ORLANDO, Florida — Outside of Silicon Valley, tokenmaxxing is a joke. Really.
Some people define tokenmaxxing as proof that a company is maximizing AI automation by the size of its token costs — which one IT leader recently called "nonsense." Others talk about it as squeezing every last drop of value from the tokens they buy.
Either way, CIOs have reason to watch AI consumption costs. In a recent Deloitte survey, 53% of CFOs said their organizations primarily review AI usage costs after the fact, through bills or periodic reports, rather than monitoring spending in or near real time.
That raises a question Rebecca Wettemann, founder of independent research firm Valoir, has asked repeatedly this year: "Could this LLM call have been a workflow?"
Meaning: Could good old, less-expensive, rules-based automation — rather than "throwing AI on it" — be the smarter choice when the work doesn't require an LLM?
CEOs, who don't necessarily have a technical understanding of how AI works, are pressuring their CIOs for fear of falling behind the competition. But pressure to adopt AI can run ahead of results.
Only 13% of CIOs said their enterprises had realized significant or transformational value from AI, according to a Gartner survey released Oct. 19. CIOs need to examine what really requires probabilistic AI and what can be handled more cheaply in deterministic workflows.
“Even if they can, it doesn't mean they should.”
Rebecca Wettemann
Founder, Valoir
"People are realizing now that a lot of the stuff that they did probabilistically in the beginning, they spent money they didn't need to," Wettemann said. "Even if they can, it doesn't mean they should."
Process analysis is key
Leaders at the OpenText World user conference said that minimizing LLM calls — and the token costs they incur — involves putting their company's processes under the microscope, fixing them and routing as much as possible through traditional workflows.
Such analysis starts with visualizing a process in a tool such as Microsoft Visio and understanding it before AI even enters the picture, said Megan McFarland, vice president of IS operations at Ochsner Health, which runs 47 hospitals and more than 370 clinics and urgent care centers across Louisiana, Mississippi and Alabama. Her team is automating and organizing asset management, among other processes, with both AI and traditional software tools.
Adding AI to inefficient processes is the path to higher consumption costs. Also, process analysis may uncover outdated workflows designed for one — or a few — employees who might already have left the company. Without looking closely at process, adding AI needlessly compounds consumption and incurs technical debt.
“You should check-mark the previous things before you get to AI.”
Megan McFarland
Vice president of IS operations, Ochsner Health
"AI should be your final result in terms of what you're going to use to quote-unquote 'fix a problem'," McFarland said. "It needs to be after you've gone through whether it needs to be a process improvement, an automation opportunity, or heck, sometimes it's a people problem — they're not following the right process. You should check-mark the previous things before you get to AI."
Vendors are hearing it, too
OpenText, which lets its customers choose their own LLMs and is staying out of the multi-vendor agentic orchestration game, has observed that its customers are examining AI costs more closely.
At a customer forum at OpenText World, some CIOs discussed which processes were better suited to traditional automation and which needed AI, said Shannon Bell, executive vice president and chief product and technology officer at OpenText.
"A lot of enterprises are focused on wanting to address different AI use cases," Bell said. "But automation can be a solution that delivers the same outcomes at a lower cost point."
Waqas Ahmed, vice president of AI engineering at OpenText, said that this whole discussion boils down to avoiding a five-dollar solution for a two-penny problem.
"That judgment is part of the enterprise learning that we are going through," Ahmed said. "It doesn't come as a principle that you can simply state and people are going to remember; it comes through a disciplined practice within your enterprise where you identify the use cases and decompose them."
Don Fluckinger is a seasoned B2B technology journalist with more than 30 years of experience specializing in enterprise IT, digital experience and content management. As a senior news writer at Informa TechTarget, he delivers award-winning analysis that helps IT and business leaders navigate complex technologies to enhance customer and employee experiences. Got a tip? Email him.
