Customer service AI pricing just turned into the biggest shell game

Pega's economic argument

The Pegasystems model is interesting because it also remakes agent architecture processes designed to minimize large language model calls during the design phase, writes Keith Kirkpatrick, vice president and research director for the Futurum Group.

"There is increasing concern in the marketplace about the amount of money that's being spent on AI and the actual value that it's returning," said Don Schuerman, CTO and vice president of marketing and technology at Pegasystems.

"People are realizing that if you're not careful, you can send agents off to burn a lot of tokens without them making a meaningful difference in the efficiency of your business or in the experiences that you return to your customers."

Constellation Research analyst Liz Miller said the current fog of tech jargon around outcomes-based pricing has obfuscated that point. Pegasystems has figured out a key to sustainable AI economics, she said: repeating processes that should be repeatable with business rules after AI figures it out the first time. One doesn't need to invoke Anthropic or OpenAI models anew for every single case.

"Once you have that the first time, you can then run through good, old-fashioned 'what we used to call AI five years ago' -- deterministic, runtime models that are still AI-driven processes," Miller said. "It's just not going through that whole costly agentic token cycle."

Defining the 'AI outcome'

On the surface, these ideas appear to put the customer at the center of the pricing equation -- and vendors seem to tie their AI pricing to the customer's business goals. But the fact is, vendors still control the definition of that subjective word, outcome.

"Until we get better at monitoring -- on an ongoing basis -- whether AI is executing effectively, outcome-based models are going to be problematic," said Rebecca Wettemann, founder of independent research firm Valoir. "They have always been problematic in software. They're even more problematic when I don't actually have the ability for a human or AI to evaluate at scale."

The vendors will tell you that they listen to their customers on their various boards, panels and surveys. They do -- and will continue to do so.

But the hard economics of generative AI only give them so much wiggle room: No one, ever, shed a tear for a SaaS vendor, but the reality is that they have back-end costs from AI companies like OpenAI, Anthropic, DeepSeek, Google, xAI and Meta. Someone's got to pay them.

That's the bad news. Is there any good news?

Tech buyers, stay hopeful

Customer service tech companies know this is a market sea change. Each vendor is trying to come up with the best-sounding narrative that proves they're the ones making AI costs the most stable and predictable -- and that they really care.

They have to.

Kishan Chetan, executive vice president and general manager of Agentforce Service, and his colleague Prasad Raje, senior vice president of product management, said that Salesforce vetted with customers its ideas for the outcomes-based pricing it plans to roll out for Agentforce Help Agent this month -- and that it will continue to do so as the company bends its cost models to plug into their business goals.

"The reality in this agentic market is: All of us are continuing to evolve as we work with customers," Chetan said. "There's no precise answer."

For customer service -- and agentic AI, in general -- you could call the current environment an AI "hype cycle," the "formation years," "unfinished business," the "messy middle" between the early innovators and the mass adopters or whatever.

CIOs and their buying teams can take solace in the fact that, amid this chaos, if past IT pricing trends around service-oriented architecture (microservices), cloud computing and data lakes hold with AI, both pricing and barriers to entry will come down. Several factors could influence the race to the bottom, including constantly improving, more efficient compute power and data centers.

Also, Wall Street could play a role in this. xAI just went public. Meta is public. OpenAI and Anthropic plan to go public. Stockholders, demanding as they are, will insist on sharper competition to drive higher revenue -- and that's likely to commoditize generative AI further and drive down prices.

The caveat, vendors and analysts alike said, is the race to make AI more and more powerful. As models evolve, they will drink up more compute power, energy, water and costs. This opposing economic force could cancel out some of the pricing benefits of AI companies going public.

This might seem like a time when "fear of missing out" on AI -- sentiments sometimes driven by CEOs who don't really understand how AI works or how legal and data readiness constraints slow adoption -- is rapidly turning into "fear of messing up," as Wettemann frequently puts it. CIOs are caught between lofty AI expectations and the difficulties of executing on them.

But even in this upside-down world where it doesn't seem possible that the customers could be in charge of the AI situation, you all are. You hold the power of the purse. If you don't sign that renewal, your vendor pals don't get paid. And if they don't negotiate on your terms, you don't have to sign the renewal.

That is, after all, how enterprise IT works, even though this time around the negotiations might feel kinda like you're buying a used car from a pushy salesperson -- or worse yet, playing a shell game.

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.