Insurers and their technology vendors have spent the past two years asking what artificial intelligence can do. Now, many of them are asking what it will cost.
Sergio Zuniga, CEO of Blitz, a commission management platform expanding into the US from Latin America, told Insurance Business at ITC Vegas that the price of AI models has become a real worry for his company.
“You know, everything’s about the model, the tokens,” he said. “We are pretty scared because we know those companies are losing money. So at some point they will start charging more, but we cannot identify that, right?”
Tokens are the units of text that AI providers use to measure and bill usage. Zuniga’s concern is that the AI work Blitz does for clients could suddenly “cost us or our clients double or triple.”
“So those hard conversations are like, well, how we can try to contain the use of AI within some cost parameters and to see how we are going to move forward,” he said.
The research points both ways. Gartner predicted in March 2026 that by 2030, running inference on a large language model with 1 trillion parameters will cost AI providers more than 90% less than in 2025. The same Gartner forecast on falling AI inference costs warned that those savings will not be fully passed on to enterprise customers, and that overall inference costs are expected to rise because token consumption is growing faster than prices are falling.
In August 2026, Gartner went further, predicting that inference costs per agentic AI workflow will increase more than fivefold through 2028. Agentic tools, which carry out multi-step tasks on their own, are already being built into insurance software, including agentic AI platforms for small business benefits.
The pace of change makes budgeting harder. “I mean, every 3 months is something completely different,” Zuniga said.
Shift Technology, the fraud detection and claims AI company, has turned cost control into a routine. Jeremy Jawish, its chief executive and co-founder, described the approach during an ITC panel on fraud.
“But our job every week is to benchmark these models to the point where now what we do is even we trade from these models,” he said. “Their price is changing, their efficiency is changing, their models are changing.”
Simple tasks go to cheap models and the most complex work goes to the priciest ones. “I’m going to send it to the most expensive one,” Jawish said. “And this is how we’re able to get really good results and making sure the cost stays under control.”
Gartner recommends a similar approach, which it calls inference tiering: routing each part of a workflow to the model it actually needs.
The same panel heard warnings about AI-generated fake claims outpacing insurers’ fraud defenses.
Moderator Contessa Brewer of CNBC told the panel about a chief executive she had met at another conference. “They found an employee just experimenting on the enterprise platform and it cost the company $70,000 in one month,” she said. “But that, that issue of now token costs are a real headwind.”
Cain Hayes, a former health insurance chief executive, said companies need to run at different speeds. In his view, “you really have to have more than one speed as it relates to AI,” moving quickly on experiments and more deliberately on decisions that affect customers’ health or finances.
The spending is meant to pay for itself in output. Shift has about 700 staff, Jawish said, and the target is for them to do far more. “So our job right now is making sure that with the 700 people we have, instead of producing 700 FT, we want to produce 7,000 FTs.”
Other executives at ITC described similar shifts, including Cambridge Mobile Telematics CEO Bill Powers in his remarks on telematics, consumer trust and how AI is changing staffing.
For Jawish, managing that spend is now part of the job. “So we have a lot of benchmarking control and compliance, which means our job now is benchmarking, testing, training.”
