The AI pricing conundrum remains Unresolved

The AI pricing conundrum remains Unresolved
We are a little over two years removed from OpenAI’s release of ChatGPT, a moment that will be remembered as a pivotal shift in the technology landscape. Almost immediately, we felt the impact of that change, and it wasn’t long before every software company out there had to adjust and add generative AI capabilities to their platforms.
But AI doesn’t come cheap, not for the LLM vendors, and not for the people tapping into those models. Many companies have piggy-backed on the ChatGPT API, for example, which was released just four months after the release of the first ChatGPT app. As with any commercial API, the company using the service has to pay each time their application hits the gateway, and it can add up fast.
The issue isn’t that OpenAI is charging for access to its API because that’s how the industry works, it’s how software companies recoup those costs – and companies have struggled to find a pricing model that works.
Just this week we saw this struggle playing out in real time. Microsoft, a company that has fully embraced AI, is still having trouble figuring out how to price it. For some time, it has offered Copilot with Office 365 for $30 per user per month, over and above the Office 365 monthly fee you pay.
This week, the company introduced a pay-as-you-go option, designed for folks who don’t use Copilot enough to justify $360 per year subscription price, but want access to some of the features on an occasional basis. The thinking is that once you start getting into the habit of using the advanced capabilities, perhaps you’ll eventually sign up for a regular subscription.
Not to be outdone, Google also announced that it is bundling Gemini, its generative AI toolkit, into the business-focused Workspace product for an extra $2 a month. At the same time, it eliminated the $20 per month Gemini add-on, apparently concluding that not enough customers were willing to pay that.
Consider that when Box introduced generative AI features across the platform in October 2023, it used an unusual token system, letting each user have up to 20 generative AI action credits as part of the base plan – an action was any interaction with the generative AI tool such as asking a question about the content in a document. After they spent those 20, they could dip into a pool of shared credits, and if they went over that it would require a conversation with sales.
The company was trying to find a way to let people try the AI features, while recovering their costs, especially if power users were using the AI features more frequently. Box wisely abandoned that approach last June, opting instead to bundle the AI features into their enterprise plus pricing tier, which had to simplify the accounting for everyone involved.
All of this maneuvering points to a pricing model problem, and as we enter the agentic AI age it’s only going to get more challenging. When agents are undertaking a series of actions and producing a series of outcomes, does it make more sense to pay by the outcome? Perhaps, but outcome-based pricing leads down a rabbit hole of issues that remain very much unresolved.
It’s a nice idea in theory, but when you potentially have dozens of agents moving through an organization from multiple vendors, crossing multiple systems and undertaking numerous actions, how do you determine which vendor was responsible for any particular outcome, and how do you begin to think about how to charge for that kind of approach? I have the feeling that CFOs and CIOs are going to want to have some idea of what their bill is going to look like every month.
As we saw with spiraling cloud infrastructure costs, customers are reluctant to give vendors carte blanche, then find their costs spinning out of control. There needs to be some sort of controlling structure in place.
That, friends, is for smarter people than me to figure out, but as we’ve seen with generative AI functionality, it’s no simple matter coming up with a plan to recoup your investment in AI, however you chose to implement that tooling. As we saw this week, that means that companies are going to continue experimenting with different pricing strategies for some time to come until they land on something that works well for them and their customers.
-Ron
Photo by Angèle Kamp on Unsplash