Johnson Controls rethinks IT for the cloud-native and AI era

For years, Johnson Controls Internationals developed its digital architecture primarily with IT partners. Vijay Sankaran, newly appointed chief digital and information officer, has new ambitions for the fire, HVAC, and security equipment manufacturer, with an eye toward developing its in-house capabilities for a new era.

As chief digital officer for the past four years, Sankaran doted on the customer-facing side of the business. Since taking on the additional role of CIO in January, Sankaran has pivoted hard to address the internal digital business needs of the $47 billion building products and services empire — including its own data centers.

Johnson Controls develops building controls and applications for energy management, physical security, fire detection and suppression, heating and cooling, and video surveillance for skyscrapers, multi-campus enterprises, airports, and a growing number of data centers.

As such, the company is both a service provider to and customer of Microsoft Azure, Google, and AWS. For example, Johnson Controls’ Silent-Aire division is developing chilling and cooling solutions for data centers and modular data centers.

The company, which employs 100,000 globally, relies primarily on Azure for cloud and data services, has deployed a variety of ERP packages, and maintains a mature Snowflake data lakehouse. Johnson had even dipped its toes into some early generative AI applications for document summarization and content creation, but Sankaran has since set to work on a comprehensive overhaul.

Even as the company develops new lines of business for its data center customers and many enterprises returning to office post-COVID, the CDIO’s primary objective is to transform Johnson’s in-house IT operations for the native cloud and AI era after years of being overly reliant on external IT suppliers, Sankaran says.

“We’re focused right now on building our internal capabilities,” he says. “We’ve been a heavy externalized organization, and we made a decision that we needed more in-house talent to give us a solid understanding of our business processes, our data, and the analytics as we begin to drive forward. We’re looking at our overall set of investments across all technology stacks and making sure we’re getting a good ROI on that.”

Bringing transformation home

Johnson Controls now has roughly 4,500 IT pros internal and external, many based in India, working on the next transformation. The goal is to streamline, simplify, and align the company’s cloud-based workloads, data, and analytics to be far more efficient and “optimized like a technology organization,” Sankaran says, adding that cybersecurity is another key focus.

The CDIO established a centralized AI team that will build applications using Microsoft OpenAI Azure as its core platform, adding related tools for efficiency. The software engineering teams are now using tools such as Github Copilot for greater efficiencies and are working with Microsoft to teach all employees about using Copilot for document summarization, though it’s still early in the process.  

In Snowflake, Johnson Controls has a flexible, cloud-based, “high spike utilization” platform for processing a complex variety of data workloads, including data transformations, as well as machine learning and gen AI modeling, Sankaran says.

Aside from building up his technology organization, one of Sankaran’s key challenges is maximizing the company’s use of the large number of ERP systems across the organization. The CDIO is tasked with “figuring out” whether those programs are delivering the intended value for making business operations more efficient.  

There are many ways to approach this challenge. The company could standardize on an ERP package, but Sankara is currently evaluating “using data integration technologies to provide unified views without necessarily doing large scale ERP transformations,” he says.

Palantir AI — which offers AI-powered automation for real-time decisions — is currently in testing as well.

“If you’ve got data structures, it can derive what the meta matching is across multiple different data structures that might not be in the same format and then present a unified view in real-time,” he says. “And then it can take unstructured documents, like a bunch of PDFs, derive the semantic meaning out of them, and then bring it altogether using its own proprietary LLM technologies to summarize five or 10 different data sources and provide a unified view.”

Johnson Controls, a field-based organization that is distributed globally, is still in the early stages of AI development but has a handful of generative AI applications deployed for its field technicians and one for customers to monitor their building systems in real-time.

One of those gen AI apps significantly helps field technicians troubleshoot a piece of equipment or perform a fix much faster and more efficiently, resulting in significant savings. Another gen AI application was developed to hire and train technicians globally and accelerate upskilling.

“What we’ve done is use a pattern of retrieval augmented generation, or RAG, which uses a vector database to query our internal, proprietary manuals, and then use OpenAI to produce the summarization we need, and potentially translations with a link back to the original manual so that a field technician can refer back to the original manual as well,” the CDIO says.

“It’s a completely closed loop in the way we’ve built it so there is the ability to reference while protecting our own domain data from being exposed into the public domain,” Sankaran adds. “Given the number of technicians we have globally, even getting a savings of 30 minutes of productivity per technician each week is pretty significant. We are seeing productivity gains.”

Sankaran was formerly CIO of a financial services company at the forefront of technology. Switching to Johnson Controls in 2021 was a major challenge for Sankaran.

Founded in 1885, Johnson Controls has acquired many smaller companies and there is a lot of fragmentation across its portfolio, Sankaran says. Modernizing the stack and moving much of it to the cloud has enabled Sankaran to optimize IT’s cost structure and focus on value-driven delivery for the top building solutions and services supplier globally.  

Agentic AI is up next, for staff and for customers, but the CDIO’s goals in making the user experience more efficient is far from finished. In the end, taking the time now to prepare the company’s internal IT to complete that journey will prove invaluable, Sankaran says.