Google’s AI Comeback, Enterprise Agents, The Real Path to AI ROI — W/ Promevo CEO Karthik Kripapuri
Channel: Alex Kantrowitz
Published at: 2026-02-11
YouTube video id: hgpvEvuAugE
Source: https://www.youtube.com/watch?v=hgpvEvuAugE
Power companies finding value today from cutting-edge AI models. Let's talk about it with Promevo CEO Karthik Keyur Purri, who joins us today in a conversation brought to you by Promevo. Karthik, great to see you. Welcome to the show. Thank you, Alex. Thanks for having me. So, a little bit about Promevo, you work extremely closely with Google's AI technology, which has been like a pretty good place, I would say, over the past few months to be. So, I'd love to hear just from your perspective because no one watches this as close as you do. Your Your company works extremely close with them. Um what have you seen over the past few years from Google? And what do you Do you think that they're building a lead over the compet- the competition? Well, I hope so for the last part of it because we are we are all we do is Google, right? I mean, that's all we're focused on and there's a good reason for it. Um kind of probably should take a step back, you know, for me as we look at it and we just came off of our annual kickoff with our company and way we look at this is where the technology has been in the market and what Google has done. I mean, what like 2023, you know, late part of 2023, Chat GPT took the whole by storm and it's what's big novelty act. I mean, by the way, it's real, right? I mean, not just novelty, but it's a novel for for the end user, right? From a com- commercial user um adopting this technology. And yeah, Google probably got a little caught flat-footed at the beginning, possibly, but the pivot that has happened in the last couple of years, I couldn't be happier as a as a partner of Google and the support that we have gotten from them in uh driving this with the clients as well. And the technology stack itself, right? I mean, so uh you know, the the approach of being open uh in you know, you know, the technology itself is open, it's secure, and the hard pivots into model like, you know, um you know, you know, Gemini Pro 3.0 that where we are in right now and we're just giving multimodal opportunities, it's been it's been very um rewarding both for us, me as an operator of a company, as well as our clients as well, right? So, I think it's you know, the overall the biggest thing AI for our clients as well has been a big thing for us in terms of providing security in our full stack. So. Yes, so just to explain what this means. Basically, Google will come up with these new models. They're not going to be able to help the whole world implement them. So, what you guys do is you have built expertise within the Google models and then you will help the implementation on the business side. >> That that is exactly correct, right? Like the last year alone about 18 months ago, right? We started work partnering with Google to provide a lot of adoption services, especially around Gemini for Google Workspace. We have done about 250 to 300 workshops at this point, which which is pretty incredible from the investment standpoint for Google investing in clients and us providing the service for them. And we see the adoption like all things in all things in the world at different scales with our clients, but there is a really good reason for that, right? I mean, but we are seeing the adoption hitting mainstream and, you know, the demand for it increasing day by day. Yeah. So, yes, we know that Google was a little flat-footed at the beginning. Yes, everyone knows, I think, theoretically that things have gotten a lot better. But from your perspective, just talk to us about the tangibility of this stuff and when we hear, "Okay, it's gotten better." What have you seen on the ground? The models have gotten better at handling complex queries, they're hallucinating less. Talk a little bit about the progression. I think the progression the big progression, right? It I think is going from a novelty, like you know, "Hey, give me answer some questions, do some basic stuff." To where we are right now, I feel like with with the advent of agents at this point, where you can do agent assembly lines. The models have gotten very good, you know, you know, very close to being good at discern discerning. There's still some work to be done in terms of hallucination. And but Google's done a very good job of grounding, you know, the models, providing visibility into it. And that's the reason why we are super excited to partner with them. And look, there's still some work to be done here, right? I mean, you know, like and and this is not just a Google only race, right? I mean, today the models are at a point where next, you know, you're going to get announcement with OpenAI. There's going to be, you know, there's going to be leaps that are happening back and forth, which overall I think like like one of the things we like to say internally is this is as worse as these technologies is going to be. It's only going to get better as things go. And a lot of this still we feel like is grounded in the fact that organizations still struggle with data integrity and that is fundamentally what is the primary blocker, right? In terms of like, you know, really AI mo- I mean, models and agents working and reaching the full potential of this point. That's what how we feel about it and that's where we come in with our consulting services and provide that help for our clients. Yeah, so what does that mean that the data integrity is not there and that's the main thing holding them back? I mean, the big thing, right? I mean, I mean, I'm a I'm probably aging myself. This is the same thing that happened in the 1990s and, you know, the the the digital transformation. The biggest blocker that we see with digital transformation is where data is very siloed within an organization. It is not really really available. And then who has the proper definition of the single source of truth uh for that data that exists in organization. So, once you're able to align as an organization on what that is, models can get better about grounding themselves and providing that level of uh insight and analysis that I think agents can then act upon and that provides better outcomes for our clients. And that's what what we see, you know, uh when in even internally like as a as a as a COO at Heartwell, I find that to be the biggest thing that because we want to be able to act on it in the most uh in an efficient manner possible. So. Okay, and then one more question about this. You said they've gotten better at discernment. Yes. >> What does that mean? That they're making better decisions? Well, they're getting better decisions based based on the information that's available, right? And you know, and I would say that is you know, especially with Google uh you know, where with you know, with with the models that gotten very efficient, very you know, and then there's a lot of transparency that's in there which allows you know, have a visual into how the thinking process that happens in that. We are able to see it in you know, inside what is happening, what the decision is and you're able to tune how much of the hallucinations, you know, you can tolerate, what you don't want to tolerate. I'm I'm obviously, you know, speaking in very basic terms here, right? There's a lot of you know, you can data sovereignty is a big thing so you don't want it to go outside, you know, like things that you don't you know, want to base decisions on and that level of you know, um you know, control that it's given to the user has gotten them at least the Google models gotten be very, very effective. At least internally that's what we see and this is what our clients are saying. So. Okay. Great. And there was, you know, it's interesting. A couple months ago there was this discussion is AI a bubble and is there any enterprise value uh in artificial intelligence and now it's like the entire conversation has shifted to what software companies can actually, you know, survive this moment. And I think that that to me shows that there's there's been this moment where there's companies like yours have taken the AI models, used the improvements and translated that into real business outcomes. That's happening now. So I just want to hear your perspective. When the model gets better, how does that translate into your ability to do more with a with a customer? Well, um I our CTO John Pederson and I we each spend a lot of time thinking about this and talking about this and one of the things we believe, right, the idea of the AI platform, right, is getting closer and closer. I mean, I know you know, I'm going to quote Satya, right, who talked about, you know, the idea like a few years ago he talked about the idea of, you know, these the platforms like CRM platforms that sit on top of data that's existing in the organization and that's what we talk about AI platforms that are on top of it and how we work is going to fundamentally change. You know, in the I really I think we're and I'm not trying to like you know, scare anybody to think that PowerPoint will go away, but you are seeing PowerPoint the people whose reliance on PowerPoint is less than it used to be, right? And you know, I we feel like the like as a as an operator of companies, most people we don't want a plethora of AI tools out there, neither do we want to have a IT sprawl with a whole bunch of, you know, platforms that exist. What we want is the ability to action and do our work in a you know, with the data that's already existing in our organization and how do we interact with it and how do we make decisions on it? And I feel like that's where we are right now and you know, I don't know what the time frame is, but I feel like it'll be a day when I can just ask what I need to do and I don't have to go check my email, I don't have to go check my CRM, I don't have to check my financial systems. I'll be able to interact with the data that exists in our organization. Right. And so then as the AI models have gotten better, that's allowed you to do more of that or what exactly have the model improvements translated to? Like one of the things I will say here, right, an example of that would be your you know, we were talking about before about how we have a you know, internally to our organization, we are asking our people to be 30% more effective, right? I don't I never use the word efficient because efficiency means like all kinds of weird things and people are like, "Oh, what's about my job?" It's not about it, right? I mean, like the idea is, okay, if you got 30% of your time back, you know, let's take our finance team for in our organization. We know they cannot take PTO the first week of the month because they're closing. Right? It's impossible for our CFO and our financial controllers and our FP&A team. So, what we have challenged them is how can we work in a way in which a lot of the tedious mundane task that's working in our own bespoke billing system that we have in there, how can that be automated so you can actually take PTO? Right? So, that to me is what, you know, the agents that we are building and you know, it's sitting on top of Looker or BigQuery, you know, that can action the mundane and gives time back. Whether you be use it for PTO or taking time off that much deserved or do something else interesting in your career is what we talk about when when we talk about the gains that we want to make in people's growth and in our company's growth. That's what That's what internally we have challenged our organization to be good at. Right. And that's working today. Well, we we are we are in the journey. That we starting we have started on this one that, you know, we we have we have kicked off a couple of projects at this point and more mostly focused on, you know, people ideas coming from our organization not top-down. So, that is one example I gave you about the close process. How do we get that tedious done? One of the things that the other thing is we one of our co-workers she brought up the idea of how do we automate a lot of RMAs. We do sell a lot of Chromebooks for education and you know, we are a full stack Google partner. And one of the things that happens is Chromebooks sometimes don't work which you know, people don't want to hear, but you know, it happens, right? Technology doesn't work. So, we she came up with an idea of how she got herself first of all GenAI certified, right? And then she wrote the prompts to kind of figure out how do I automate this and then working with one of our practice leaders, she wrote an agent to figure out how to automate the process and we're about to launch it. So, the idea is can we get eight hours back for our plan in a on a weekly basis? That's the whole idea and that it's working. Yeah. So. Okay. All right. And I definitely have more questions about that, but um I want to continue along our our line here talking a little bit about um what Google enables you to do because you said another thing that sort of uh made me perk up, which you said that the the model or Google models, you're able to tune them in a way uh that you might not be able to do elsewhere. That if you you can have like a dial basically that says, all right, less hallucinations, but maybe a more rigid answer, uh more you know, potential for more hallucination, but more creative answers. That's what I'm I'm guessing. And uh one thing I've heard is that maybe that that Google is not a a black box, but maybe a gray box. So, talk a little bit about that and how that enables you to implement the technology better. Yeah, I think, you know, when you're when you're interacting Vertex, you know, Vertex AI, which is at the core of what how Google, you know, exposes the you know, the thinking, you know, to us, you know, like or to any of the users in there, you're able to dial back and, you know, how much, you know, broad that you want the model to be or how narrow you want it to be focused in the grounding that happens. All of that to me, you know, you want to the whole idea of using these models, right? Is one based on developing trust. And what is the model using to make the decision it's it's uh you know, it's going to give answers uh based on. And what we're trying to figure out is the more transparent it is, the better it is for the user and for the model to even understand the context of who you are and how you work and all of the stuff that happens in there, right? So, I think that is what I mean by that. I mean, like the idea of like with Google has been has done a very good job of it in the last year or so. What we are saying is, you know, first of all, one is making sure that the data that you are, you know, inputting into the model, it is not used to train models globally, right? Which is the most important thing. As, you know, as we develop we are on we are we develop our own SaaS platform. The last thing we want to have is, you know, any code checks that we do is opened up, as you've seen in the news that's happened for a couple of big enterprises. Like the IP indemnity that Google's providing, it's I mean it's impressive. And, you know, that's that's Those are the things that we are When we interact with CTOs, they are talking about how much they appreciate and the confidence level is growing with what Google's providing in terms of how to interact with the models better in a more safe and a secure way. Can I ask you, like, why Google versus open source? Well, first of all, like to me, uh it kind of goes back to, you know, the idea of like thinking about managed services, right? Like Like one of the most 99% of the companies, I would say, what they need is probably need to consume this as a managed service in my mind. And the full stack of Google's investment, right? If you If you look at the full scale of it from infrastructure, models, and open I They have ability to have interact with other models, the investments that they've made in, let's just say, in developing agent A2K, which is, you know, open to everybody, right? And agent-to-agent uh integration, all of that, you get almost the best of both worlds. Like, why recreate the wheel with the infrastructure that and spend that you don't need to do? And the work has been done, so consuming this and you get the latest, you know, learning uh the LLMs that you are already getting, it's from a managed service provider like Google that's already invested in that, you know, and then and you can calibrate it the way you want to consume that. To me, yes, there are some bespoke cases where you, you know, where there's You know, we have a couple of clients who are building their own models and we are totally uh it makes sense for them. But 99% of our other client with the clients that we have are probably just okay with using just, you know, what I'm not quite off-the-shelf, but, you know, what is generally available to, you know, for consumption. Yeah, [snorts] we had uh Arthur Mensch from Mistral in recently, and he was also talking about how AI is effectively a managed service, that it is a an intense technology to implement, and you could be a technology company and sort of, you know, take a model and customize it yourself. But, for most businesses, the way to consume it will be working with some other company as an implementer and an optimizer, and having them build the AI for your use case, and that's where you get ROI. I I would agree with that. I mean, and I think, personally, I feel like this is one of those buy versus build decisions that face us all operators in the world. Yeah, I mean, like I We know, when we talk internally about investment decisions, what we talk about is, is this something we want to be good at, right? Yes. And if you want to be good at, there's two ways to think about this, whether do we have practitioners building it for us, or do we want to have an experiment and try to build this ourselves? And the time to market and time to effectiveness, all of that stuff, you know, comes into play here. And that is That is a tough call to make, right? But, most cases, I think it's made tougher by people who want to say we are probably the We are We don't need to be We have to learn, continue invest in it, continue nurture it. That takes a whole lot of thing that takes maybe away from the core value of what mission of the company feels like to me, right? Yeah. So, then, tell me, where do companies get ROI, or when do companies get ROI from this versus when do they not? Like, what are the setups where, you know, a company goes full steam ahead and says, "All right, we're going to implement AI." And they actually end up realizing a return on their investment versus when do they not? Because there's definitely been, I think, the people who aren't getting value out of it are louder. Yeah. Maybe it's a competitive disadvantage to, you know, shout to the world that this is working for you. So, talk a little bit about the the divergence here and and where people get the most value from it, or when they're able to exploit that. I I think, look, I think, um, I start with We said we've done like 250 workshops, you know, like what adoption we always talk about, right? Whether it be take, you know, we find the best use cases, right? Where the best success clients have is when they start small. When they have a clearly defined KPI in mind on what is it they want to accomplish, right? The two examples I gave you are tangible. And we have actually have said, "Okay, we're going to start small like an RMA automation." Or you know, a finance close operation. But there's a whole bunch of things in there that you can talk about how you want to automate. So, when and then there's also buy-in from the top down about the success of what, you know, what, you know, something needs to happen. The biggest thing I will also say here that we see where ROI works and not work is, you know, the opposite is also true where we don't see ROI is when you try to boil the whole ocean. We're like, "Oh, we're going to be an AI-first company." I don't know what that means, right? Uh when when somebody says that, it sounds great in a board meeting or in a, you know, press clipping maybe in a release, but I don't know what AI-first means. I mean, unless you're a software developer, it's fine. I mean, you want to be grounded in how you develop your your software. But the idea is to start slow, be very clear, and then, you know, ensure that you know, you understand what outcomes you want to drive in there, and then expand to the next use case, and then, you know, and then be methodical about it, right? Like all uh you know, transformation work, this is also at at the core of it, right? I mean And the last thing I'll also say is it's not, you know, while I think we are there as consultants to come and help, the SMEs who run the business are key to, you know, making sure that the use cases are clearly defined, and we can help facilitate the technology portion of it, how the technology works for the business operations, but it's important that the key stakeholders are absolutely engaged in, you know, driving the, you know, the results that we need. So, yeah. That's That's That's how we follow it. And uh we try to punt when we we don't see the buy-in and we we try to recommend it gently, but with the clients who do the right way, you see tremendous amount of uh returns. So. Have you seen the the percentage of implementations where clients are getting ROI go up or have you seen even uh cases where it used to be pilots don't go into production, more of those pilots are going into production? Yeah, I mean so I can well, I can tell you a few examples of where this has happened, right? Like Bold Bond, I think we have some personally we have a few personal with them and they're a great example of a company that has, you know, started out, you know, in Google Workspace Gemini for Google Workspace and they saw 70% adoption, right? Which is the first key metric that culturally there's a massive of adoption across the organization. Then they went to the next use case about how do you visualize, you know, provide visualization of products on the website, you know, so that takes freeze up more time to sell and then then they moved then we're helping them automate a lot of their, you know, um order processing and everything like that. And they are the great transformational organization that really understood how to start it and then how do you expand it because and that's why I would say is probably what more of 2026 has in store for us like that. I mean, because that that is if somebody can replicate that, right? Over and over again, you're going to see a lot more outcomes in there. And the other thing we also is a good barometer in my in our thing is, right? We, you know, like you know, I I can one more thought on this one about outcomes that we need to drive here. But when we see massive engagement after we do a pilot, right? when when it's not limited to like the managers or an executive organization, when you see the attendance that's happening and then the usage build up even using Gemini for Workspace, right? You know culturally that is getting adopted because people do not want to do the boring tedious stuff over and over again, right? And then when that catches, you know, fire, it it's going to, you know, really adoption is get driven. Innovation starts at the very edge, and then it's nurtured and fostered there, right? So, that is that is how we think about this. This If you want to use cases I can give, but that's kind of, you know, that's a good poster child of how we want to see this roll out. So. Yeah, very interesting, especially the range of use cases. Just in that gold bond example, AI for workspace, order optimization, um you know, trying to visualize products. Do you have a sense Are the models today sort of equally good at all different things, or do you have a sense as to like what the models are best at, and where they could improve? Look, I think the models today I I will I'll zoom out and give a general answer of what I think the models do really well today, right? Maybe that's probably specific, so I don't upset any product managers anywhere. But >> [laughter] >> But what I would say the, you know, the models are very good at reasoning, understanding of given context, and it's it's it's they're very good at that. Where I think they can be a little bit better is and, you know, it's obviously we want always limit hallucinations, and, you know, it's the idea of becoming a little bit more uh It's a just like people, I would say, is becoming more self-aware, right? As in, you don't need to give an answer. If you don't know something, say you don't know. It's something we tell our kids [clears throat] and, you know, people meetings, right? So, don't feel like you The idea of like you have to provide an answer needs to probably be minimized a little bit, and that's probably where I think the models in general all probably need to get better at in my mind. And do you ever see them say, "I don't know"? Or is that still They're still kind of >> I when when the context is uh off, they're better at it. But more often than not, it tries to connect dots that don't need to be connected in my mind sometimes, yeah. >> Yeah. It's like the the friend who's read everything, and and does seem like they always have a response. Maybe they don't, though. >> Exactly. Um And I I I think that >> I know how you feel." It's another favorite example of that. >> [laughter] >> Yeah. Oh, yes, I know how Yeah, sure you do, AI model. Okay. Uh I said I said we were going to get back to this. I want to talk a little bit more about what's going on inside your company. So, you actually have an initiative where you've tasked a good chunk of the company to create their own agents. And we're kind of coming at this moment where um where we have this surge of people building their own, some on uh you know, Mac minis and setting them loose in the world. Uh and so, I'd love to hear an update from you about how that's going. Well, um what we did actually is we we have not tasked people to do build agents, okay? What we have told them is be we are going to be grounded in AI because if you're going to be consulting with our company with our clients, we want to be lead by example and show use cases that we actually want to show them because what we want to show is have our clients see a first meet with our CFO and say, "How do they, you know, close books and how do we integrate?" Same thing with our finance team, our client success teams, all of that the visualization, right? So, what we have really It's one of the ways that I can think about how do you accelerate the learning curve because most people, right? This is new to a lot of people including our own co-workers internally. I mean, you know, I you know, so the way we are trying to do this, right? Like a couple of years ago, you know, one of the things that we do our employee feedback with our companies annually and um you know, one of the things people are like, "Oh, you want us to get all certifications and everything, but give us time." So, on Friday afternoons, we don't have any uh meetings. So, we have block learning blocks. And [snorts] I will say we're I'm very proud of the fact that people actually have taken that and then they most of them are GenAI certified in our company at this point, right? Google's uh courses and uh certifications that it provides. And what we have seen, right? The I So, we asked people to like, you know, you don't have to build this yourself. What we want is you to think about the things that the use cases that you go through that you think this technology now can be applied. So, we actually did in I you know we had our company um outing in Austin and most of our company ended up doing a hackathon along with our developers and our practice leaders and they worked through a few of the use cases that we have and then we are going to use the next few months to go and implement this just like we tell our clients how to implement it, right? Look, like all things it's a question of prioritization, right? And we went through three of the use cases in Austin and there's about 20 more that's behind it, right? And so but the coolest thing about this is you now realize there's so much opportunity to be unlocked at this point and and the people are thinking about it the right way and you know about how to solve things that people usually don't like to do on over and over on a on a day-to-day basis, right? I mean, you know, like I don't want to create a power you know Google Slide every time and I want to be able to have a starting point that just it's more conversant with us. So that is one thing that we have done internally. The other thing I'll also say here is it is you know this probably you know not that much of surprise, but we are a software provider as well, right? We we have G Panel which provides, you know, workplace automation and management software for all of our workspace clients and we use you know generative AI to actually accelerate our road map you know our code development in our software development life cycle process and which has led to like acceleration and it it started last Q4 and we're seeing the productivity happen a lot. I mean, so you know, those are the kind of things that we see a lot of that's happening operationally as well as our own software development you know on both of them it's been very very useful and productive for us. Yeah. Okay, fascinating. All right, let me ask you one last question here because we've talked a little bit about it, but I think we should go a little bit deeper. Sure. You're talking about how you're working to implement AI in workspace and all these different these different knowledge work tasks. How do you think AI changes knowledge work economy? Hmm. That's a that's a very broad interesting question. I mean Can you elaborate a little bit more when you talk about when you said the impact, what do you mean by that? Like >> Yeah, no thanks thanks for clarifying. I think that like I wonder how our work day changes, how the workplace changes as we include more and more AI into it. We won't hold you to it but I you know we can dream a little bit here. >> I think I I kind of I've been touching on this throughout this conversation, right? >> Yeah. Um Look, I think the the doing more with less is not going to end anytime soon, okay? I mean, you know, even the weird ground you know, we're hiring but you're also seeing some uh I don't think it's all AI related, it's people related personally, but that's a different conversation all together. But within the workspace, right? Within the work day, how it changes is this for me. So I'll can give you a tangible example of this. We had an offsite in Q4 to start our planning for 2026. Right? We knew what our strategic objectives were and that you know, you've been enough board meetings and you can say, "Oh, now we're going to convert that into some OKRs and it's going to the wordsmithing we actually used Gemini to actually feed this into it. It we actually fed this into our plans for 2026 and the OKRs by department aligned with our overall strategic objectives for a 3-year time zone and the 2026 plan, we did this in 1 hour. Amazing. So it is something that >> is that is incredible, right? That's tangible and we actually showed the board our how we did that that process. That is pretty cool stuff, yeah. It is pretty neat, so. Yeah. You know, it's interesting cuz there are some organizations that are out there that are like, "Do we do this? Do we not?" And I think being someone who's leading a company who is so far forward in this you You really have to wait for permission and that must be freeing. Yeah, I It's one of the reasons I work in a smaller organization, right? There's not we we can do what we are The one advantage that we have as a 100% company, right, is we can move fast and we can adopt technology faster, and uh it's just what That is advantage we have, and we want to be we always never want to lose that. And this is such a transcendent uh you know uh thing that is happening in our in our in our world. I mean so awesome to be part of it, right? You know, for all of us. I know it's scary, but it's also pretty exciting that, you know, you can um take advantage of it and also use the experience we have as an organization from the the the tribal knowledge we have and use it to accelerate this fast. That's how we look at it, so it's pretty exciting, you know, I think I'm also a I'm a nut for change, so. >> [laughter] >> Well, you have to be in this industry. All right, so >> we are a growth equity backed company, and that's all that's what they pay us to do, so this is great, so yeah. Totally. If people want to learn more about Promevo, where do they go? Well, there are a couple of things, right? You can go to our website, promvo.com. We and then we're on follow us on LinkedIn. We we you know, we provide a lot of blogs, and then you know, and I love to do a plug. We have one of the best ISV um uh you know, platforms that we have for the Google Workspace. G panel.io is its own website, where we have great tools, ROI tools, and everything as well. And yeah, and our goal is to be integrating Gemini into you know, our G panel, so you can have conversion Workspace administration. You know, right uh with with our which is a Nobody wants to be writing scripts anymore, right? We want to be able to interact with it directly. The G panel already does it today, but you know, we also do you know, and then talk to look for our services blogs and our you know, we are we are pretty Our team provides a lot of it from our experience with the clients, so yeah, please do. Okay, good stuff, Karthik Ruppapuri. Thank you for coming on the show. Great having you. >> Alex. Appreciate it. Yeah. All right, everybody. Thank you for listening and watching, and we'll see you next time here [music] on Big Technology.