It’s not easy pricing a new AI product. Given all the work involved in creating an LLM or AI tool that actually functions well, it’s tempting for companies to charge as much as humanly possible. However, companies that want to actually make a profitable buck need to take customers into account when building a pricing model.
The biggest players in the space offer some idea of what to charge for a do-it-all LLM. Currently, ChatGPT’s pricing ranges from $20 for the Plus version to $200 for the Pro version of the model. Others, like Claude, are sitting at $17 and $100 for Pro and Max versions respectively. Google AI Ultra is available for customers to purchase for $249.99 a month.
But how does the sausage’s price tag get made, especially with more specialized AI tools?
Michael Friedman, SVP and head of finance at GenAI provider ASAPP, said that his company sticks to “codified industry best practice,” which is a three-pronged approach to pricing: customers, competition, and costs.
But the customer part of that framework is getting even more attention these days, both from Friedman and other AI company founders.
“When you price based on value in a construct that helps your customer realize a rapid payback, rapid cash flow, positive return on their investment, and quick time to value, you have alignment there,” Friedman said. “The thing that’s easy to forecast is the thing that you probably have the most consistency on.”
Is the customer always right? Customer research takes place across finance, market research, sales, marketing, and more to see how they’re feeling about the return on investment for other vendors and market pricing at large.
Cosmin Ene, the founder and CEO of AI pricing infrastructure company Supertab, said that to make pricing appealing to customers in the market, he prioritizes choice. That means pricing becomes granular and gives people what they want, “but at a price at which it becomes not economical anymore if they do that repeatedly.”
Top insights for IT pros
From cybersecurity and big data to cloud computing, IT Brew covers the latest trends shaping business tech in our 4x weekly newsletter, virtual events with industry experts, and digital guides.
“The past has shown that people usually go for the low-friction model, and once they become loyal customers, once they understand the value of the service, some of them will become subscribers,” Ene said. “You cannot lose a subscription that you will never get, but you can make money with 98% of people [who will never] subscribe.”
Arjun Pillai, CEO and co-founder of AI platform Docket, said while he sees AI companies making the most of their money through using consumption models, he doesn’t believe subscription models are eliminating any choices from the customer.
For example, Microsoft provides users with a “pay as you go” model for Copilot, in efforts to allow administrators to “better manage costs” and provide users with features of the model “as needed.” The company also wrote on its site that the plan “allows administrators to enable usage-based billing for specific Copilot scenarios, providing users with the ability to use Copilot features without committing to a full license.”
In addition, Copilot is available to users in a free, limited version, as well as one that costs a monthly subscription fee of $20.
“For most of these cases, we see that the subscription-based pricing is hitting a ceiling,” Pillai said. “Pricing is actually enabling more people to choose more things, and that is what is creating this competition and innovation. Because if ChatGPT fails tomorrow, I go to Claude. It’s the same.”
Friedman told IT Brew that it’s critical to talk to existing customers to understand the value of a product when they use it at scale. He said that, based on that feedback, the company can work to “quantify the value” and begin to research “how can we structure financial models to turn their language and their ways of buying into [return on investment] calculations.”