“When we started this journey a year or more ago, we thought of AI as a tool or a platform or a system,” said Guru Bandekar, chief procurement officer and VP of enterprise systems and security at Carrier. “When we changed that mindset to think about AI as our operating model, I think that really unlocked a lot of things.”

Embracing agents 

Carrier has over 20,000 employees using Microsoft Copilot and, increasingly, agents, said Bandekar. The company set up preconfigured agents it allows the entire workforce to use, and allows a cohort of trained employees to build their own agents for specific problems they want to solve.

When Carrier hires welders, electricians, and different tradespeople for its factories, an agent now automates scheduling for the in-person job interviews. Carrier also launched an agent to continuously monitor the evolving global tariff situation and calculate how any changes will impact the company’s costs.

One standout agent translates purchase requisitions into purchase orders, which Bandekar said has delivered “tremendous ROI.” With Carrier constantly buying components and services from thousands of suppliers, this used to be a severely time-intensive and tedious task. The agent reduced that processing time from approximately 15 minutes per transaction to less than 30 seconds, according to the company.

Realizing ROI in this use case came down to not just using AI to automate an existing process but as an opportunity to reimagine the process in its entirety, said Bandekar.

“When we are able to reimagine the entire workflow with an AI-first kind of mindset, then we get a lot better ROI,” he said. “If we try to automate the workflow that existed, that may not be the most ROI-friendly agent.”

Keeping AI fit for purpose

When it comes to Carrier products, AI has allowed the company to move from individual pieces of equipment to integrated solutions it can better monitor and optimize. 

One example is what it calls “permanent inspection” of residential devices. This involves using a series of AI-based diagnostics algorithms to monitor the efficiency of each piece of connected equipment—for example, a heat pump—to continuously calculate and compare how the device should be performing versus how it’s actually performing.

“With AI, we can do it targeted down to the individual device, the operating conditions, and the ambient conditions, and you get really a significant benefit because you have basically product-individual optimized diagnostics,” said Klausner. 

If a deviation is detected, the system will also detect the root cause. If it’s a problem with the user settings, an alert will be sent to the user. If it’s related to other conditions like the installation, it can then provide guided advice to the company’s installer partners to fix or optimize it. Carrier was previously able to accomplish some of this with physics-based models, Klausner said, but AI is allowing them to do it in much more narrow, targeted ways.

Ironically, Carrier is also using AI in data centers to optimize efficiency in both the design phase and the active cooling of the very equipment that trains and runs AI models.

“If you put AI into that, if you apply that to that unique data foundation, you can really optimize the operation of a data center in a way that [otherwise] would not be possible,” said Klausner.

Despite its embrace of generative and agentic AI systems internally, Carrier has primarily been combining machine learning and physics-based models for its products, as this has consistently delivered the greatest benefits, said Klausner. 

In terms of where Carrier is getting its AI, the answer is “all of the above.” The company is using all of the frontier models as well as fine-tuning open-weight models for specific purposes.

Overall, remaining open to all forms of AI and all models has been a key part of the company’s AI strategy, according to Klausner, and Carrier remains focused on always selecting the right model for the right job. 

“That’s a very deliberate decision in how we go about that,” Klausner said.

Lessons learned 

As with any technological transformation, Carrier’s AI journey has tested some commonly held assumptions and delivered valuable new insights. The first lesson Carrier learned was that AI is only as good as the quality of the data that goes into it. Klausner’s team created a standardized, unified data architecture that allowed the organization to curate the company’s vast quantities of data for more effective use by AI.

“That’s basically the core,” said Klausner. “And then the second [key lesson] is that data in itself has no value. The outcome you generate with the data has value.”

This means not just assuming that agents or any other reimagined AI workflows are in fact going to deliver on the intended benefit. To stay focused on defining and evaluating outcomes, Carrier leadership formed an AI council chaired by the CEO, Klausner, Bandekar, and other key leaders. Together, they examine every AI business proposal through the lens of direct, tangible outcomes—whether that is making a process more efficient, reducing costs, or improving the customer experience.

“That is the mechanism at which we select those projects, and that same mechanism then is used throughout the life of those projects to evaluate whether they’re giving us the intended outcome,” said Bandekar.