#353 The Data Team's Agentic Future with Ketan Karkhanis, CEO at ThoughtSpot

Building agents is going to be as common as using an Excel file. Agents make you a super team. Agents are your force multipliers. You see, the biggest limiting factor to adoption is not the intent of embracing AI, but it is data readiness.

AI agents for analytics recently started to deliver on their promise, but there's a huge difference between the happy path of the data team being wildly productive and the unhappy path of the data team creating AI slop. And people trying to make decisions based on nonsense.

Don't let perfection be the enemy of progress is absolutely critical. The ability to talk to your data is now not an option. It is an imperative. Your column description should be agent-friendly, AI-friendly, and not human-friendly, because humans are never going to read that anymore. Agents are going to read that for you.

I want to know how the role of the data team is changing in this agentic future, how data teams can become AI-ready, and how we can bring data-driven decision-making to the whole of the organization. Our guest is Keitan Karkanis, the CEO at ThoughtSpot. Keitan is spearheading ThoughtSpot's transition from a business intelligence platform to an agentic analytics platform. So naturally, he's very keen on the topic.

I think so. It's the most exciting time to be a data analyst in the industry right now. The big thing we are seeing data analysts do now is becoming AI stewards.

Let's find out about the future of the data team.

Hi, Katen.

Welcome to the show. Hey, Richie Cotton.

Good to be here. Thanks for having me.

Yeah, great to have you here. I'm very excited for this. Now, to begin with, I want to talk about agents in data analytics. This seems to be the biggest thing right now. So talk me through what's the most impressive use case you've seen for agents in analytics?

Oh, I'll give you 4 because it's like a whole new world has opened up. I'll tell you, this is the reason I came to ThoughtSpot, because we are an enterprise AI company and we are transforming BI with our AI agents. And I'll tell you one of the most common use cases which everybody maybe could resonate with. You know, I'm sure, Richie, you've heard of this concept of self-service BI. Like, the word self-service gets thrown— it's been thrown around for more than a decade or even more, like, if you think about it. But if you really think about it, When we talk about self-service, what are we saying? We are saying you can create your own dashboards. You can self-serve your own dashboards, right? Now, tell me, Richie, who wakes up in the morning and says, I want to build some dashboards today?

There's only maybe a few business intelligence analysts.

Okay, let's leave them aside, right? But the point is, right, I genuinely feel, and I say it with utmost humility and respect. But self-service has been a hoax. It's been a hoax because what people want when they wake up is, or they go to their job and they plug in and they're like, I just want an answer. Now what you need is, and that's where our agent Spotter comes into play. Spotter is like your personal AI analyst, your data analyst, where you can ask any questions. I'll tell you how I start my, how today I started my day. You know, it's the middle of the week and I'm tracking my top 10 deals. For my quarter, right? You know, as a CEO, I kind of do that, right? How are we progressing on the top 10 deals? And I asked Spotter, hey, show me the latest on the top 10 deals and give me your sentiment. And it's my first deal is already closed. The second is happening. Third, it kind of flagged as yellow or red or something like that, right? And I'm like, great. And it said that there's a lack of engagement and activity. And then I said, well, that kind of seems odd. My sales team was telling me it's kind of a hot account. Like, what's happening? But then I asked Spotter, well, why don't you look at all data sources and come back to me with it? Look at Slack also. And then Spotter not just went to structured data, went to my Slack data, connected to the channel, figured it out, joined all of it, and told me, you know what? I think so there's a lot of Slack activity happening. We see this, this, this, this, this happening. So are you seeing this? Like, I could get my answers. I'm the CEO, but I don't have a data person. I don't have an ops team. You know who's my operations team? Spouter, my agent. That's my operations team. But that's just one simple use case where I have truly become self-service, if you may, because we run our staff meeting like that. We run our forecast call like that. But now you can multiply this across use case after use case after use case. Who doesn't have questions? Because that's the second important point about BI or analytics. You get an answer. You have one more question, then you have one more question, and it's like that question and answer tangle, right? And in the olden days, I actually am calling it olden days because it is, it just feels like an olden day. In the olden days, back in those days, every follow-up question would either become a filter on the top. Have you seen those dashboards with like many, many, many filters on the top? You know, what are those filters? They're basically follow-on questions people have, right? And you have to select it. But Sparrder, you can just have a very engaging conversation because Sparrder is not just your analyst, it's also your brainstorming partner. So that's an example of one. I'll give you another example, which I think so is way more exciting than this, is a lot of people have a lot of data, but let's— and we all say it's messy. It's all over the place. How do you get your data ready for the AI world? You see, the biggest limiting factor to adoption is not the intent of embracing AI, but it is data readiness. It is having your data in a governed fashion, metrics, semantic model. That's why we have another agent, Spotter Model. And Spotter Model is like your data engineer agent. So now your data engineers can just tell Spotter, to do all the semantic, automated semantics for their data. And now your data is becoming AI-ready 10 times faster. We have then gone ahead and we said, but yeah, people will still need dashboards sometimes, right? It's not like you don't need a dashboard. Like, you know, you need some charts for your QBRs, for your presentations, for your board meetings, blah, blah, blah, blah, blah, blah, all that stuff. Or every day you don't want to start your day with just asking a question. You probably want to start with a chart and then go from there. Which is fair. That's why we created SpotterWiz. SpotterWiz is like Lovable for dashboards. It'll just create your dashboards for you. So now imagine this world wherein every persona, every in the analytics workflow can be augmented or rather amplified. Their impact can be amplified with agents. I think agents make you a super team. And I'll tell you, Richie, everybody is going to have 5, 10 of these agents working for you. You're going to have them working for you, Richie. You, you, you need your own data analyst, Richie. And you, you need that 24 by 7, and that's your agent. So we're very excited about what's happening in the world right now. And we are taking it one step further. In the next couple of months, we are getting ready to launch the working name for it is Agent Spot.

It's a—

but I'm sure my CMO will have something to say about it, and I'll let her think about it. I've got a wonderful CMO. But the idea there is not to stop at there. The idea is to take it from insights to automate actions and outcomes. And that is the world of autonomous analytics. So we are going to move. So we are in agentic today, and from agentic, we are going to move to autonomous. Which is gonna be a whole another set of innovations we can talk about.

Absolutely. There's a lot of exciting things to talk about, so I'll forgive you a long answer there. But it's interesting. So, I mean, you talked about self-service analytics being a hoax, and I guess it's maybe been an aspirational thing for a decade that we're gradually sort of getting closer to. Now, all the agents you mentioned, they sound like employee replacements almost. So I'm just working at trying to work out how close are we to having an agent that is equivalent to a real employee and where a human's still sort of useful No, Richie, I may have to take a different perspective out there.

I don't think so. They are replacements. I'll tell you, the biggest problem in the data team has been a shortage of talent. Every company has highly constrained data teams. Have you seen their backlogs? If you go to a data team and you ask them to show them your backlog, they'll be like, high, strong. I mean, they're wonderful people. They're really smart people, but no human being can go through all of that. Why? Because the data needs of an organization are infinite. It's a long tail of data needs. So what happens? You take your 2 people and you put them on the most important one. And what happens to the rest? Eh, they get ignored. Not because you don't want to do them, not because you don't like those people, it's just you physically, you can't get to it. This is the very important point. What we are seeing is customers are coming to us and they're saying, Whoa, for the first time in my life, I can actually get through my backlog and keep my entire business stakeholder team happy. This is the idea of having an augmented super team. Is the data analyst plus their agent SpotterViz. It is the data engineer plus their agent Spotter model. So I think so that's how we think about it. Now over time, will certain things that an analyst or an engineer or a business user do themselves be not needed to be done? Yeah, absolutely. So for example, I'll give you a great example. A lot of times analysts are spending time on dashboard layout and structure and charts and this and that, but they're an analyst. They are not a dashboard creator. See, we have made analysts— analysts need to analyze the data and give you strategic inputs and help you think through your business. So actually, it takes time. It frees them up from the mundane. Allowing them to really focus on the business context and the strategy. Over time, I think so it's a massive, what I call, force multiplier. Agents are your force multiplier. So if you are a data team, you're like, whoa, I can now do 10x more. I mean, who doesn't want to do 10x more?

I mean, yeah, I'm sure everyone wants to do 10x more. And I do take your point, actually. I've spoken to people in a lot of different data teams, like in different companies, and there's no one who's like, yeah. We're having a slack week. We've not got much work to do. Yeah, everyone in this industry is busy. So I like the idea that more productivity is definitely needed and it's not necessarily going to be like, oh, we're cutting jobs because agents have taken over. You do make it sound like the roles are changing though. So perhaps do you want to start with data analysts? Are you looking for different skill profiles? Are people doing different tasks? How does the data analyst role change now that you have powerful agents?

I think so. It's the most exciting time to be a data analyst in the industry right now. I genuinely believe so because for a change, they are no longer just churning out dashboards. They don't like that. All day, all day long, they're just churning out dashboards. That's not fun. That's not fun. Oh, add a filter. Great. I feel very accomplished. I added a filter to it. They don't like that, you know, but we make them do that. So number one, I think so AI helps them actually have an extremely amazingly fulfilling career. That's number one. But let's take that role persona. What changes for them? Two things change for them. Number one is. They absolutely have to embrace new skills. And that's why we are investing a lot. We have our community, the ThoughtSpot Champions community, where we are helping data analysts embrace the new world of AI. And new dimensions start getting added to their job. And I think so this is still getting figured out. So let's not say everything is set in stone. There's a little bit of learning we all have to do. And I think so it's important to be humble about it. Like, We, some stuff we need to discover still, right? But I'll tell you the big thing we are seeing data analysts do now is becoming AI stewards. They're becoming stewards of the data governance semantics. Two, their role is not about now I give you 5 dashboards, but I give you these curated governed data sets, which then you can do anything you want with. But three, they also become agent architects. They become agent whisperers. They are the AI enablers for your data platform. So of course, that will require a little bit of new skills and a little bit of exciting new things to learn. But I think it's like the best time to be a data analyst, right? I think so. It's the best time to be in the data business, period.

Yeah, I agree with you there. It's definitely exciting times to be a data analyst. So you mentioned the idea of curating the semantics of the data. So that requires some domain knowledge then to understand what does the data actually mean. So is there a shift then, data analysts becoming closer to commercial teams then?

Oh, two things. Yes, no, absolutely. I think so. This does allow analysts and data engineers to come closer to their quote-unquote business teams, right? And really ingrain themselves in the business of the business versus the technology of the business. And I think so that's a very important strategic point to make. But two, I think so what's also happening is they can now use AI or Spotter model in this case to automate semantics. Like semantics is hard manual work, right? Column descriptions, metadata, metrics, keeping everything neatly organized. And then one thing changes and it has to ripple everywhere else. And then you've got data here with one semantic model, data here with another. There's a lot of work people have to end up doing. But I think so this, our agents actually help them speed semantics, but also that there's a subtle part which we are picking up from our customers is the biggest value they're getting is also understanding what is in their data. Do you know one of the biggest problems, and sometimes that's referred to as data literacy and all that kind of stuff, but the favorite question people ask Spotter, we kind of poll people on what kind of questions you're asking, what's bubbling up, and what's hot, if you may. The hot question which we see so many customers ask is two questions. Number one is like, what is in my data? What's in this data? Just tell me what's in it. And then they're asking brainstorming questions like, I want to measure pipeline velocity. How do I do that? So, it's less about the what, but it's like the how. And I think so this is very exciting, very, very exciting because now Spotter is becoming your thought partner, your trusted advisor. So, I think so it's all these things put together that that I think so are going to be, in fact, next week, I think so. Yeah, I think so. Yeah, Monday, March 9th or something, we are launching a whole new platform. We are calling it Spotter Semantics, and I'm sure we'll make a big splash out of it. But I'm very excited about Spotter Semantics because that comes included with the ThoughtSpot platform.

Okay, wonderful. Yeah, I can certainly see how Being able to have AI that helps you figure out what the methodology should be in order to get the answers is just equally as useful as just giving you the direct answer and not explaining how did it get there.

It also builds trust, right, Richie? Because when you see how certain things are done, you start trusting your AI. Because look, there are lots of people out there will be like, well, you know, here's the answer. Well, how do I trust it? How do I know you followed the right I think so. Yeah, it's quite important.

Absolutely. I mean, it's certainly, uh, if you've got, um, some kind of regulatory, uh, compliance, um, thing of— if you've got to comply with any regulations, then you're going to have to explain how was the analysis done. You don't want to have just an answer, you got to show your working as well.

Yeah, and, and auditable, uh, auditable and traceable, right? Like you, you, you should be able to walk back the answer and see exactly how it was arrived. So yeah, absolutely. Absolutely. Okay, wonderful.

And so we talked a bit about how data analyst roles are changing. How about data engineers? Is there a similar sort of shift there?

Yeah, I mean, you know, I think so. Again, the same concept is how can they use an AI-first data engineering approach? You know, it's less now about us writing the SQL or manually updating the metadata and all that. And it's more about like the best part is like if you look at Spotter Model, our agent for data modeling. And data engineering, you start using Spotr Model by not saying, these are the 3 tables I have in Snowflake. You start by saying, my data is here. I want to create a lead performance dashboard. How should I model my data? And Spotr Model will then identify your data and then go through all of it. So I think so the shift happens on, again, to a very important point you mentioned, wherein a data engineer is no longer just a data engineer, but also a business engineer. They're understanding the business problem, the context, and designing the model to give the business team the insights they need. But I see all data engineers to be becoming AI-first. They don't have to now write scripts and all that to update. One of the biggest lacking things in people data is column descriptions. I know it sounds like a simple thing, but we've observed that every Everywhere. Everywhere. And it's terrible. It's terrible. I'll tell you why it's terrible. Because we have designed our world for humans. We have designed data and metadata structures for humans. The biggest shift that is happening is now you have to design it for agents. Your meta— your column description should be agent-friendly, AI-friendly, and not human-friendly, because humans are never going to read that anymore. Agents are going to read that for you. So that's why it's a, it's a very exciting shift that's happening out there.

No, that's fair. I think historically, yeah, humans hate writing documentation. They hate reading documentation. And so yeah, it's always lacking. I suppose that's the good thing about that. Yeah, maybe an AI data analyst is okay with reading documentation. That's fine. So is that something you need to get set up then before you start making use of agents? Like, do you need to have all this documentation there? In place before you get going with agents?

No, we have been working hard on what we call the cold start problem. What we are focused on is not just a shiny demo. A very core principle we have is our AI will lead you to ROI. And if it doesn't, then that's a conversation. So to set up or to get started with ThoughtSpot, I think so the biggest thing you need is is curiosity. I'm sure we love if you are on already a cloud data warehouse. Snowflake is an excellent partner of ours. Databricks is a great partner of ours. Google BigQuery, great partner of ours. And we are so seamlessly integrated with them that if you are on Snowflake, you can get started in 2 minutes. Because for example, I'll tell you, if you have all your semantic model in Snowflake, ThoughtSpot will just inherit it. You don't have to worry about it. Or if you're using so many other things out there. So yeah, no, to get started, we just encourage people, I think so this is also an age wherein it's not just about technology, but it's also about imagining a future and having the curiosity to imagine that. And what we are doing is spending time with our customers, so many of them across the world, whether it's Schneider in Europe, whether it's Macquarie Bank in Australia, whether it's EasyJet in London, whether it's Navant right here in the Valley, down the road, all of them are focused on one thing and one thing only. How can they use data to transform not just their internal operations, but their customer experience? This is a very big thing that's happening. I was just before coming to your recording, I was seeing on LinkedIn, somebody posted one of our customers. That's always the best thing for me is when my customers are posting videos of how they're using ThoughtSpot. That's kind of pretty cool, you know? But Thrive has embedded AI into their own application, transforming their customer experience with our Intelligent Apps product. So it's becoming an imperative that you transform. Data is going to be the driver of the AI enterprise. It's going to be right at the center, and data is going to be powering not just Analytics and insights, agentic, but applications and agents. And that triangle, at that center of that is where ThoughtSpot is. And that's why we are having so much fun, as you can tell.

Yeah, I mean, definitely there's lots of different kind of, I guess, outputs. You say you can have agents as the output, you can have applications as the output. So lots of different possibilities there. You've been working with Snowflake, with Databricks, with BigQuery, Google for their BigQuery product. It sounds like all this kind of governance then is happening in the data warehouse. So is that the idea that you get all your data warehouse in order and that's going to enable the use of agents?

Yes and no, both. We want to give our customers choices. Every customer is at a different maturity curve. Some are at a place wherein they have everything neatly, beautifully organized and they're good to go. Some are like, well, I'm not even in the cloud, Ketan. And I need to go. And that's a reality. Nothing right or wrong about it. It's just you— what we are doing is, so for example, no, I mean, no, it's not necessary for you to have everything organized. That's where our tooling and our technology comes into play. We really focus on how can we get you to value in 30 days on one use case? How can we get your first agent live in 30 days in one use case? And then you go from there. What we are noticing is a lot of customers Once they see that first proof point, once they see the first use, it just catches like wildfire. Then they don't require any prodding. It's like then the tail wags the dog and then the data starts becoming clean very quickly. But no, it's not required that you should have everything organized or anything like that at all. Not at all.

Data quality, data cleaning, data management is one of our sort of regular topics on the show. And it's like, it's somehow a never-solved problem. It's like, The tooling is getting better, but there's always work to be done there. I'd love to zoom out a little bit and talk about data strategy. So now we've got all this different tooling, just at a high level of your organization, do you need to completely change your data strategy?

I think so. Let's start. Strategy is an important part, but I think so. The first aspect of data strategy is going to be your data culture. And I think so culture becomes the key multiplier for whether it's your data strategy or your AI strategy. And what do I mean by that? I'll give you a couple of examples, just a couple. Not that I'm going to give you a culture doctrine out here, but one of the most important things we have seen our customers struggle with is just the enormity of the job. They're like, oh my God, I've got this 10 years of like, what am I going to do? And then some of them actually come to us and they say like, Kethan, I've got like thousands of Tableau dashboards, which nobody even knows what they do. And like, what do I— it just seems enormous, right? Getting out of it. But the culture of don't let perfection be the enemy of progress. Is critical, is absolutely critical. And I think so that becomes center stage. Leaders who are driving with that culture are making the most amount of progress and achieving the most amount of success in their organizations. Two is realizing that when you think of your data strategy, more of the same is not the answer. You can't just say, I'm going to make, or I'm just going to, my strategy is to make dashboard creation easier. I think so that's missing the point. With utmost respect, I mean, I want to be very empathetic to everybody, but it's like saying, thinking the legacy products out there, the visualization and all those products, they're almost like the flip phones. You have to skip the generation and move to a smartphone. You cannot say, I'm now going to have a flip phone with a slide-out keyboard. I'm trying to have fun with you right now, but like, is this making sense? Like, you have to take the generational leap and you have to upgrade. Not— it's an upgrade supercycle is what we are seeing. But if customers— and a key part of the strategy is, no, I mean, you know, the ability to talk to your data is now not an option. It is an imperative. The ability to have agents with every business user is an imperative. And how do you do that? And in the beginning, just like back in those days when the smartphone launched, people were like, well, it's only for the executive. Like, no, everybody's going to have that. Everybody's going to need that. So I think so. And the third point I would say when it comes to data strategy is, look, recognizing that it's going to be not all in one place. We are seeing a lot of our customers with what I call multiple data estates. And it's a very important fact that allowing them the ability, for example, one of the capabilities we are working on is having the ability to have a semantic model that transcends multiple data estates. Imagine that, like a common understanding layer across data estates. So recognizing that is important. And ultimately, The key part of the data strategy, and I'm not saying anything technical if you observe, it's not because I think so we've got the technology, you got to bring the culture and together it's kind of then a party. The final point I'll make out there is people. A key part of your data strategy needs to be your people strategy. How are you going to organize your data team? What are the new skills you're going to invest? How are you going to redo your approach to managing your business stakeholders? And that has to be AI first, agent first. And that is like, if you think about those things, the next year you're going to still see, it's like all these things, the progress seems like slow at the beginning and then it's sudden. And I think so we are at the tail end of that slow period wherein people are still coming out of it, getting a better appreciation for it. Understanding it, and then boom, you're, it's by next year, you're going to see everybody having their own data agent working with you. Like you're going to have one. If not, I'll send you one.

No, I agree that like, it's amazing how quickly it's gone from AI agents are kind of coming soon. They are the future to now it's almost table stakes that you have to have an agentic way of working for anything with data. Now, you asked some great questions there, like how do you organize your data team in your organization? How do you upskill all your people? Those are actually going to be my next questions. So I want to make you answer those questions that you just asked.

Which one do you want first?

Okay, let's go with the organizational design stuff because, yeah, do team structures need to change? Do you need different ways of teams interacting with each other? Like, how should the all this work?

Oh, and we are shaping this up ourselves. So, we are working with a group of ThoughtSpot champions. These are key members of our community. As I said, our community is our strength. And we are kind of— this is another thing, like all my best ideas are not mine. They're from my customers. I just want to be very clear about that. I'm simply an assembler of all of their ideas, if you may. But one of the key things, a great idea they had, and maybe it's a a small thing, but the first thing you should look at is how are you hiring people? Go look at the job description you have and then look in those job descriptions and are the first two bullet points AI readiness and AI savvy? Like you can use AI tools. Like, remember, I'll give you a very simple example. If you're a data analyst, you probably have in the job description familiar with using SQL and Snowflake and XYZ and XYZ, right? Forget that. The first point should be savviness with agent infrastructure, ability to use agent-first models, familiarity with LLMs, and everybody will have their own tweaks. But the point is, first introduce that in your hiring so that you start attracting and building the right teams. Two, the next thing that we are seeing a lot of our customers do is invest in their people. We can't assume everybody's just going to learn this on their own. I think so we have to help. We have to have empathy. We have to, have to, have to give them a path to come into the future. We absolutely have to. And that's why you see, I formally say to everybody is make this training mandatory. Don't make it optional. It's like, will you hire anybody who doesn't know how to use the internet? Sounds like a silly question.

But I suppose, yeah, a couple of decades ago, there was mandatory training for this is how you use the internet.

That's the same thing. This is bigger than the internet. 3, I think, so set clear expectations and goals. The big thing we are seeing in some companies is like they get lost. They get like enamored with demos. Somebody will come demo them, "Hey, I can do text-to-SQL conversational." And what they're doing is text-to-SQL. Text-to-SQL was done 20 years ago. You don't need AI to do text-to-SQL. Okay? Like, seriously. And the reason I say that is connect your AI initiatives to business initiatives and ROI. Take 3, 4 tangible use cases. That I want to do this supply chain. Nike is a great example that's working with us, or Cisco is another great example. So connect it, connect AI to ROI. And frankly, I would say that be prepared to experiment together. And I'm not giving you a playbook, but I'm giving you certain ideas because that's the very important point. Don't wait for a baked 5-year strategy. If you're trying to write down a 5-year strategy, you are gonna miss the train big time. Because there is, this thing is changing every 3 months. You wanna be very agile in how you're approaching it.

Absolutely.

Yeah.

I mean, it's very difficult to predict 5 years in the future. I think if you go back to 2021, no one's gonna have predicted this year very accurately. Okay, so you mentioned about the job adverts and how you want AI readiness, AI skills right at the top of the list. You know, the more I find out about hiring, like, it amazes me how much like job adverts, they just copied and pasted from last year and it goes, keeps going on the same. So I like the idea of writing something from scratch, deciding the skills you want. Would you still keep things like SQL, Snowflake, all those things? Are they going to be in that job advert just later down or it's like, less focused on the AI skills.

Yeah, they will be. But like, look, if you're— if your ability to just write a great SQL is lesser value than your ability to transact, to think business, be AI forward, be familiar with using, uh, you know, uh, technologies like Spotter or Spotter model and so on and so forth, uh, and I think so, it's like, um, it almost becomes like, yeah, of course you know how to write SQL, like, you know what I'm saying? That's not the tilting factor for you to get the job. But also you have to think about the workforce of the future. And the workforce of the future is an AI-first workforce. It is happening. And we all need to reskill and upskill into that future. We all need to. Some will do it sooner, Some will do it later, but all of us will have to do it. It's just going to happen. And I'll tell you, it's extremely exciting and rewarding. Extremely. I'll tell you one of the best. Yesterday, I got one of my— our HR leader, Kim, she sends me a text message. I was supposed to talk to her and we were just back and forth. We were on text. And she texts me, is like, can you give me a few more minutes? I'm a little tied up right now. I'm like, sure, what's up? Like, you know, what's up? She's like, no, I'm building this agent, which is going to help people onboard in the company. And I want to experiment with it. Like just that conversation she was having with me was incredible. And she was excited. She was like, whoa, I now have the power to fundamentally alter not just the employee experience, which is very important to an HR leader, is the employee experience, but also the candidate experience. So from her job fulfillment standpoint, AI is helping her be a much more incredible HR executive and leader. So what I'm trying to say here, this is a micro example of it's actually going to help you feel more fulfilled and in control. Of what you're doing. So I think so it's going to be exciting. Embrace it is my mission, is just message. That's it.

Yeah. I mean, I love that story because HR teams, I guess traditionally they're not highly technical people. So you go into HR because you love working with people, not necessarily technology. So yeah, this is someone who is using cutting-edge technology. They don't necessarily have like, they're probably not a programmer. But they're doing something which has a high impact both for your internal employees, but also new employees and candidates.

I'll tell you, building agents is going to be as easy, as common as using an Excel file.

That's brilliant. I'm not a huge fan of Excel. I think this is more fun tech. Yeah. But yeah, it's everywhere. Wonderful. All right. So I'd like to talk a little bit about change management as well. So how do you get to this state where everyone's adopting AI, everyone's actually making use of this stuff? Because building those new habits is difficult.

Habits, that's a key word you used out there. We touched upon some of those in our strategy topic, but the key out here is, um, there are two ways to do change management. You know, one is, uh, is just, just force-feeding it in your organization and, and all that stuff, and let's leave that aside. Some might do it, some may want to do it. I'll tell you the biggest way to think about change management is, number one, leaders need to embody it first. It's really important. If you are a leader, whether you're a data leader, an IT leader, a line of business owner, an executive, a CEO, doesn't matter whoever you are, you need to first show vulnerability by saying, yeah, I need to go learn. You know, it's okay. Not everybody is born with AI fed into their veins. You know, it's like some of us have to go learn and it's okay. So, and it's okay to say that is my point. And you have to give your organization the permission to do that. So number one is live that AI-first value. I'll tell you a great example. In my staff, in my leadership, my ELT meeting, every Thursday I have my ELT meeting. I do it on Thursdays. I don't do it on Mondays. Sundays. There are many reasons why some, and that's another podcast. But Thursdays I do my ELT and the whole meeting is run on, no, you're not allowed any slides, Excel, nothing is allowed in that meeting. You cannot do that. You are, you, we run the whole meeting on Sparrr, our agent. Like my agent is like an ELT member for me. Is this making sense? And this agent sitting at the table with me and my, my LT. So it's my, you know, it's a C-suite. My agent is my C-suite. It's kind of funny. It's kind of really funny because then funny in a very positive way, because then you start using it in real situations, not in theoretical demos. So for example, one of the first ways we start the meeting is we look at our NPS. Every meeting, every Thursday we are tracking NPS both by segment, by industry, by many different aspects. But the first question we ask is, how are our customers doing? What is last week's NPS? We track it at that micro level. And so, my usual question is, Spotter is like, and we'll be like, what was NPS? And then there's a follow-on conversation. And then the conversation is like, somebody will say, well, that happened, so that happened. And then Spotter will say something. That this happened and this happened. And it creates like this interesting dynamic in the meeting. But very soon, at the beginning, people resisted. I saw like my LT was a little bit like slightly uncomfortable, if you may. Like, what is Spotter going to say now? You know, what is this AI going to say now? But very soon, they love it because everybody's now on the same page and answers are instantaneous. So living it. Is important, however imperfect, however imperfect it might be to begin with. Live with it. Live it. Number 3, I would say for change management, I'll make this point and I cannot make it enough, is ROI. Every AI project has to be tied to ROI. And that ROI could be we are going to eliminate 5,000 Tableau dashboards that have been lying around and we are maintaining and wasting time maintaining Or that ROI could be we are going to reduce travel policy cancellations by X amount. Or that ROI could be we are going to drive 2% supply chain efficiency in this country for this product. Doesn't matter what it could be. It's there. And the final point I would say is change management is be prepared to let go. It's a subtle point. But be prepared to let go because you can't be in this world of hyper command control data in the organization. What you have to think about is governed center of excellence spawning innovation of data at the edge. So as long as you have the governed center, let the edge, let a thousand flowers bloom, and that will help you figure out which area to focus on. But we could talk change management for hours. I'm not an expert, but I'm just reflecting what I'm seeing. In the market.

It's a whole episode in itself, but I do love your idea of agents as being people in a meeting. I mean, people's not quite the right word, but this is a concept I've not heard before. So I'd love to know a bit more details about how this works. Like, does everyone get their own version of Spotter within the meeting, or is it like, is this a voice activation and it joins your Zoom call on the thing? How does it work?

No, I'm not fully at the Jarvis level yet, though I would love to get there. My team is building voice, by the way, so we should have that very soon. No, we have it up on the screen. You know how when you're in a staff meeting, you have a presentation and somebody has a slide with data or some metrics? So no, we just have Spotter on the screen. That's it. And somebody will, my chief of staff or me, I would just ask a question or anybody will ask a question and then people jump over each other. Then they're like, no, no, no, ask Spotter that. No, no, no, I spotted that. So it kind of becomes a game, if you may. But, and it's not just, the idea is to provoke the right conversations. That is the key. 90% of the time, we as humans, we know what is the right thing to do if we are debating the right things. And bringing focus to what matters is what Spotter does. And I think so that's really, really exciting. But that's just one way of using it. I'll tell you another way of using it. My head of— we've got an incredible sales leader, Scott Parsons, who is like our SVP of North America. He leads all of North. He's a sales leader, okay? Not exactly a tech-first kind of a personality, but I was with him yesterday and he's like, Ketan, here's what I did with Spotter. They all come and tell me, here's what they did with Spotter recently. Like, it's kind of fun. And he's like, Ketan, I'm taking a trip to New York. So I told Spotter and I was like figuring out what to do. And usually I would've asked my admin and everybody to figure out and reach out to my local sales leaders so that they can organize meetings and so on and so forth. But instead he went to Spotter and he said, I'm traveling to NY/NJ next week. Look up our customers and tell me whom should I be meeting based on our pipeline.

Okay, nice. Boom.

And it looked at the pipeline, it looked at deal progression, it looked at all of that. Then it looked at relevant Slack conversations for those things, knowing which is hot, which is not, and prepared for him a list of top 10 accounts to visit while he is there. Then he says, okay, great. Now he's like, okay, great. Send a message to all the AEs to try and organize meetings for these. And then, but this is an, I'm giving you another example. Like the examples are numerous, but the idea is what would've taken him I don't know, 20 emails and 10 phone calls and coordination. Boom. He just did it himself.

These are not from sort of tasks like calling lots of people and having backwards and forwards and things and just organizing all that stuff. It's real kind of, unless you're really into administration, which not many people are, these are the tasks you want AI to be taken care of so you can focus on the more fun stuff. So we talked a lot about the positive of AI. I want to know, are there sensible ways of dealing when it makes mistakes? Because particularly a lot of these agents, they're in their infancy, and if your data's not perfect, then it's gonna make mistakes. So how do you do this safely?

Mm-hmm. Mm-hmm. Mm-hmm. It's a very important point. You know, I was in the beginning of the meeting, I was asking you, I was telling you what's hot, what's not, what people are saying to Spotter. You know, one of the things that I see a lot of people now I mean, we teach people to do that, is check your work or show me your work. Is this making sense? Like you literally have to say to spotter, show me your work. How did you do this? Or sometimes I say like, huh, that is odd. Can you double check please? I'd say please also because just, I'm just a nice guy. And the idea there is. Accepting the fact that we are not in a perfect world, that AI is not perfection. The job of AI is to give you directional guidance, 70%, 80% there, and for you to always have that human-in-the-loop idea. But it's also supported by what I call fundamental architecture choices. And I'll tell you why we can do it and why others can't do it. So in Sparrow, when you ask a question and you get an answer, we are not just plopping a chart. We are giving an explanation, but most importantly, Spotter will first show you how it arrived at the answer, what tables it looked at, what formulas it used, blah, blah, blah, blah, blah, blah. And it will give you all of that. Second, you can then click edit and actually see the tokens that we are used to generate the SQL. A big part of our technology is we are not using an LLM to generate SQL. It's a very important point, Rich. Everybody else is doing that. And that's not good. That's not good at all. So I know you want to say something, but I'll— the bigger problem is AI and LLMs by definition are not— are probabilistic. AI is just probabilistic. What we are doing is we are using the LLM to convert language into tokens. But from token to SQL to your database, which could be DirectQuery, we don't create cache, we hit your Snowflake, is all our proprietary technology, TML, and our search data technology that we have built. And that allows us to create a trusted, repeatable fashion. The same answer will be— you will get the same answer. And plus, we should go all the way down to not just show you the SQL we generated, but the query path and the query plan that we created. So there's a lot of transparency. All these are examples of transparency. And sometimes in that, you will notice like, whoa, there's something wrong. This is just a wrong answer. Oh, that's because, you know, AI was not smart enough to figure out that Rachel's maiden surname is not there in the database anymore. Her married name is is there. Like, how will AI know that? So it's trying to find Rachel Bogner while, you know, all that kind of stuff. So yeah, I don't think so. We should assume the key is to architecturally design for transparency. And if you design for transparency, then trust follows. Even if the transparency will show you AI is making a mistake, you will actually be like, good, I found you were making a mistake. That's actually a good thing. The idea is not to be perfect. The idea is to be human.

Yeah, I love it. And I think that's maybe a common misconception about a lot of these AI applications. People think it's just like one LLM doing everything, but actually you're gonna have a bit of deterministic software in there. You're gonna have other technologies in the LLM. It's only gonna be one part of that. So I love that. Architectural distinction there. Just to wrap up, I always want more people to learn from, so whose work are you most excited about at the moment?

I'm getting really excited by the work I'm seeing some of my customers do. I'm really getting excited by what they're doing. They're very inspiring to me. Some of these users, I give you the example of Macquarie, or I give you the example of EasyJet, or I give you the example of Roche out here. They are fundamentally, or Elevance in UnitedHealth Group. They are fundamentally taking the leap into transforming their data experiences. They're not just thinking, hey, I will create more charts. They're thinking, how do I drive more growth in my business? And they're solving real problems, which are not just technology problems, but they're solving culture problems. They're solving change management problems. And in many ways, you're not reading those stories in TechCrunch and on Twitter. And I know we probably get enamored by this new algorithm was created by this person and that, and yes, that is very cool. Power to them. Great. Good job. Congratulations. Including my team. But I'll tell you, the real work is happening by these data teams everywhere. And I wish I could write a TechCrunch article every week on them or a Bloomberg article, but I'm quite inspired by them. I'm quite inspired by my ThoughtSpot champions. In my community. And, you know, it's— every day is a new day. Every day is a new day. And we just look forward to having a very exciting time because we are liberating the world from legacy visualization and BI. And it's quite exciting as we do that. And I'll tell you, the thing that inspires all of ThoughtSpot is not just our customers, but And maybe we could close on this point is we are a mission-driven company. Our mission is not AI. AI is means to an end. Our mission is to make the world more fact-driven. That's our mission. And that's a very inspiring mission and very enduring mission. And AI and technology is just means to that end, but we have our sights set on a bigger prize. which is to make the world more fact-driven. That's why it's fun.

Uh, yeah, I mean, it's a, it's a wonderful mission and I agree there's been this sort of shift from, uh, generative AI first came out, everyone was just talking about like the technology, the models and gradually it's moving towards, well, how do we do stuff with it? And it is going mainstream. So yeah, uh, certainly exciting times. Uh, wonderful. Uh, thank you, Caden, for all your time.

Thank you so much. And, uh, hope to hope you're well. Anytime you're in the Valley, come by. We are in Mountain View. I would love to host you.

Absolutely. Thank you. All right.

Bye. Have a good day.