Many companies are adopting some form of credit as part of their pricing model. This is especially true for B2B AI and agenticAI companies. This can take the form of tokens, or credits, or some arbitrary unit constructed by the company, but it is emerging as one of the key pricing models for business AI and AI agents. Why is this?

Credit pricing is the best way to price general purpose technologies early in the adoption cycle.

General purpose technologies (GPT) are technologies that can be applied to a broad array of problems and that have many uses. Early on, many of these uses have not even been imagined.

Examples of general purpose technologies range from writing to steam power to electricity. Databases and virtualization are also general purpose technologies. Today one of the most compelling general purpose technologies is deep learning and more generally artificial intelligence.

The challenge in pricing general purpose technologies is that there are many different use cases, all with different value drivers and that many of the use cases have not been discovered. Pricing needs to be flexible and careful not to discourage the emergence of new uses. Credit based pricing is one way to achieve this.

Eventually general purpose technologies become commodities, conventional economics kicks in, and the prices converge to the marginal cost of producing an extra unit. The marginal cost of building and accessing foundation models has been coming down quickly. (Remember the DeepSeek kerfuffle back in January.) Eventually they will become a commodity.

But they are not a commodity today. Foundation models are still in the early stages of development and adoption. Credit (or in this case token) based pricing makes sense and we can expect different companies and services to continue to take advantage of credit based pricing as it provides maximum flexibility. Current projections suggest that foundation models will be commoditized by around 2030.

Credit pricing is in the interest of the buyer

Buyers can benefit from credit pricing. It allows for maximum flexibility. Many buyers are not clear on how much value AI will provide or how AI will provide that value (the exception is certain AI agents like Intercom’s Fin AI where the outcomes are specified and predictable).

Of course flexibility has its own costs, and one of the drawbacks to credit pricing models is lack of predictability. This can be managed through clever pricing models like the approach taken by Box.com.

Box AI Pricing

Early on, Box had a sophisticated approach to AI pricing. Individuals were provided with 20 credits to use for AI functionality like content creation or document queries. There was an additional pool of 2000 credits that could be applied by any user. Of course credits could be topped up.

As adoption picked up, Box simplified its approach by creating an Enterprise Plus package that included AI.

Box has recently introduced Box AI Units, a metric that tracks and manages AI API usage. The transparency can help companies to scale and optimize AI spend without unpredictable costs. It will also make it easier for companies to adopt the Box AI Studio that was announced at the same time. See Content + AI: Transforming intelligent workflows and user experiences, February 19, 2025.

Credit pricing is in the interest of the vendor

Credit based pricing, where buyers prepay for credits that they can use at their discretion, is also a big advantage for vendors.

• Revenue is moved forward, from the time of use to the time of purchase

• Revenue can be locked in, helping with predictability (both sides value predictability)

• Usage incentives and rewards can be designed to encourage use and explore what buyers really value

• Flexible uses are enabled, making it easy for buyers to experiment and use the functionality where needed

Key Benefits of Well-Implemented Token/Credit Models

• Flexibility: Customers can scale usage up or down without renegotiating contracts

• Value Alignment: Pricing is tied to measurable value metrics rather than arbitrary feature bundles

• Simplified Pricing: Complex service offerings can be bundled under a single "currency"

• Customer Control: Users have more agency over their spending and resource allocation

• Revenue Optimization: Companies can capture value from both casual and power users

Best Practices in Designing Token/Credit Systems

• Map tokens to value metrics: Tie credits to customer-perceived value rather than backend costs
  (Credits should be based on variables used in the value model)

• Implement usage transparency: Provide real-time dashboards for credit consumption

• Offer hybrid pricing tiers: Combine credits with subscriptions for baseline access

• Design for ecosystem participation: Allow token transfers or staking in partner-led models

• Anticipate token liquidity: Consider potential secondary markets, especially in blockchain-based systems

When to use credit pricing

Use credit pricing when

• You are pricing a General Purpose Technology in early stages of adoption

• There are many use cases, too many to price independently (you can capture the value of different use cases by how you assign credits - how may credits are needed for each use case)

• The technology is approaching commodification

In most cases, credit pricing will be part of a hybrid pricing strategy and will be one of two or three pricing metrics. Credits can be a very effective way of capturing more value metrics than can easily be captured through a simple pricing metric like users or some one dimensional measure of usage.