theCUBE + NYSE Wired: AI for CFOs & AI Leaders in Silicon Valley | Bhavin Shah, Moveworks

Clips
More from theCUBE + NYSE Wired: AI for CFOs & AI Leaders in Silicon Valley

Bhavin Shah

Founder & CEO

Moveworks

play_circle_outline Introduction to AI innovators and business model transformation

play_circle_outline Evolution of AI models from 2016 to gen AI

play_circle_outline Role of agentic framework in accessing multiple systems and domains

play_circle_outline Importance of quantifiable impact in AI adoption

play_circle_outline Balance between buy versus build decisions

play_circle_outline Momentum and metrics of Moveworks

play_circle_outline Improving AI Models Through Constant Fine-Tuning for Efficient Automation in Resolver Groups

Info
Transcript

Bhavin Shah, Moveworks

Bhavin Shah

Founder & CEO Moveworks

TheCUBE covers AI innovators in Silicon Valley and CFOs adapting to AI. Moveworks' CEO discusses the company's evolution in AI technology, focusing on efficient business transformation and ROI. The shift to transformer models in 2019 led to advancements in neural retrieval and semantic search, improving user experiences. Moveworks helped companies simplify processes and interactions as the world shifted to remote work. The focus on business model transformation and cost reduction led to the development of an agentic multi-hub reasoner capable of navigating va... Read more

explore Keep Exploring

What events do we cover and what topics are currently being discussed in Silicon Valley? add

What significant advancements have transformer models such as GPT-II and GPT-III brought to natural language processing and understanding tasks? add

What type of AI assistant is described as being able to interact with multiple systems to help users find information and take action, regardless of where the data is coming from? add

What is the difference between business transformation and employee productivity and why is it important to the average CFO? add

What is the potential downside of focusing solely on the idea that AI will deliver pure productivity gains over time? add

What are some metrics that Moveworks can publicly share about their company's performance and growth? add

What processes and decision-making strategies are used for fine-tuning models and updating experiences in this company? add

bolt Powered by CUBE AI

Bhavin Shah, Moveworks

search

>> Welcome back everyone to theCUBEs coverage here in our studio in Palo Alto for special presentation. theCUBE plus the NYSE wired community, which our open source contents are Silicon Valley west Cube Studio, and we got the East now new Access Point Studio in Manhattan at the show floor of the New York Stock Exchange. Of course, we cover all the events where the action is. And here it's about the AI innovators in Silicon Valley and also we have a special edition around CFOs and AI and how business model transformation is now on the table because of the economics, because of the cost reductions and the revenue, everything's changing and the business model transmission is the top story. Having shots here. The founder and CEO of Moveworks is on theCUBE. Thanks for coming on. Appreciate it. You're a neighbor right down the street. Thanks for coming on.

>> Yeah, thanks for making it easy for me to come by.

>> Well, we're going to do a lot more studio coverage. We're going to open up more flow on the editorial content, the video, so we want to definitely do the deep dives. One of the things that we've been talking all about, and you're in the middle of it, is the folks that were on the pre-gen AI wave hit gen AI with perfect situation. It's like having those right surfboard ready when the big wave comes. To use the California analogy. And then people who just trying to catch up. It's still early days, so it's not bad to get into the game, but you guys have been at Moveworks on the front end. And remember we started theCUBE 15 years ago when Hadoop was the big data wave.

>> There you go.

>> That thing was not as big as people thought, but it did set the table for Spark, Databricks and a lot of the machine learning stuff with data. It opened up the market a bit, but now it's blooming.

>> Yes, yes. We started in 2016 with a lot of the same goals and premise that we have today, which is businesses are always striving to transform themselves. They're always striving to be more efficient, they're always striving to give employees a better experience. And of course at that time, 2016, we had statistical machine learning models to use to take employees questions, their searches, their different actions they needed to take. And to then prosecute those and take actions within various systems of record and business services. And it was over time that we then adopted transformer models in 2019 with the Burke model being the first of many to come. And I think now that we are in this full-fledged gen AI world and agentic world, there's a lot more possibilities, but it still keeps coming back to how do you drive hard ROI? How do you sort of deliver business transformation? And that's what we continue to serve as our mission.

>> Talk about the days back in 2016. Because at that time state-of-the-art machine learning was unsupervised, supervised kind of dynamics. You had pipelining and machine learning, and you're very specialized field. Not many people except for the big banks, Morgan Stanley, JP Morgan Chase, and they're still doing it. They actually haven't brought in a lot of gen yet because of their risk framework. But they will. Agents are coming. What was it like in 2016? And then what was it like inside the company when the whole world woke up when they saw the visualization with ChatGPT? Because I think that was the interface side of that convinced... I mean, I was, remember, screaming at the top of my lungs on theCUBE. Like "Dave, finally people will get it. What we've been covering on theCUBE." Because insiders knew this was happening. Take us through that mindset of what was it like at the company and then what happened next?

>> Yeah. Well, at the beginning was we were referred to this space as natural language processing or natural language understanding NLU, NLP. And the idea was we could use models to do discriminative tasks. Ones in which you would sort of look at something that someone would write and you would figure out the intent and you'd figure out what the entities are and what the action was required. And we had a single shot reasoner that was based on the ability to go and take what you're needing done, go take those, actually come back. That was state of the art. And transformer models made that more sophisticated. And we were working with GPT-II and GPT-III.

>> Explain that needle moving moment for the folks that don't have the inside baseball. Because entity extraction, that was kind like open source stuff. You can do that. You can go in and look at how things were clustering in texts for example. But then what did Transformers do? What did it level up and what did it put away?

>> It allowed us to more quickly train these models on the different topics and domains that we were interested in having them perform. They could also be more fine-tuned for different use cases and they were just higher quality experiences. But still in the era where we were using them in the back office, if you will, they weren't being used to communicate directly with the user as in terms of a conversation. And so that upleveled what we could do there. It allowed us to expand to many domains. We were first focused on IT support and transforming the help desk, and then we moved to HR and facilities. All that was facilitated by the advancement of transformers. And we started doing neural retrieval and semantic search, and this made finding information easier. It allowed us to use concepts instead of keywords. So that was itself a big step function. And if you keep in mind, this is now 2019, so we started the company when productivity tools like Slack and Teams were starting to become all the rage. Everyone was using short form messaging to get work done. WhatsApp and WeChat and all these things were starting to become commonplace. And then as we moved into this next phase of technology, we were then helping companies really transform their user experience. And that's also when Covid hit. And people were like, we need to simplify what it takes to get things done around here, otherwise employees are going to not stick with the company. So that was sort of that era.

>> And you had enterprise customers at that time?

>> We did.

>> And so obviously no one could have forecasted the VPN usage at that time being a hundred percent. I mean all the risk management profiles went the hell on a handbasket. So then all of a sudden you're now make it better.

>> That's right.

>> Take us through the Covid period?

>> So Covid actually helped put, front and center, this idea that employee experience was actually a real area of investment and concern for the business. So you started to see the rise of VPs of employee experience, Chief employee experience officer. And these were roles that didn't exist before. But it showed that companies were keen on improving this. In fact, even on Wall Street, you started to hear public company investors saying to CEOs, we want to see your retention data, your employee survey results to ensure that you are doing everything you can to retain the talent that you have. So now if something took three hours to get done, in the Zoom world, three seconds, three minutes felt like an era. So people were really gung ho on advancing that. Plus everything was now driven through software. Your entire interface with your company was not about walking through the lobby and experiencing the vibe. It was about all these pieces of software and these tools. So that brought us into a lot of large enterprises who were looking to do that. And one example of that was Broadcom, where we started working with them five years ago around that time because they really wanted to continue to reduce costs but also improve that employee journey.

>> Yeah, it was interesting. You bring up a good point. There's going to be a historical view of Covid either made companies or broke companies. And if you look at, I mean we're a covid survivor. I mean the studios exist. You made a comment about this being theCUBE. Well, this is thriving part of our business model, because out of necessity we needed to make our studios work. You had acceleration because out of need in the enterprise and some people were playing catch up, Amazon brought up call centers with Connect. That's now being productized. So you're seeing some productization come out of companies who were just nimble and had to be surviving. I mean, that was really a game changer. So I have to ask you now going forward, when did you see the momentum with the Gen AI and the Agentic piece? And are your customers ready for that now? How would you put the progress bar on that and what made you ready to do that?

>> Yeah, so one other bit that sort of came in between that Covid era and the sort of advent of more popularized gen AI products was the hyperinflationary period that we went through. So with hyperinflation, everyone was focused on efficiency. So now you had this, Hey, employees need to have a great experience and this stuff needs to happen with less cost and more leverage. So automation became a big play. And so a lot of our focus was around how do you take actions within the enterprise, swiftly and adeptly, and so the users can self-serve? So then as these models were sort of brought to life with OpenAI's ChatGPT 3.5, all of a sudden they took what we were using in a discriminative fashion and turned it into a generative experience that we actually didn't have. We didn't have that sort of dynamic conversation capabilities. Interestingly enough, about a month before that came out, Google had their paper about the react architecture. And it was an insight into how you can start to think about LLMs to do planning, and how you can then use that to then observe and take action. And so we started to actually have that conversation ChatGPT comes out, we're like, "Hold on, there is a new frontier." And that's when we sort of made a decision to focus all of our energies on building an agentic multi-hub, multi-step, multi-shot reasoner that could then prosecute across more systems, more topics with more agility. And that sort of was inspired by this whole wave that we're in.

>> And what's interesting too, that is what the top state-of-the-art conversation is today. Not agents, but multiple agents, sub-agents, collection of agents across domains, different companies maybe. I mean this is a domain challenge, is the data challenge. What are the key things that need to get done from a requisite standpoint?

>> So one of the things that we have been trying to play a role in is we are a company level AI assistant, not an application layer, a level AI assistant. What that means is we are interacting with lots of different systems every single minute to help users find information that they're looking for to be able to take an action. And they don't care. Is it coming out of your CRM? Is it coming out of your customer success management system? Is it coming out of your finance system? Is it coming out of your inventory management system? We have to navigate all that. And it's only possible with an agentic framework, because this is one of those open-ended problems. Once you ask the question, I don't know the answer until I being the AI goes and checks, and it could maybe check 19, 20 different systems. When you say, Hey, what's the status of my backpack? What are you talking about? Is that a ticket in your ITSM system? Is that some purchase order item in your purchasing system? Is that some other thing that you filed with the marketing team requesting a backpack? Is that another sort of ticket? So the open-endedness can only be solved by you have an agentic framework that can go out, talk to those systems, come back, assemble the information, decide if that's enough to satisfy the answer, or if it's getting closer to the goal, it will then go out and seek more information, assemble it, and pull that together. So that's sort of the reason why people have been talking about this, but I think the hard part has been for many to actually fulfill that vision.

>> Yeah, I mean, I think what you just said, I mean if you put your computer science hat on for a second, you say, Oh, well, okay, how do we do that?" And then you put your IT hat on, "What are you talking about?" And you put your business hat on, you say, "What does it even mean?" So I think the point of CFO, AI summit we're doing here with the AI innovators together is the first time blending, Hey, this is going to happen. I mean, there's no doubt in my mind. I mean, knowing the tech and looking at the years on the trajectories, many cycles, and you guys are in the middle of it. Again, that's why I'm excited to talk to you, is that this will happen. Agents will be part of the human in the loop. It's a matter of time. So the question is that next year, fully steady state or is it going to take a couple more years? But there's no doubt, so CFOs have to be ready. Number one. Because this is business model transformation. You go back 20 years ago, it was consumerization of IT, make it usable, easier. Digital transformation was go to the cloud. And this other definition is there, I'm oversimplifying it, but for the most part it was kind of, I don't want to say boring, but compared to this, it's boring. Business model transformation is technical transformation, business transformation all at the same time. And so the data and the infrastructure got to be locked and loaded, ready to go, and it's not ready now. Well, that's my opinion. I looked at your opinion. Because you're seeing companies like Uber doing it. You're seeing JPMorgan Chase ready for it. Okay. They've spent 17 billion a year budget, I mean, nice customer to have. Uber has huge large scale. Amazon's got some capabilities. So if you're at large scale, you have a different take. But the average main street enterprise, they're retooling. I mean, they're definitely on it, but are they truly ready? That's the question.

>> Yeah. So a lot of what you said really resonates with us. In that, first of all, there is this difference between business transformation and employee productivity. And those two things are quite distinct and quite important to the average CFO. When ChatGPT came out, everyone was talking about employee productivity and wow, I can do this in three minutes. That used to take three hours. Easy-peasy and it's great. But it turns out that most CFOs are quite hard-nosed about the impact that they actually want to drive for the business. And when you save someone three hours a week, most CFOs will say, well, the only beneficiary is your dog in the back nine of your favorite golf course, because I still pay you a full salary for your work week. And so where a lot of the focus has been is how do you improve business processes so much that then it transforms how everyone in the company does X or how they do Y. The thing with employee productivity is when someone leaves the organization, that improvement leaves with them. Versus if you transform how employees update their OKRs, how they can check on the status of a project, how they can update PO, so you no longer have to call someone in finance to say, "Hey, I need to update this purchase order." This and that. These things actually systematically create a lot of cost for organizations because you have people interfacing with other people to get these small tasks done. Now, to your point, who's ready for this? You have the biggest companies in the world who have potentially the resources-

>> By the way, just to interject on the productivity, I totally agree with you. The impact is the critical thing. How do I quantify the impact. So if I'm getting three hours back, I got more beer time with my friends, I got more time with my dog, what else did I do? That's not quantifiable. There's just no tracking of that.

>> That's right.

>> Where is that?

>> And there's a term for this. CFOs refer to it as weak ROI, versus things that they can be tracked can be brought back to staffing levels and true change and processes. That's called strong ROI. And I think that's what people are now judging a lot of their initiatives around. Because last year, '23, a lot of hype, a lot of excitement. People were like, let's do a demo. Let's show off what is possible. Now in '24, we're getting the call because people were like, "I got to deliver something before the end of the year. I've got to deliver something by next quarter that actually shows us as being more efficient."

>> Well, let me run something by you and then want to get into what you guys are doing in more deep way. And of course we can do a follow-up. Since you're right down the street, come back in. There's producers of AI, people who make AI like you guys, you have people building AI, playing with the transformer models and innovating with it, but kind of having fun with it. But you're delivering AI to people. Then there's consumers of AI like theCUBE. We have our own AI cloud we call it, but we're consuming other people's AI to use our data and our workflow. So then there's the data providers, which is us. So if an enterprise has the data, now they're kind of a hybrid. Once they start doing some AI, they might do a little bit, put the toe in the water. But if I have data and workflows, that's intellectual property, that's what we think. That's what we're saying. If I have that and you come to me and say, great, I'm bringing you AI, I consume it. It's a simple concept. So now I have to be better at what I do, not get more time. So the strong AI angle there is, okay, I consume AI and I got to be doing something demonstrable. That's evident.

>> That's right. That's right.

>> That's obvious. Which is for us, do content faster, do more volume, more accuracy. That's our business model.

>> That's right.

>> So to me, where the... What's your reaction to that?

>> The area that we've spent a lot of time in, what we think is a place that the market has been asking for someone to own and come in is all of these things that occur at every company. So there will be proprietary things that theCUBE does because of your specialty with certain types of data, certain types of media, that you will build and sort of conform and offer a better service to your market. But there's a lot of tasks that come and go, whether it's employee onboarding, open enrollment or annual budget cycles or people completing audits and tax returns and things that it turns out these are common across the enterprise. The homogeneity of what we all do, despite us being very different businesses, it's actually shockingly hot.

>> Expense reports. Fill them out.

>> You fill them out. And there's glitches and there's issues. And when I talk about things that show up in our data that people use us for a lot are things that are common but are infrequent. And what I mean by that is your taxes are a common thing. All citizens fill out tax returns, but it's infrequent. We do it once a year. And there's always something like, wait, where is this? And you got to find this other thing and search comes into picture. But there's also these actions that you have to take or things that you have to complete and find and then act on. So what we find is that that then becomes an area of expertise. That organization like ours can come into the largest companies in the world, like a Honeywell, like a Broadcom, like a Unilever, like a Toyota, and right out of the gate, turn it on and deliver tremendous amounts of impact.

>> Because they've documented it.

>> Well, because we've seen what it takes. So there's a deeper insight from an agent standpoint, what it means to find something within your ITSM? What does it mean to search across 24 different-

>> So you have a low-hanging fruit use case. So your success right now is coming in, being pragmatic and saying, "Hey, we kind of know the future, but right now the market wants this. And we can deliver value."

>> Exactly.

>> Which is the mundane tasks.

>> Which is what the CFOs are looking for. I think there's a lot of angst towards showing up with something that can be offset with other costs. And I think the idea of, hey, this thing will deliver pure productivity gains and will realize that over time just doesn't cut it for a lot.

>> Dave, and I always use the baseball analogies on theCUBE. It's like AI, just get on base, get a single, maybe stretch it to a double, but don't try to swing for the home run. Because that's where people are getting over themselves on over their skis or whatever you want to call it, because they go for the home run and it's not gettable. It's hard.

>> I think there's a lot of paralysis that goes into this, which is people are like... And I think this is sort of the build versus buy. There's things that you will build. You build things that are proprietary business, but when you have to interface, like we do, to hundreds of underlying systems, all of a sudden that single use case demo breaks down because you realize, oh wow, identity's not aligned between these systems. The namespace is not correct. We have to understand these different table structures. We have to put all it all together. And I think that's where third parties can help play a role for common types of actions. But where you have your own insights, you can kind of continue to go deeper and .

>> Talk about the company, Moveworks, you guys have 300 plus million in funding. You're over a couple billion dollars in valuation. Congratulations. Obviously you got some momentum. Can you talk about some of the momentum you guys have in terms of metrics that you can share publicly, revenue customers?

>> Yeah. So over the summer we announced that we had surpassed a hundred million of ARR. So we have seen scale, and it's not just the one-year scale. I think one of the things that I'm really proud of is having customers that have been on the platform for two, three, four years whose procurement teams are continuing not just to renew, but to expand their relationship with us. So that's part of it. Over the sort of 350 and growing large enterprises, we serve about a little over 5 million users.

>> Enterprise users.

>> Enterprise users and employees basically of these companies from all sorts of categories. From employees and knowledge workers to frontline workers and the likes. And we are known by our customers as sort of the AI assistant/co-pilot for all employees. So if you go to a place like GitHub, for example, GitHub uses Moveworks. Now, GitHub has their own co-pilot, but it's for engineers to write code, and they use Moveworks for all their other systems that people need to navigate and get help with and act-

>> It used to be called knowledge systems from the past. Expert networks or something. There will be always some database of-

>> Systems of record. Whatever term you want to use, but they're applications that the business is sort of founded on. Those will continue to persist for a long time.

>> Of course, they're resources. They all get smarter too.

>> Yeah. And they'll get smarter. As I said, there's application level AI for each of those ecosystems and each of those products, but there's also a need for an independent unbiased provider that can then peer into all of these, pull that data together and do that. So servicing over-

>> Give us a good example of that difference between the application layer, co-pilot and yours. What's the distinction?

>> Yeah, so we use GitHub for example, since I brought them up. If you're an employee there and let's say you're on their sales team and you're preparing for an enterprise, a sales meeting enterprise customer, maybe they have a few licenses, some engineers are using it, you want to expand it, they go to Moveworks. And Moveworks can help that sales rep prepare for that meeting by pulling data out of their CRM, by pulling data out of their case study repository, by pulling data out of other systems, assembling it and preparing them for it. They also use Moveworks to also navigate calendars, which is a simple task, but we all find ourselves spending a lot of time like, Hey, when's John available? When am I available to catch up on this deal? We can now go to Moveworks and we navigate that. So it's about us knowing where we can add the most value.

>> You're horizontal across.

>> We're very horizontal. And as I said, things that everyone has to do once in a while that keep coming up. Every day there's something you do once in a while versus trying to be a system that tries to be everything for every use case. I think we are this co-pilot for all, agent for all. But we also know that at the right time, you will use your Tableau GPT to really analyze some data sets. You'll use perhaps a tool from your CRM or even your office suite to help with a presentation that you're working on. But where we fit the model and where the GAP is is something that CIOs and CHROs can bring into the company. They can roll out to all employees as a single plane of blast that you can talk to. You can get search actions and you can look up information and get help. And mostly, or one of the important aspects of this, they get to control what it does. So they also name it. So if you go to Toyota, they call us Asian Ask, if you go to Unilever, they call us Una. If you go to Honeywell, they call us Red. And there's a reason for that. Because the customers have told us they don't want to use a third-party name for a tool that they ostensibly want to make their own inside their company. And so that's how we've gone to market, is allow them that full kind of agency.

>> And this is the key buy versus build decision. And this has come up, we had meter on before we had catch and release on, we have all these other services. We had one for accounts receivable. If you're not a core competency in the old model of going to business school, hey, never outsource your core competency. That was an old conventional wisdom. But when the business model is an operating system, you think like a CS person, put your CS hat on. Yeah, I got plumbing. It's a subsystem, but it's got to work. Because the whole thing is core, but the core IP might be something else. So I don't want to internally build something that the engineer leaves or get technical debt. Because a lot of these homegrown projects turn into black boxes and they never get used. And with AI, that cannot happen. Because it'll be so obvious, because you said about the business value, nothing will be adopted. So there's a trend back to rewiring the operating system of a business with managed services, core on-prem, connected to the cloud. And by the way, some people have scale that can't be replicated like AWS. As you guys get more scaled, you're going to see things before the customer.

>> Yeah.

>> You'll have visibility into, well, we kind of know that maybe expense reports or onboarding is huge for you. You have a girl's spur, you might want to look at this piece. I mean, you guys have that view, right? I mean, that's the key advantage.

>> Yeah. We see the demand and we see what is that very easy to tackle, but also powerful, strong ROI that come into picture. To your point though, around this whole tech debt and build versus buy. Look, it's happening every day with new models, with new capabilities. Your team builds something using GPT-4.5 turbo, and then there's GPT-4.0, and then there's model, there's Llama and its next version. And you're like, oh boy, who's going to now refactor what we built to do this? And I think what we've been able to help our customers with is by giving them a single interface. We are handling all of the evolution for the use of these new models. And right now we're rolling out the capability that customers can now upload screenshots to our assistant to say, Hey, help me with this. I'm stuck. Right? Before it was more chat. Well, we're innovating because the models support that. And the customers don't have to rethink anything. They just are like, wait, the future gets turned on. And all of a sudden now it's-

>> And by the way, the other issue that comes up that looks back like in the computer revolution was backwards compatibility. When you have these advancements, backwards, compatibility-

>> Very big deal.

>> Is a huge deal, and that's when you have these massive evolutionary phases where there's another processor, in this case, another LLM, more data is coming in. Having change out is huge. I mean, what's your reaction to that and how should CFOs think about this isn't just buying a SaaS subscription or even a usage pricing, if the things can't ride and then have backwards compatibility or resilience their toast?

>> Well, I think that's where I think sitting in a place on a perch where we can see all of these systems and continue to manage our interface with those. Systems change all the time. Even just not just forward-looking, but things break, things don't do as they're published to do. And you're constantly keeping up with these nuances that does make for the requirement to create a robust system. I think what you'll find is that a lot of things that get built in the AI space will work really well in the happy path, but they're very brittle and you can easily sort of do something and it sort of stops working. And that's because a lot of this is not just an AI problem, it's a systems and engineering problem and an access control problem and many other factors that go in. So to your point, there's a lot of nitty-grittiness that we have to have had to build and work-

>> And the AI infrastructure is the hottest area. I was talking to Robert Nishihara at AnyScale, and they're doing ML pipelining infrastructure. Because the big problem is that at certain scale, it's hard to sandbox large-scale machine learning pipelining. And then nevermind going into production. Talk about brittle, don't touch that button. The engineer who did that is not here anymore.

>> That's right.

>> I mean, it's running large-scale production. You just don't iterate and push code every day to a large-scale production, especially with machine learning stuff. What's your reaction to that kind of ?

>> So we're constantly fine-tuning models. We're constantly delivering different experiences, and some of these models can be updated in runtime based on user interaction, based on feedback, thumbs up, thumbs down. It improves a search ranking model. Other models have to go through a series of tests and evals that we have to then decide on. We have know a 70 human annotators that look at data to help us generate an opinion on are these models performing well? Are they drifting, are they overfitting? That then give us calibrations back to what we rolled out? And so it's all the above. Some of this stuff takes a committee decision to update. Some of it takes some of it's auto ML and we have these pipelines, to your point, where new customer comes on board, how you route things if you can't resolve it as an AI, the resolver groups at one company are very different than the resolver groups at another. So we've had to sort of automate a lot of that.

>> Well, I'm glad you guys are well-funded and successful, because this is at the forefront of pre-agentic. I mean, I don't mean to say it that way, but we are coming fast into the agentic infrastructure where agents that are coming, I mean obviously the co-pilots are great. But full task-driven agents where someone goes, "Hey, I'm in the New York Stock Exchange. I'm flying in every month. John wants to walk to the exchange. He wants a hotel. This is his preference. I need his flight and here's the Uber ready for him." When is that going to happen?

>> Well, if it's internal kind of tasks, that's happening today, and I think that's what we're really excited about. Our customers are too. Where someone can say, "Hey, I'm going to New York. I want to meet with the team for dinner. Can you tell me how much I can spend?" And the system actually will show you. First I'm going to figure out who Bavin is, how many employees of Moveworks works in New York? Okay, there's 14 of them. Now I'm going to do a rag lookup. I'm going to figure out what the policy is. Policy is a hundred dollars per employee for dinner. Comes back and says, you can spend $1,500 because 14 plus 1 is 15 times a hundred.

>> Can you ask, who else can I invite that has more credit power? Get the boss in there.

>> Well, it's interesting you say that because it obviously follows all the rules, but it will perhaps-

>> It's certainly a guide. Yeah.

>> It'll perhaps respond by saying, "Hey, here's the policies that I can share with you." And you might be able to see what the different policies allow for, and then you could make interpretation.

>> You certainly can make a move on that big time.

>> But we do a lot of guard railing, and we do a lot of sophistication where customers really want to make sure that we're not hallucinating, we're not giving out incorrect information. In fact, some of these things, customers don't want them summarized. Because they're like, look, that's the policy. Don't summarize it. Just give it back verbatim and let the user then decide, the employee decide what they want to do next. So a lot of safety controls that have had to be thought through, and not just thought through, we are running this at scale with millions of users that are giving us this implicit feedback of what's working.

>> I love how you get that data. Just keep pouring it back into the innovation. Thanks for coming on.

>> Thanks, John.

>> Great conversation. Certainly we'll have you back in since you're neighbor down the street. Six minutes away by car. By foot, no. By car. Thanks for coming on. I'm John Furrier here at theCUBE. We are here for the AI CFO event as well as the AI Innovation Silicon Valley. Thanks for watching.