Pricing patterns for generative AI

Steven Forth is a Managing Partner at Ibbaka. Connect on LinkedIn

Most companies monetizing generative AI are using multiple pricing metrics. In the survey, we provided a standard list of pricing metrics (enhanced to include some of the emerging metrics we are seeing for generative AI). Here are the response options and the raw data from the morning of September 10 (N=216).

• Token based (a combination of input and output tokens) 35%

• User based 41%

• Model based (complexity and number of models accessed) 29%

• Speed based (latency of response) 15%

• Value metric based 30%

• Outcome based (based on results delivered) 20%

• Connection based (the number of connections between nodes) 6%

• Process based (number and complexity of processes supported) 23%

• Specific to the domain and how value is created 25%

• Managed services to build and operate the solution 23%

• Configuration, training, and set up fees 10%

• Other (please specify) 4%

It is always interesting to see what comes up under “Other.”

• Threats intercepted

• Customers served through the AI

• Customers supported

• Server based

• Consumption based, difficult to measure outcome

• Who cares (we usually get some cynics responding to these surveys)

It is interesting that User Based is the most common pricing metric even for this advanced set of applications. User-based pricing is deeply encoded in the SaaS mindset, frequently to its detriment as the number of users does not generally track the value being delivered.

The second most common pricing metric is Token Based. This is generally a cost-based metric and is used by most of the large foundation model companies in some part of their pricing (Open.ai, Anthropic, Mistral, Google). In terms of actual real-world pricing for access to models, this may be the most common metric as these models underpin most other applications.

Most companies are using more than one metric. How do these metrics cluster? What metrics are most likely to be used together?

We used Ibbaka’s proprietary clustering analysis software Geode to answer this question. We wanted to see if there are clusters of pricing metrics that tend to be used together. We did this by applying the Leiden algorithm to the answers to the pricing metric question to see if there are clusters and what these clusters are. We then looked for other patterns related to these clusters.

Geode found three clusters: User Based; AI Cost Based (Tokens, Models and Speed); Domain Based.

Ibbaka Geode Screenshot

One interesting thing to note. User Based was the highest single metric but the User Based cluster was the smallest group! This is why one needs to look at the clusters and not just the individual metrics. The largest cluster was Domain Based, even though no single metric was the highest. Interesting.

User Based 26.76%

The only metric that correlated with Users was Value (which in SaaS pricing is often taken to mean a Usage metric).

People in this cluster tended to see generative AI as a modest or incremental improvement and not as something transformative. This suggests they see generative AI as more of a sustaining innovation, and with sustaining innovations pricing metrics tend not to change.

The capabilities of generative AI are expected to grow linearly or plateau and not to grow exponentially (also signaling sustaining innovation).

The most common approach to pricing was to base price on Willingness to Pay (WTP) followed by Value Based. Cost Plus is not common in this cluster.

AI Cost Based 36.36%

The metrics in this cluster here are the metrics used by the big foundation model companies: Tokens, Models, and Processing Speed. Managed services also play a role here.

This group sees generative AI as transformative, disruptive innovation. The capabilities of generative AI are expected to grow exponentially.

Pricing is cost based. There is an argument that if you don’t understand how an application will be used and there are significant operating costs (which is how some companies see generative AI) then cost based pricing is the simplest way to price for a wide range of different use cases. As generative AI is still in the exploration stage for many applications cost plus is the best option.

This will change as the use cases get more explicit and value comes into focus - see the next cluster.

Domain Based 36.86%

At this point, this is the largest cluster (by a hair). These are the companies that Bessemer Venture Partners is talking about in The Future of AI is Vertical.

Here pricing is domain specific and process based. A surprising number of people in this cluster called out Outcomes Based pricing. This is what Intercom and Zendesk are doing when they charge per customer interaction successfully resolved with their customer service bots.

Similar to the previous cluster, respondents in this cluster also see generative AI as transformative and expect its capabilities to grow exponentially.

The dominant pricing process here is Value Based Pricing.

And what about Generative Pricing?

Ibbaka is developing an approach to packaging and pricing second-generation generative AI applications that we call generative pricing. Begin to read about generative pricing here.

Interest in this approach is growing and 20 respondents said that generative AI should be priced with generative pricing.

All twenty said that generative AI is transformative and sixteen said that it will fundamentally change how B2B software is developed and priced.

Importantly, this is the group that is most committed to outcome based pricing.

We are still early on the path to generative pricing, but these are encouraging signs.

Book a demo today and discover how our unique B2B SaaS solution can increase sales, improve retention, and drive growth.