Why Trimble’s VP of Technology Innovation Embraces Chaos while Reducing Friction

Speaker A: Trimble started in 1978, and today software is about 70% of our business and the rest is connected hardware. They often describe construction as chaotic and beautiful, and I emphasize a beautiful part here. I assume many might feel the same about their IT landscape. I think the key here is to embrace the complexity but reduce the friction. You don't eliminate silos overnight. It's about culture, it's about contracts, about the structure of the industry. You cannot just eliminate silos, but you can create interoperability by introducing common data environment, by adopting open standard, by using platforms that allow different systems to talk to each other. The next challenge for us is really adopting agents at scale. The challenge here is to move from focus, uh, on productivity improvement, which are critical but not the full picture, and to focus on how we fundamentally rethink workflows. It's not anymore just about doing the same thing faster or better. It's about transforming how work gets done with this technology.

Speaker B: Welcome to the new automation mindset, where AI automation and integration come together. I'm your host, Markus Zern, and as chief Strategy officer and part of the founding executive team at workato, it is my mission to find these top innovators in AI automation and integration and share their journeys with all of you. So today I have with me Aviat Al Magor. Aviat is the VP of Technology Innovation at Trimble, and he's also a founding member at Trimble, uh, Ventures. And you know, Trimble, uh, has been around almost 50 years. You know, founded 1978 in Silicon Valley. And we'll hear from Aviat, uh, you know, how data really mattered at Trimble. And data, of course, is cornerstone of an AI transformation. That's what everyone is, uh, interested in these days. So, uh, you know, Trimble, you had to deal with lots of different data and also just in general, lots of data, uh, since the very beginning. And so I think it'll make for, uh, a really, really, uh, cool case study for all of us to learn from. So, Aviat, welcome and maybe you can just introduce first, uh, to the audience Trimble, uh, what you guys do and explain how data matters to Trimble.

Speaker A: Sure. And thank you, Markus, for inviting me. It's really a pleasure to be here. We might need more than 45 minutes to address this first question, but I'll try to be short here. So as you mentioned, Trimble started in 1978 in, uh, Silicon Valley. Originally mainly a hardware company, actually the first to deliver, uh, commercial GPS technology to the market.

Speaker B: Ah.

Speaker A: But over the years we've grown from uh, this kind of hardware focus to uh, a technology company serving industries like construction and transportation and geospatial. And today software is about 70% of our business and the rest is connected hardware. Think about laser scanners, machine control, mobile, uh, mapping technology, IoT or GPS hardware. Actually hardware that produce tremendous amount of data to support our customers through our software solutions. That's about Trimble. And for your second part of the question about data types, we are handling, uh, unstructured data and structured data, physical world data and digitally generated data. Uh, think about positioning data point, cloud operational data, um, BIM, building information modeling through our 3D modeling application, uh, video images. So really a diverse set uh, of data, uh, and obviously business data, contracts, schedules and finance. So the real value um, is basically in connecting those different data streams, uh, from the subcentimeter, uh, precision in the field to workflow in the office, to decision making by stakeholders. Mhm.

Speaker B: Now this is really cool. I do think overall maybe we'll dig into that later. Uh, what you mentioned, connecting data. I think that's where the value is. Right? You know, you have, often enough we have like different data streams and they're there but they're not correlated, they're not like put together and then there's very little value, uh, even though you have the data. I'm really curious, I want the audience to kind of learn from that long journey that you guys had at trimble, you know, 50 years, you yourself over 10 years, like 13 years I believe, right at Trimble. And I believe, you know, CIOs out there, you know, data quality matters to them, that interoperability, uh, that you talked about matters to them. And they're all thinking about data as that foundation to trustworthy AI also, you know, because everyone's talking about AI, um, so maybe let's get straight to the point. You know, if, if I have fragmented organization, I have legacy systems, I have data silos, you know, how can I make them future ready? And what do we need to do as IT leaders to get data right to then also get AI transformation? Uh, right. I'm curious, any thoughts of wisdoms that come to your mind from your experience?

Speaker A: Yeah, I think we're touching here many aspects. Uh, maybe start with the beginning, um, at least my personal beginning. So I joined Trimble, uh, as part of acquisition of a startup company dealing with integrating 3D modeling data with cost data and schedule data. And this relates to your comment earlier about how to connect the different data streams. And the idea behind this startup was to really, uh, deliver a solution where people would understand the impact of any type of change, a change in the schedule, how this impact the cost, a change in the 3D modeling and the design decisions, how this will impact scheduling costs. So this kind of uh, interrelated data streams was a key concept. What we learned uh, is that um, data is only valuable when it is trusted, when it is connected and when it is tightly contextualized to your specific domain. So if I'm looking back 13 years with Trimble, um, looking at this journey that the company evolved through from a set of uh, isolated solutions to uh, integrated workflow with basically a common data environment. And with better data comes better planning. So uh, what we see already uh, with this kind of integration is a massive impact on the customer. Less rework, improved efficiency in the processes, better resource allocation by understanding what is required, what is coming next in the project, um, and improved collaboration which is critical in the industries we are serving.

Speaker B: I really like that. So trusted, connected and contextualized. I think these are kind of three really, really important, uh, kind of thoughts you have to have about data. So let's deep dive a little bit into that, uh, trusted. Maybe let's start, let's start there. I mean everyone talks about, I think, you know, most people are really concerned, you know, what will happen with AI when AI agents, uh, start doing things themselves and they have access to data and that maybe we don't want any learnings you have to, to make data trusted. Well, I think some of this has to do with data qual also. Right. One of that is like what can you do with it, like governance. Uh, and the other thing is data quality. Both I guess fold into that uh, trusted thing. What have you learned? I think you guys build data products also at Trimble. Any learnings from that?

Speaker A: Yeah, certainly. The short answer is that uh, AI, if we look at AI as a future, uh, is only good as a data, you fit it. And this is obvious I think, um, and to get this kind of transformation, uh, right, uh, you need this trustable data we mentioned. And this means improve on the quality, improve on the context and improve on the accessibility of data. And let me kind of uh, uh, provide more color for this. So uh, when I'm talking about quality, I mean accuracy and consistency. You must rely on the data. You must be sure that the data that is provided to those AI systems, um, is in good quality, uh, because bad data just leads to mistakes and by the way, mistakes that are typically communicated with confidence, uh, by the AI agents. So we need to be careful with those mistakes. On the context side, it means understanding um, your specific domain, sometimes about fine tuning the model. And we don't want to go into the details here, but it's about understanding your specific industry and connecting the dot across domains from in case construction, from design to pre construction to construction to operations. So data isn't trapped in silos, but carries meaning into the next step, across the organization, across the project. And last uh, I mentioned accessibility, uh, giving AI agent or even people um, who are part of the project the ability to access and use data in a transparent way, in a secure way and of course with the right permissions to make sure we are uh, protected. So if we improve those three dimensions, AI stopped being a uh, toy for experiment and really becomes a scalable tool for us. Mhm.

Speaker B: Uh, let's look into it. I looked up some of what you said before on the Internet. One of the things that stuck in my mind, you said the construction industry is chaotic but beautiful. And so it looks to me like there are these silos you mentioned. Silos, data silos that I think we all know, uh, and have in our companies. I'm curious, what does that look like in the construction industry? Uh, maybe so that people listening in here can build bridges to their own worlds. Because I think data silos everyone talks about, but the question is like what do you do and how do you tame it and how do you actually get from kind of data chaos to like really trustworthy AI at the end.

Speaker A: Yeah, and you're right. I often describe construction uh, as chaotic and beautiful. And I emphasize a beautiful part here. Um, and I assume many might feel the same about their IT landscape. I think the key here is to embrace the complexity but reduce the friction. That's a key aspect here. So you don't eliminate silos overnight. It's about culture, it's about contracts, about the structure of the industry. You cannot just eliminate silos, but you can create interoperability by introducing common data environment by adopting open standard, by using uh, platforms that allow different systems to talk to each other. So one example for this is Trimble, uh Connect, our AECO or Architectural Engineering Construction Owner and uh, Geospatial Collaboration environment. Triple uh, connect, basically aggregate all the different data types I mentioned earlier, scans and 3D models and PDF, document and schedule and cost and so on. And all this is going into a common data environment where stakeholders, no matter where they're coming from, can analyze, can understand impact of changes and uh, really communicate around single source of truth which is critical because data is changing so fast as the project is progressing. Because, and you must know that if you are in this environment, what you see in front of you is what others will see as well. So now back to the AI discussion. The beauty of AI is that it really thrives on connected data. So every step you take to harmonize and contextualize information move you closer to this kind of landscape where this chaos as we named it earlier becomes um, coordinated collaboration.

Speaker B: You mentioned before. I think it's very important but uh, you know in the past I believe a lot of us when they, when we think about data it's always structured data. Because in the, in the past you know there was some of the data that you could deal with and unstructured data was difficult. Now with Genai it's actually, I mean maybe one of the biggest strengths of Genai is that you know, that's based on language and it actually does provide understanding for unstructured data. Also I'm curious is how you think of the topic with the silos and now thinking about not just structured but also unstructured data. Does that. My feeling is there's probably new silos that we don't even know about yet or you know, maybe more contextuality that that's possible that we don't even think about yet. I'm, I'm curious how you think about that.

Speaker A: Yeah, we see the two uh, the kind of unstructured uh, data, unstructured data as kind of complementary data sets for our customers and they need to use both. That's the nature of the work. So in this case technology, um, and uh, predictive AI and generative AI are supporting us to uh, enable it. And yes I agree with you that generative AI has the ability to really overcome some of the major challenges we encounter with unstructured data. And mainly it provides us the ability to uh, carry information that was processed, sometimes with generative AI, sometimes with predictive AI into the next phases. Because there is understanding of the context. Um, that's part of the nature of generative AI that allow us to do this kind of uh, transition between different domain. And this is a key for uh, us to open the silos and basically create end, uh to end workflows.

Speaker B: One of the age old situations that I believe there's some, some, well some element of friction in, in companies there's like you know, the different silos but then there's this one I guess call it larger conceptual silo which is, you know there's it and then there's the line of business, right. The, the people who actually do the operations, the work and the people who kind of know the technology that supports the work. And you know I have to tell you, you know in the, in the 90s, uh, consultant I did business process redesign. So I was on that line of business side and I had my frustrations with, with it. Uh, um, actually one of the reasons why we started Workado was kind of like break through that a little bit and make, make it better. I'm curious about your experience. My, my. So my thesis would be that for Gen AI to really work well for an AI transformation, that collaboration between it and the business has never been more important. But um, I'm curious your experiences maybe also where you saw this work and maybe where it didn't work and what we can all learn from it.

Speaker A: Yeah. So uh, funny you're asking it now, uh, we will have our global hackathon um, in October, next month. Um, and one of the topics that we are uh trying now to uh, raise awareness to is those kind of hybrid teams because we want to make sure that the hackathon is not just an engineering celebration but actually bringing internal functions as well to the party and working together with engineers to address specific domain problem. And when looking at the work we did and part of the motivation to kind of promote this in the hackathon is that I think the most successful hybrid teams uh, that I've seen worked well um, because of three major conditions. The first one was um, executive support, understanding the value of the hybrid team and really enabling collaboration. And this is critical specifically in the age of Gen AI because this is where this kind of collaboration is really required. Bringing the domain expertise, let's say for finance or legal and the engineering with a know how, uh, on delivering AI and making sure they work well together so the executive can really help with removing roadblocks which are critical when you try to kind of change those uh, traditional systems and they can create the momentum that is required. So this is uh, the first aspect. The second one um, is really about uh, shared objectives and outcomes. It's not anymore about my uh, department and your department. It's about shared OKRs or shared KPIs. The team should understand that they are solving a problem together for the best of the company and not necessarily just for their specific division or function. And then maybe the third one, a bit kind of a uh, hidden aspect but I think is critical is a psychological safety because when you're in this type of um, new engagement type of environment where you Enter an unknown field. Engineers have nothing to do with finance, finance has nothing to do with engineering. You need to be able to show some sort of vulnerability, uh, say I don't know, I don't understand the specific domain. You need to feel free to experiment, to ask naive questions and sometimes even to challenge assumptions without fear. So if you have those uh, kind of three condition met, the executive support, the um, uh, shared objective and the psychological safety, the team start to be not just a random cross functional group, it is really a uh, cross learning group. Uh, and that's where the magic happens. Mhm.

Speaker B: Yeah. I think it seems like you made some really important learnings there. I wish we all would get down that path. I think somewhere, I mean it's getting better. I definitely see it. I mean it is so much better now than it used to be in the 90s. And that's uh, a good thing. So people always want to see like what does AI transformation look like in another company. So yeah, I, I uh, so I, I'm really curious and I'm m sure everyone wants to hear like how did that look like at Trimble? Because you guys uh, given the data that you have and so on, I think there was a lot that was AI even like the machine learning, the good old like more statistical kind of processing. And then now you're doing things with uh, Genai. So really would like to understand how that unfolded at Trimble. And maybe to start with uh first question, like if you can port yourself back first AI initiative at Trimble, uh, like how did that go? Maybe also was there resistance that you had to face? Uh, how did you get the first beginning of AI Gen uh AI. Well let's look at both actually AI and Genai. How did that get off the ground and what made it maybe difficult to get it off the ground?

Speaker A: Yeah, I think there are two questions here. What was the motivation to begin with? And then naturally when you're doing a change at this scale, uh, what was the resistance? So let's start with the first one. Um, and, and our journey started around uh 2010 uh when we introduced machine learning uh for point cloud feature extraction and classification. Just to give the context. Our customers are using laser scanners to create very ah, dense point clouds. Those are files with points in the right location in space. And those point clouds represent the environment that the scanner uh captured. Uh in a very, very precise world. It's kind of sub centimeter, uh so it's a very um, it's a huge uh data and uh, very tough to Analyze. So this is where we started with feature extraction from this point cloud and classification of the points to objects. Uh, imagine uh cracks on the road or pavement or lighting fixtures, stuff like this. So the challenge here was clear and this is where the motivation came from. The challenge was a customer pain where collecting terabytes of point cloud and aerial imagery, but making sense out of this data was nearly impossible to do manually. So by applying machine learning we really enabled them to our customers to do things like frequent uh assessment of infrastructure health. If you do use mobile mapping, um hardware capture all hundreds uh of miles of roads and then process it. With machine learning you get uh, um the outcome in a very uh digestible way. Ah, we could do monitoring of production in construction compared to the design schedule. We can perform quality checks. How is the design compared to the actual work? Any differentiation, any discrepancies between design and actual during the construction and also uh, identify any um deformation during the operation of the facility. Um, so this was a clear value proposition with generative AI. With predictive AI, Sorry, with predictive AI the classical machine learning processes, um, and it is still ongoing and deliver tremendous value to our customers. I can provide some more use cases, uh if you'll be interested. Now with the arrival of generative uh AI we expanded the portfolio beyond analysis into content generation, user experience which is dramatically changing and process automation which is critical for our customers and uh, an opportunity to really connect end to end workflows with automation. So today uh generative AI um is embedded in workflows such as uh generative architectural renderings from a uh simple prompt, um or streamlining invoice processing um or enhancing user experience through natural interaction. Uh, all this is really based on generative AI. So we see in a way machine learning and generative AI as a continuum. Classical and machine learning processes help um analyze and structure the data while generative AI takes that output to generate new content, help automate workflows and transform a user interaction. Um now you were asking me about resistant um and naturally, and I will focus here maybe okay with you on the generative AI part. Um so when we first introduced generative AI um initiated at Trimble, the resistance uh we encounter was um, a natural mix I would say of uh skepticism and fear which to be honest was uh reasonable concerning the maturity of the technology back then with a lot of hallucination and performance issues and limited um modalities and so on. So some team members thought AI was just a hype and uh, too generic for complex high precision domains and other uh Were concerned about um, the risk, uh, deploying generative AI as part of our solution. So what actually helped us move forward um, was Trimble, um, strong innovation and collaborative culture. So it's not about technology, it's about the culture. Team members, um, don't wait for top down instructions. Uh, they bring ideas forward, test them quickly and turn concepts into working pilots. M and this is critical um, in this kind of early phase, uh, it helps us bring uh, the confidence in the technology, understand the limitations, um, adjust to the specific use cases which are low risk and, and provide low hanging fruit. It helps create the awareness required, uh, and even the excitement that we need in order to move uh, the organization toward adoption uh, of generative AI. And once we had those first success stories to tell, um, the conversation shifted. Ah, so that's when we started uh, to scale more intentionally to set clear goals and expectations. Everyone needs to use AI and move from this kind of experimental, um, very agile state to enterprise impact and scale.

Speaker B: Sounds good. So when you're doing something new, when you're affecting change, I don't know where it was but I remember one book that I read said you got to look for the bright spots, you got to look for the obvious improvements or the things where it's crystal clear and then use those as anchors to then do bigger change. So, so taking that as a framework, as a conceptual framework, um, I'm curious if there were any AI driven projects that just had so much business impact that they were almost like, couldn't argue with them. Like the, the small things that, that were just uh, crazy in terms of the impact Any, any of those that you, that you remember.

Speaker A: There are many small things that created an impact and the accumulated impact is kind of um, getting us uh, to where we want to be. And um, this is kind of moving side by side with the maturity of the technology. The more uh, mature the technology is, the greater the autonomy we can provide um, and the greater automation obviously. So let's take some examples here. Um, maybe start with internal, our internal function. So a great example here was how the product management teams transformed our product development cycle by combining large language models, uh, historical data, uh, PRD document and all this kind of data for product and AI assisted coding or vibe coding. Uh, by combining those three together, teams can now move from customer input, which is a start to product requirements, to working prototypes in a fraction uh, of the time it took them in the past. So in some cases cycles have been shortened by 70 or 80%. That's tremendous impact. Uh, it really Changed the way the team uh are working. And the implication of this is faster iterations, you can do more iteration at the same time and much quicker time to market. So that's on the internal processes. One example, there are many but this is one example with very significant outcome on the customer side. Um, one interesting example is our tunneling ah business um and uh, actually it's from the classic machine learning processes. I can give another example from the generative AI maybe um on this uh specific business, a tunneling business, the team targeted a very specific narrow workflow and sometimes this is critical for success. Target a very very clear workflow. Uh and the goal this workflow was customers uh perform 3D scans after every blast in the tunneling process. So with AI based feature extraction and object classification customers were able to reduce analysis time by 90% 90 from about 30 minutes per scan to 3 minutes per scan. So this is not just efficiency, it's a dramatic change in how they operate day to day. Just imagine how many scans are needed in a tunneling project and the potential saving um, um with this kind of technology. And then I promised something on generative AI maybe so um, an example from this genre um maybe come from the architecture and design business Trimble. So SketchUp is our 3D modeling tool and the SketchUp team introduced last year uh SketchUp Diffusion. It is a generative AI rendering tool. It really creates high quality rendering from A3D models based on a text prompt. And this takes seconds instead of hours. What maybe really excited me and I'm an architect by profession, what excited me um, in this specific use case um is how it expand the focus and the value proposition from uh, just um productivity gain ah to improved creativity and client communication so designer can explore designs very quickly, instantly communicate with the customer and with visualization almost on the fly. And it's really democratizing the process and changing the nature of design conversation.

Speaker B: Yeah, that's really interesting because you know in the old world it was always around automation and so on but creativity wasn't part of the equation. It was almost like you know how can I take what I'm doing, just do it faster and do it without people. But I think now the genai uh, I would agree with you is uh, kind of does certain things that just weren't even possible uh before. So look, let's go on the other side of things. So I think this was actually really really insightful and uh, very uh inspirational and positive. Now at the same time MIT came out with this report and said 95% of all AI pilots fail. And that almost seemed like a wet blanket on the whole Gen AI thing. I'm curious what you make out of this. Based on your experience. Can you understand why a report like this comes out? That's maybe number one. Maybe you have some idea of like how people can avoid that and then also maybe tell us in the opposite of you know, here's easy gains, almost like here's maybe the things. If you apply AI gen AI to that and you want to transform that based on your experience, you found it actually really really hard. Like certain areas where you feel it is fundamentally hard. I'm curious, I'm trying to get both sides and get people also an idea where things can go wrong.

Speaker A: Yeah. And I saw this article from MIT and um, I was surprised a bit but I think I can identify the reasons. And it is a similar pattern to what we experience when we introduce for example mixed reality to the construction market or robotics. It's really about setting the right expectations which should match the uh, technology maturity state. That's the first thing you need to make sure both internally and when working with customers. And if you don't do it you get this kind of hype which turn into disappointment. And it's not that everything is working perfectly with our initiative. Um, I will not maybe describe it in terms of failure but um, you know we embrace a fail fast culture at Trimble. So it's really been a series of exciting challenges, some of them successful, some of them are less successful that have uh, evolved along with the technology. So maybe kind of to give you a bit of color here in the early phase when we just introduced the technology, the challenge, um, and I'm talking about internal and customer facing uh, interaction. The challenge was building awareness what technology can do, what it can't do, uh, what our expectation for the future. So side by side with a uh, long term vision you want to also to demonstrate and show the low hanging fruit. So we really focused internally on upskilling, demonstrating value in very basic workflows and showing that AI could make real difference. And we made two decisions during this critical introduction or early phase. The first, um, and when we look at the adoption inside the organization, the first was uh, to slice the community based on where they were in the adoption journey and tailoring the communication for each group. The last thing you want to get as an employee is to be bombarded with content which is not related to you, which is not aligned with where you are uh, in the technology uh, journey. Um, and just to give an example if you are not, we can see who is not using the technology at all. Um, and it would be good to understand why. Is it because logistical reasons cannot access a license? Is it because, uh, not enough management support? We can address those challenges, but we need to understand what is blocking you from adoption and then take it to the other side of the spectrum. Those who are using the technology every day, those are champions, those are excited with technology. They can be our driver for adoption. They can share their success story and help us build the confidence and the best practices to be successful. So communicating the right messages with a different group by slicing the community to those kind of groups was critical, uh, at the beginning. The second decision, uh, which was highly important in the adoption of generative AI was to develop our own internal tool, Trimble Assistant, the tool I mentioned here, essentially eating our own dog food to build confidence and credibility, to understand the limitations and there are limitations, um, and to see where we can uh, um, take this technology in order to support our internal customers. So as technology matured, the next phase for us was about becoming more intentional. As I mentioned earlier, that meant uh, consolidating all those early initiatives which is again it's a challenging thing because it's your baby. Why should I give up uh, for the, you know, and change my uh, uh, technology vendor or the process that I developed to something else. And the reason is that we want to make sure there is alignment and intentionality behind the development. The technology is now more mature. It is also about setting clear expectation. Now it's not an early phase for adoption. Now we want everyone to use AI. We want you guys to report back, what did you find out for the good or for the bad. And then it will help us align resources and fine tune the upskilling efforts, uh, to support um, our strategy. And again one of the, in this phase, uh, the phase that we are in now, actually one of the most important outcomes uh, from this stage was to develop or build our agentic AI platform. So providing foundation, uh, the plumbing if you will, that is required like security and access control and LLM access and sandbox environment or responsible AI. All those uh, are put together into the platform so the businesses can actually focus on building their vertical or domain specific agent without needing to recreate or focus on things that the platform will provide them. This is basically how we scale the use of AI. And now looking for the future, the next challenge for us is really adopting agents at scale. And I think the challenge here is to move from focus, uh, on productivity improvement which are Critical but not the full picture. And to focus on how we fundamentally rethink workflows. It's not anymore just about doing the same thing faster or better. It's about transforming how work gets done with this technology. Mhm.

Speaker B: I mean that's a central thesis of the book actually we wrote uh, the new Automation Mindset, the innovation uh, mindset. Uh, I totally agree. I'm actually curious. The, I mean you made a good point earlier about predictive AI, the machine learning and then the generative AI. And I believe if I interpret that correctly, uh, correct me if I'm wrong but I feel like you need both. Right. It's not like uh, I don't think Genai on its own can solve all problems. There's certain things that ML and the predictive AI is actually very suited for. Um, and then I would say again, this is just my perception, I might be wrong. But I felt that with the machine learning, the predictive AI, the adoption at scale, I'm not sure I would give it maybe a 5 out of 10 in the world. I don't think it is fully being adopted. I think it's been kind of a little complex. Uh, adoption journey. Uh, my sense is that with Gen it may actually be easier. But I'm curious about your opinion. Where do you see that and maybe what are the factors that are ah, maybe different for the adoption of scale of both AIs?

Speaker A: Yeah. Um, my view on predict. And first I completely agree the two technologies work together. I mentioned a continuum but actually it can also be reviewed as um, a loop here because what you create, the output you create with classical LM processes, um, can be used by the generative AI to create new content which can in turn feed back into those classic machine learning processes to improve and to help with the training. So it's really um, those two technologies working well together. We do see on our side, predictive AI is a solid. There is no hallucination. It is delivering exactly what you wanted to deliver in a very consistent way, in a very trustable way. So it's a powerful technology which our customers are actually using every day in the geospatial space and also in the construction environment. This is already there and it's tightly connected to our hardware ecosystem. Uh, because output from the hardware is actually being used and processed by predictive AI. With generative AI it's easier to adopt because of the easy interaction and the natural language processing and um, the excitement from the um, cool aspect of the technology. I think we'll face different challenges There the monetization is one unlike um, predictive AI where the cost was known in advance, uh here with generative AI, um, the cost is growing all the time. And there is some paradox here. If you are using the traditional monetization models, uh, like subscription, the customers who are your best customers, those who are using your solution the most, are the most costly customer for you. Because every action in generative AI costs you money. It's a computer um, cost, um, and those who are not using just sign on a subscription. Not using your solution, you gain uh, the maximum revenue from them. So it's not a healthy balance. And this monetization challenge um, I think will evolve into, and we already see it in the market to a more um, or to adjustable monetization models, um, consumption based or premium based or any other solution that um, would be a good fit for the market. And so this is something we need to resolve in order to really scale the adoption of AI both internally, uh, on our side and on the customer facing solutions.

Speaker B: Yeah, yeah. The crazy statistic I saw was when if you compare like the calories that we have to eat as people to think with our brain and the energy, the electricity that goes into an LLM, I think it's like a factor of 1000 or something like that. I mean we human beings are still way more efficient energy wise, uh, with what we do at the end. That's what reduces the cost and that's what I guess is not good for environment and something we got to probably solve over time.

Speaker A: Yeah, and I see it's certainly being improved over time. We had discussion with the major vendors last several weeks because this um, environmental aspect is uh, very close to our values. Uh and we want to make sure that we are doing the right thing with AI and what we heard from the vendors is uh, improvement in processing and optimization which really dramatically reduces the compute needed for um, the generative AI processes. And, and the other side of it if you will look what you gain out of um, uh the technology, what are the improvements you can bring, whether in optimization of shipper carrier, um, kind of relation and the optimization of the pass of trucks or heavy machinery in the field, um, uh, or optimize processes during the construction. The gain from the sustainability perspective you gain out of the generative AI, uh can really compensate for uh, the impact of the compute. And we see tremendous uh, value here. Small you know, small improvement, uh, which really impact um, uh, the industry.

Speaker B: No, totally agree. Uh, I mean, so aviat. I mean first of all I wanted to Say thank you. Uh, I think this is, I mean it's been super interesting to me and I'm, I'm sure I can talk on behalf the whole audience. I think this will help people really think through this AI transformation challenge. Maybe to bring this to closure. Um, just wanted, we talked a lot in giving people examples and ideas and what to think of. But if you could give them some concrete advice, maybe to a CIO in front of you, one thing you feel they should be doing now, that in two years time they're positioned properly so that they're going to succeed with their AI transformation, what would be the one

Speaker A: thing I would say the highest priority, uh, for CIOs? And by the way, we just appointed a new CIO for Trimble, um, Jim Palermo. So uh, he should answer this question maybe and I don't want to uh, step on his uh, toes. Um, but I would say that um, if I need to give advice, data first with ah, interoperability and governance in the core, that's certainly uh, the baseline and a must to have in order to scale AI and scale mindset. So think moving beyond pilots or single process automation to system level automation. And if you will think in this kind of mindset, I believe that you will be able to get your system ready to scale AI

Speaker B: Aviat. Thank you. Thank you so much. Now I think this has been super insightful and uh, really want to say thank you. Sure.

Speaker A: Thank you very much for having me. I enjoy this discussion.

Speaker B: Thank you so much for tuning into today's episode. Please make sure to rate today's episode. Leave us a comment with your thoughts and subscribe to the show so you will never miss an episode. I'll see you next time.