Usage Based Pricing Models for Gen AI

April 23, 2024

An Intellyx Brain Blog for Amberflo

Everyone’s rushing to deliver generative AI based applications and products to market and cash in on the latest trend or gain competitive advantage.

At the same time it’s not clear that everyone understands the right way to set prices to ensure a return on investment.

If you are rolling out a generative AI-based application, incorporating AI into an existing application, or creating a new product that incorporates AI, taking a few minutes to develop the right pricing model will be an essential part of determining your success.

The pricing model for AI is important not only for the revenue it brings, but also because it has to be something customers easily understand. Finally, if your pricing model isn’t competitive, your customers will go looking for alternatives. This is a tricky balance to achieve.

Setting up usage based AI pricing

First, configure usage metering for the gen AI tool you are using, as explained in the previous blog in the series. For an LLM-based tool, this means capturing the number of words sent in the input prompt and the number of words received in the output response, and calculating the backend price charged by the model provider according to the per-token rates.

Popular generative AI tools such as Chat GPT, Anthropic, Google Gemini, Cohere, and Mistral charge on the aggregate token (i.e. word) level over a period of time, with varying price tiers depending on the quality of service you select or on an API call rate limit.

So you first need to analyze and understand the pricing model for the gen AI tool you’re using, establish the metering, take into account any variations imposed by the particular AI tool, and structure your pricing model accordingly.

It’s also important to figure out the right margin to add, if any, to cover your costs and generate revenue.

If the application is internal to your organization, there’s probably no need to include margin – especially if you are just passing along or dividing up the cost across different departments.

However if you are reselling the service and adding value to it, such as security scanning for the prompts and responses or training the LLMs for industry specific needs, it is standard practice to to add a reasonable margin to the gen AI usage fee to account for the additional value-add and generate a reasonable profit.

Billing should be transparent, understandable, and flexible enough to adapt to the customer’s current payment processes, if possible…