The shift has thrown some CIOs — and their budgets — for a loop, grabbing headlines in the process.
Starting June 1, GitHub shifted from offering a flat price for its premium request model to charging based on token usage for “input, output, and cached tokens, according to the published API rates for each model,” per the company’s announcement. Zendesk and Workday also rolled out changes to their pricing structure as AI features expand.
“Many people look at AI as a technological upgrade,” said Marko Markov, technology, media and telecommunications senior analyst at RSM. “It’s not just a technological upgrade. It’s really a fundamental shift in how SaaS companies will operate.”
Where before, SaaS vendors would price on a per-person model, they’re now moving towards charging either partly or totally on consumption. More than half of technology executives say they expect usage-based revenue to grow by 2027, according to a Revenera report.
Dhaval Moogimane, software industry lead at West Monroe, said that charging on a per-head subscription model may not make sense to AI companies if the technology is used to automate processes. For example, if a software provider works in the customer service space, the number of agents will reduce over time, he said. Those vendors have to “find another metric for pricing, typically moving to a consumption or outcome based pricing.”
For enterprise technology leaders, the updated pricing structures can mean going back to the drawing board in terms of strategy.
Michael Corrigan, CIO at World Insurance Associates, told CIO Dive that AI pricing models with their vendors have changed “very rapidly” over the last three to six months. “We’re a big SaaS user for most of our systems, and we’re used to paying a per-seat cost based off of growth of organizational headcount.”
Many vendors the company works with are switching to a combination of seat license and consumption based pricing, Corrigan said. And while seat-based costs are easy for them to plan for, token burn is not.
“What we have to do is a little bit differently than in the past and establish those baselines and forecasts,” he said, then work with vendors to “really flesh out what your expected costs are going to be out front.”
Corrigan also keeps in mind that while cost for AI embedded in SaaS may increase, the tools are “giving you and your employee either time back in their day or quick processes and more consistent processes.”
Getting through pricing transitions
The pricing model updates signal a moment of transition, Markov said. CIOs will need to get an understanding of the tokens they require. That’s especially true for CIOs who find that their organization either has surprisingly high bills or runs out of tokens too quickly. “They’ll be able to forecast token usage and token cost a little bit better.”
Understanding usage will also help CIOs adjust software costs based on seasonality instead of paying a per head license for an entire year, Markov said.
“If your hotel chain is using more tokens in the summer, you can forecast for that and as a CIO, you can adjust your budget,” he said. The strategy would also limit token use in the slower winter months.
Corrigan anticipates that forecasts will help them understand the cost of different scenarios, which will in turn help the IT department understand how employees are burning through tokens — not just in IT but across the organization.
“We’re trying to get our arms around that right now,” Corrigan said. “It’s a moving target.”
Rethinking spend strategies
While this pricing shift affects everyday business tasks, it can also affect innovation, which could create a process and financial headache for CIOs.
Companies with in-house developers running AI-driven pilots, might need to reframe their approach to experimentation. “AI adoption is an R&D process,” said Alex Bakker, director and primary research lead at ISG. Some of those developments aren’t going to work, which might make AI token spend seem a waste. But if something hits, it could yield big results. “If you’re comfortable with the fact that some of your research has no outcomes, you’re going to be in a better position.”
For companies that aren’t comfortable taking that risk, the best approach is to rely on commercial vendors with built-in AI features.
CIOs are still setting expectations around what AI can do, and what it should cost, Bakker said. If someone uses a free ChatGPT account and sees it works really well for basic tasks, they might expect AI at work to be able to do complicated ones with the same ease, and cost little to nothing for it.
“That positive surprise probably created a bit of irrationally high expectations,” he said. “If you go in with low expectations there’s room for positive surprise. If you go in with high expectations that this thing is going to cure cancer and should also be able to do my accounts payable processes, it can’t do that.”
