Agentic AI in the Commercial Workflow with Johnson Controls CDIO Vijay Sankaran

Speaker A: Hi there and welcome to Enterprise AI Innovators, a, uh, show where top technology executives share how AI is transforming the enterprise. In each episode, guests uncover the real world applications of AI, from improving products and optimizing operations to redefining the customer experience. I'm Evan Reiser, the founder and CEO of Abnormal AI.

Speaker B: And I'm Sam Motamity, a general partner at Greylock Partners.

Speaker A: Today we're talking with Vijay Sankaran, Chief digital and Information Officer at Johnson Controls. Johnson controls is a 140-year-old company that runs the heating, cooling, fire and security systems inside everything from hospitals and pharma labs to the Burj Khalifa. They serve millions of customers across nearly every country. When you lead digital transformation across a physical footprint that big, AI stops being theoretical fast. A few things stuck with me from this conversation. First, Vijay's team took a 300 page RFP that used to take sellers days to process and built an agent that pulled pulls the relevant specs in about a minute, running in parallel across 10 reps at once. They're already freeing up 8 to 10 hours a week per seller and he says they aren't even a quarter of the way there. Second, in just 18 months, he flipped his team from 70% outsourced to 70% internal. That in house capacity is now shipping AI agents at 20% of what a third party charged and roughly triple the speed. And finally, Vijay says he now spends more time as a thinker and a reader connecting dots than as an operator, acting as the company's internal evangelist for what's actually working with AI. His view is that the modern CIO's job is to be the chief coach for adoption, not the person running the IT plumbing. Vijay, thank you so much for joining us today. Really good to see you and excited to chat about some AI stuff. Maybe to start off, do you mind giving our audience a bit of background about your role today at um, Johnson Controls and maybe kind of how you got there?

Speaker C: Yeah, absolutely. So I, I'm the Chief Digital and Information Officer for Johnson Controls. Johnson controls is a 140-year-old company that's been focused on thermal management, um, across all kinds of different facility types, uh, mission critical environments, hospitals and uh, manufacturing facilities, pharma labs, as well as decarbonization across all different kinds of facilities, and then most recently thermal management for data centers. So data center cooling and things like that. And what my team does is really build digital solutions to support all of those different strategic vectors. I've been with JCI for about five years and a couple of months. I joined as the Chief Technology Officer where we really built the OpenBlue platform that enables our digital capabilities for our customers and grew that significantly through several acquisitions and organic growth focused primarily in decarbonization and mission critical verticals around how do you drive energy efficiency, Net zero based optimizations, predictive maintenance and monitoring, uh, and now we're taking that all to the next level with the capabilities that you can get from LLMs. I took on the enterprise digital transformation about 18 months ago and as part of that now we're thinking about how do we take data and new AI capabilities to really amplify the effects of the business process transformations that we're doing across the ecosystem. So we are on a journey right now in terms of implementing a uh, business system across all of our key processes across the enterprise. And AI and data uh, are clear enablers in terms of how we amplify that to the next level. So it's pretty cool to see all of this come together, you know, both from how do we consume AI, how do we support the growth of AI while keeping in mind decarbonization efforts, you know, as, as energy consumption grows. So I feel like we're you know, in a unique place in history to help the planet evolve in a responsible and yet technologically advanced way.

Speaker A: I think like in our world Johnson Controls is super well known. It might not, I don't think people really fully appreciate the full kind of scale of operations. And you can share like obviously not asking for like super top secret numbers here but like anything you can shares give perspective on, I think that like how big of a role you play, you guys play in the world.

Speaker C: Well we have millions of customers all over the world across every, almost every country on the planet, certainly every continent with the exception of Antarctica. Not sure if we're doing anything there yet. And then if you just look at the types of buildings that you know we are in, whether it's the Burj Khalifa in the Middle east or whether it's you know now a uh, major financial headquarters in New York City or different mission critical hospitals all around the country. Millions of customers all across the world with a world class service technician team that has expertise and leading thermal management capabilities with H Vac controls, fire, security and digital. There is no company out there that has both the breadth, the depth and the reach that Johnson Controls has. So it's pretty exciting to think about how we impact people's lives every single day.

Speaker B: Maybe just starting higher level, we have a lot of chief information officers, chief digital Officers, chief technology officers who listen to the show. And I think all of them, including yourself, are facing a flood of AI ideas coming at them, right, every day, internal ideas, customer facing ideas. How do you decide what to go prioritize?

Speaker C: There's a lot of uh, in my mind experimentation that needs to go on around what is the right way to do something and what are the right problems to, you know, uh, apply this to. Right? And so that's a lot of where, where we are is planting a lot of seeds, aligning to, to the business systems, looking for continuously better ways to do things to optimize economics. Right? Because as an outcome, right. If you think about it, let's say that you have 100 engineers, right? You know, the, the hypothesis is that either that same capacity of 100 engineers can do with a, ah, five or ten AI engineer investment through, you know, one of these frontier models, right. Or Open weight models, 10x what they were doing before, right. Or you're going to, you know, reduce your staff of engineers by some percentage so that you can get, get that return on investment and value in terms of your throughput. Same thing is true on internal processes. You know, if you have sellers and you're, you know, amplifying your commercial process, you want them to produce, you know, 3x or 5x the number of proposals that they were doing because they now have AIs to respond to RFPs and, and so on and so forth, right? And so designing the system, designing the architecture, making it very modular so you can figure out what the right models are and then real measurement. In my mind that's really where CIOs need to live into the future. I can't underestimate the importance of architecture and then also data in that equation because it all plays a huge role in whether or not the capabilities that are being generally rolled out can add value within an enterprise ecosystem.

Speaker B: What use case has been most surprising? Like either the effect it's had, the fact that it works so well, which one stands out.

Speaker C: I mean, one of the use cases that I absolutely love is so, you know, as we started out this business system transformation, right. One of the things that our CEO observed, as you know, use a Japanese term, gemba, go to, seeing where the work is, is done, right, Is that sellers were spending a lot of time doing non value added work and they were basically, you know, duplicating the entry of information into multiple systems. They were going through large RFP documents to pull out, you know, only the relevant sections. They were then taking that information and Then putting it into another system, system for configure price quote activities. And so, you know, when he first shared that, I said, hey, that, that sounds like we might be able to do something there with agentic AI, right? And this is very early days, this is last year right about this time. So we said, hey, let's just do something small and simple to prove out the con, the concept, right? And so I'm, I'm here in Ann Arbor, which is not very far from Detroit. So we use the Detroit branch in Auburn Hills as a pilot, took one of the key sellers, asked him to walk through the activities through what we call as a value stream map. And then through that we said, okay, let's reimagine this, you know, let's run an amplify workshop using, you know, agentic AI. And so we had a team there, hands on keyboards and actually just building some agents to say, hey, can we take uh, a 300 page RFP document and take the relevant information out of it, find it and with the right accuracy level and what we found is with the commercial folks actually sitting there, they were just blown away by it. They're like, oh my gosh, this would take us days to do. And you've built an agent that can do this for us in minute. And by the way, it's not serial, it's parallel. I can run 10 different RFPs at once. And then as we mapped out that, that value stream, we would challenge whether or not in that value stream, you know, that step needed to have some sort of a human in the loop or intervention or we could build an agent to do that piece of the work. Right. So the next step was, okay, let's take out the strip, the document for the drawings and you know, the specific configurational elements. Let's actually put it into creating an opportunity within Salesforce. And then let's put that into our configure price tool and let's, you know, let's set up the configure price tool as a MCP server so that we can actually serve up uh, different configurable combinations. Let's put pricing in there so that there could be trade offs around cost price, you know, based upon what the customer's parameters are. And then eventually let's give that to customers and agents so that they don't have to talk to, you know, our highly specialized sellers. Right. So we actually thought through this entire value stream and said where can we put an agent in right, that can actually do that? And the whole, the whole outcome was to double sellers time. Okay. And just through the AI part of it, right. So we wanted to go from 10 hours of seller time to 20 hours of seller time a week. Right. And just in the AI enhancements that we've delivered thus far, you know, we're already seeing, you know, 8 to 10 hours improvement just with AI and we're not even, I would say, 25% of the way, way there.

Speaker A: What would be kind of like the three things you'd say, hey, if you're just getting going AI, uh, transformation. Here's three areas to start. Get these things nailed, the points on the board, and then we can build off that to like the next, you know, maybe the next set of more ambitious projects.

Speaker C: Yeah, I mean, uh, I guess I'll speak from my industrial lens because there's a lot of, you know, industrial use cases out there. But the three quick hitters, I would say, no matter what, are commercial service and customer service. And so commercial, like physical service and customer service. And I'll say why I would do those three. One is, you know, we are all, you know, you guys have, you know, software companies, you know, in Greylock, you guys are evaluating lots of proposals from people. It's, it's an onerous task, right? And the reality is, is that the answers that you've produced before and the answers that you need to produce for response are pretty much consistent based upon, you know, certain input parameters. Right? Like you've done that in some shape or form or in a combination of some shape or form. So if I know that, you know, I'm, I live in Ann Arbor, I'm specking, you know, an H vac job for a wing of the University of Michigan Hospital. And I know that it's 100,000 square feet and it's got seven floors. And you know, these are certain characteristics, right? Instead of me going there and spending a lot of time with that person kind of walking the floors, taking notes and things like that, you go to AI, you develop an agent that you say, okay, I, I'm going to propose a system for the University of Michigan Hospital. And it's about this big. And this is the name of the center. Can you get some public information? And then you just ask AI to go do that and produce sort of an initial proposal doc based upon all of the proposals that you might have produced. And for 140-year-old company like Johnson Controls, that's a lot of proposals that you've produce. You've got all the manuals about all of your chillers and Air handling units and, and everything. Right? So that's one, two is service technicians, right? I mean all of us have, whether you're in residential or whether you're in commercial, you know, you have service technicians because some warning light goes on or the water is not, you know, you're not getting good water pressure in your shower or. And the same things happen, you know, in commercial facilities too. You know, you're not, you have cold spots in the building and things like that. Yet you have a lot of young service technicians now because you know, the um, the people who are skilled trades sort of disappeared for a while and now being in skilled trade is in vogue again. And so we've built an agent around, you know, service technician productivity where we put all of our manuals and our knowledge base, you know, into an AI workflow where you know, if there are these error codes, you know, you, you, you fault codes, you put them in, into your AI workflow. It tells you what it is and tells you how to go fix it. And if you want step by step instructions, you know, we can codify that in the AR workflow and you have it at your fingertips. And so both as a teaching tool, right. As well as somebody who you know can be on the site, take photos of different things, take photos of wiring, get an assistant to kind of help them either speed up a job or make sure that is done properly. AI ah can be a real assistant. The third one is customer service and support. Sometimes customers just have a quick question that they want to have answered and I'm still very surprised even in this day and age how little customer support is utilizing AI in a positive way. The things that are already fully automated to enrich the customer experience. Uh, know flew in an airline and I wanted to provide some feedback to the airline and I couldn't even figure out how to pro give them feedback, you know, from their app. You know there's, and I go to their AI bot and it's not really based upon any LLM technology.

Speaker B: What's the balance between what you do centrally as a technology team and the investments you make to sort of enable that versus like what you depend on business owners to do. And, and how do you, how do you sort of manage that internally? And if there's a good example of that working really well around a specific capability or use case, that would be awesome to understand.

Speaker C: When I first took over the enterprise side of jci, it was very much a project and operations kind of a model which you know, the, all the projects were done by External providers and all the operations were done by internal employees. Right. And, you know, if you're an internal employee, that's not really too motivating for you to, you know, know, only be doing the run, the run, run, break, fix stuff. Right. We have some very talented, you know, employees in the organization. And so one of the things that we did over a very short period of time, 18 months, is basically went from a 30% internal, 70% external, and flipped that to basically 70% internal to 30% external. And what, what that's allowed us to do is to have enough capacity of teams that can be very responsive to these kinds of efforts, whether it's building agents or ideating, without having to like, you know, shove pieces of paper across the desk, like sows and bidding and proposals, and quite frankly, been able to do it at economies of scale of cost. 20% of what a third party would have done for us and with, with the, you know, certainly with the speed, that was triple what third parties would do for us. And so it's been a, um, a wild success. And that's just all across the board, whether it's in terms of responding to systems enhancements or whether it's, you know, developing AI. AI agents.

Speaker A: I'd love to hear your perspective on, like, how do you think CIOs or CTOs need to play, like, a bigger role in AI adoption? And like, what's the real opportunity for CIO and CTO right now? Right. It's not just managing M IT stuff. Right. It's like, it really is this, like, business transformation that you've been talking about for a while.

Speaker C: I. I spend more of my time than I probably have, you know, in other periods in my career, being a thinker, a processor, a reader, you know, connecting the dots than an operator. Right. You know, and I spent a lot, except in the area of cybersecurity where, you know, with Mythos, you know, I've had to spend a lot more time on really, the playbook for how do we approach the issues around Mythos? But in the broader business, you know, I, uh, look at all of these new models, how people are using data. I study a lot of use cases. I suggest to my team, hey, why don't we try this, why don't we try that? When I'm in India last week, I'm doing, you know, understanding and forming patterns in my brain about how people are doing agents so that I can, I can share, I can be the, the enterprise internal evangelist for great success stories and how we might be able to solve problems in different ways, ways through AI. So I think that's the role CIOs CTOs need to play. Now is the, the really being that chief coach on how to adopt AI and then working with your teams around, how's the best way to go do that?

Speaker A: One thing we like to do just to kind of sprinkle in some kind of quick hits throughout the episode is do a bit of a lightning round and Sam and I are going to be super annoying and we're going to ask you questions that are like, impossible to answer in the one Tweet format. Trying to go for like your, your one Tweet kind of sound bite. So please, uh, forgive us and Sam, you go first.

Speaker B: So to kick us off, how do you think companies should measure the success of a chief Digital and Information officer in this AI era?

Speaker C: I think business outputs, that's the way I would measure success.

Speaker A: What's your advice to leaders about how to kind of stay up to date with the latest technology?

Speaker C: My team will love this because they just voted me as, ah, most likely to be listening to a podcast. So I listen to a whole lot of podcasts and because I can do while I'm walking my dog or, you know, doing a whole lot of things. So I do listen to a lot of podcasts. Stay up to date.

Speaker B: Is there another book you've read recently that's had a big impact on you and why?

Speaker C: The book that I loved was AI Valley. It's all about the history of Silicon Valley and how we got here. Right. And why I loved it is because I've always been a Silicon Valley geek. I, uh, you know, I watched the show Silicon Valley. I wish I had gone to Silicon Valley at some point in time. Time. But what was so cool about it is, as usual, you know, how all of the cross pollination and you know, the, the failures, the massive amounts of, of capital that's invested and just at the same time how early we are on the journey of, of AI. Right. So I think it's, we're still in, in the first or second innings and I think just to see how this journey evolves based upon how the Valley has evolved, you know, over the last, you know, 30 years. It's pretty, pretty amazing to think what's in store for us as we go forward.

Speaker A: What do you think is going to be true about AI's future impact on the world that most people consider science fiction?

Speaker C: Well, I mean, I think it's all timescale at the end of the day, right. And so, you know, I recently read a column by Peter Diamantes that, you know, all science fiction eventually becomes real.

Speaker B: Right.

Speaker C: And I think it's never a matter of if, it's a matter of when. Right. So we thought autonomous vehicles would be here a lot sooner, you know, but they're not fully here. We probably didn't think intelligent robots who can speak like C3PO would take a lot longer to manifest. But I think those are probably about five years out, you know, tops. Right? You have lots of speaking robots out there. So I think speaking robots in the near term, where LLMs, um, shrink down and processing at the edge goes up, leveraging connectivity for updates, you're going to have a lot of robotic concierges in every aspect of your life in five to 10 years.

Speaker B: What's an upcoming new technology and it doesn't have to be related to AI that you're personally most excited for.

Speaker C: I would say the thing that I'm most following closely right now is energy. I think the next 15 years is we're going to have an energy renaissance because we have to have an energy renaissance, you know, and whether it's small scale nuclear where you have 1 megawatt, 5 megawatt, you know, modular nuclear reactors that can fit into spaceships and different form factors, or you have different kinds of battery formats, or you have solar take the next leap, or even fusion to some degree, I do think we're going to be entering in an energy renaissance. And as it powers AI, it's going to be a self supporting feedback loop in the sense that the new capabilities of AI powered by new forms of energy are going to basically discover new forms of energy and materials that can be used to power society in a clean way. So I'm super excited about that.

Speaker A: What an amazing inspirational note to end on. Well, Vijay, uh, I know we're coming up on time here, but really thank you. Really appreciate you joining us. Super fun conversation and uh, let's do it again soon.

Speaker C: Yeah. And Evan, I'm sure we'll have the opportunity to meet again here in the near future. So thank you guys both for the opportunity. Thanks a lot, Vijay. Take care.

Speaker A: That was Vijay Sankaran, Chief Digital and Information Officer at Johnson Controls.

Speaker B: Thanks for listening to Enterprise AI Innovators. I'm Sam Motamity, a general partner at Greylock Partners.

Speaker A: And I'm Evan Reiser, the founder and CEO of Abnormal AI. Please be sure to subscribe so you never miss an episode. Learn more about Enterprise AI transformation at enterprisesoftware Blog. This show is produced by Abnormal Studios. See you next time.