaning customers are at risk of spending more than they intend, usage thresholds with well-timed alerts can forestall unexpected bills. 
• If bad actors run up bills they don’t pay, companies can require customers to pay up front for credits that are drawn down over time. 
• If a model usage can spike unpredictably in response to certain kinds of tasks or queries, rate limits and usage caps can keep spend in check while the customer adjusts their request. 

In all cases, proactive communication is key. Customers want to understand how their usage is translating into spend, and they want to be able to react quickly if they’re spending more than intended. 

Looking ahead: Ongoing experimentation

As AI companies grow, they’re unlikely to treat their pricing structure as a solved problem. Successful AI companies keep experimenting. Though Intercom currently charges the same price for every ticket resolved, the company is considering tiering tickets by complexity in the future.

Intercom’s outlook matches what we saw in our recent survey. Among companies that sold AI products and charged for usage—either alone or in the context of a hybrid model—92% said they had subsequently adjusted their pricing. As underlying AI model costs change, market conditions mature, and customer expectations evolve—all at record pace—companies need to stay nimble to keep their pricing well-balanced and conducive to growth. 

Watch our on-demand webinar, “How to get your AI pricing strategy right,” to learn more about pricing AI products. 

Read how Stripe Billing can support your AI pricing model—regardless of which balance of usage-based fees, subscription fees, and guardrails you choose to implement.