Insurers weighing their next investment in artificial intelligence may be asking the wrong question.

Rather than giving in to their fear of missing out with AI and deploying it rapidly, insurers should be taking their time to decide where it belongs in their workflows and where human judgment can do a better job, according to Arun Balakrishnan, CEO of Xceedance, at InsurTech Connect 2026 in Las Vegas.

Balakrishnan described the company’s approach as “orchestrated intelligence,” a model that matches each task with the right mix of people, conventional software and AI.

“Unfortunately, in the last nine to 12 months especially, we have seen organizations have a bit of a FOMO (fear of missing out) around, ‘Hey, we need to start using AI quickly,'” he said.

However, a balanced, thoughtful approach to AI orchestration “is going to be the biggest differentiator” for insurers, he said. He pointed to claims adjudication as one example in which AI can be valuable for extraction and inference in certain instances but applying it broadly can make calculations needlessly expensive.

“If instead of that, we generically put AI everywhere, AI is extracting things which it doesn’t need to extract,” he said. “Some things can be calculated. So, you are spending more tokens doing things it needn’t have done.”

Balakrishnan also sees a need to tightly define what AI can do within operational processes.

“You want it to do only what has been asked, nothing else, and do that task well,” he said. “That’s how you control output every time. That’s how you control cost of doing that work.”

Choosing the model matters, too. AI model providers can carry different costs and outputs for similar tasks, making model selection as significant as identifying the work AI should perform.

“Then the other thing is which model to use for it,” he said. “Some models on the same work can be extremely expensive. Some can be very cheap.”

The same is true for the humans behind the AI models, he added.

“Every one of you will use it differently. You don’t need that in a process. You want consistency,” he said. “And that’s why these pilots aren’t scaling up well. And when they scale, the cost goes up too much.”

He proposed going back to the drawing board with orchestration. This can help insurers assess their AI pilots to determine what needs to be done by AI, when traditional software will do the job and where human intervention is still needed.

“That orchestration will make the vision,” he said.

Despite the challenges insurers have faced in translating early AI experiments into scalable solutions, Balakrishnan remains optimistic about the technology’s long-term potential. The industry’s initial rush to adopt AI, he suggested, does not diminish the opportunities that more thoughtful implementation could create.

“I think it’s a hype phase wherein people try to do it too soon, too quickly, in a rush,” he said. “But I’m still a big believer of the tech, and I feel if done right, all these pilots will scale. And there is a very strong intent of the industry and all of us to make it happen.”