Meet ThoughtSpot Analytics Platform with CEO Ketan Karkhanis & SVP Francois Lopitaux

0:00Francois, Katam, hello, welcome.

0:02Hey Tim, good to be here.

0:04Hi Tim.

0:05Thank you so much for joining me.

0:10Absolutely.

0:11I think you're both in Silicon Valley, right?

0:13And I'm in the UK.

0:14So thank you so much for agreeing to me across time zones.

0:18It's it's fantastic.

0:19Um it makes sense probably to just explain sort of why you're on the channel, if that makes sense.

0:23So um I think I've been talking quite publicly about exploring new tools and um one of your colleagues reached out to me and said, hey, we should we should get both of you on the channel.

0:33to showcase the product.

0:35And that was about two weeks ago.

0:36Katan, you reached out to me on LinkedIn saying, hey, you need to you need to try our product.

0:40And I was like, yeah, okay, let's let's make it happen.

0:42And so um yeah, we've got this conversation.

0:44So

0:45Let me hand over to you to maybe introduce yourselves.

0:47Katan, I'll let you go first.

0:49Oh thank you.

0:49And Tim, appreciate you uh uh broadening your horizons, if you may.

0:54Uh I'm Kaitan Karkanes.

0:56I'm the CEO of TalkSpot uh and it's an absolute pleasure.

1:00I've been in this uh in this role almost 17 months.

1:04Uh prior to that I spent a lot of time at Salesforce.

1:08Uh we can get into that in a bit.

1:10Einstein analytics, tableau, all those things.

1:13Uh but listen, just glad to come and share with you and your viewers what we are doing and um and get some thoughts.

1:19Francois.

1:20Amazing.

1:21Thank you for joining me.

1:22Francois.

1:23Yeah, so I'm uh SVPS products, I'm also based in in the Bay Area and I, you know, like Catan, I spend time in startup, big company

1:33sometime at Salesforce.

1:34But the the funny story that I have actually is my first CEO when I started like maybe 20 years ago.

1:40Um it was like twenty-five company people and the the CEO actually was uh the first employee of Business Object.

1:47So I guess my my you know destiny was to work in that appointment.

1:53Bob J they call it, right?

1:58That's a funny nickname.

1:59Yeah.

2:00Alex Alex Alexion.

2:04I love it.

2:05I love that nickname as well.

2:06I didn't know what Bob J was.

2:08I thought it was it was the name of someone in the company until someone explained to me.

2:12That's a nickname for business objects and I just thought, oh my goodness, how did I know that?

2:16And I was in a I was a consultant at Extension.

2:18I didn't know what Bob J was anyway.

2:20Story for another day.

2:23I I guess it might help the audience to to just understand a little bit about sort of what brought you to the roles that you do.

2:29Like what makes you passionate, Katan, in your case?

2:31about leading I I guess a a company and enterprise in this space.

2:36And then Francois, from your perspective, you're obviously a lot more product focused.

2:39Sort of what gets you excited about, you know, the role that you do at ThoughtSpot?

2:43Uh no thank you.

2:44Thank you for letting me share.

2:45Look, I I as I said the the I I've been in I've been in enterprise software, data, AI and analytics all my all my career

2:55Uh I started uh you know, I was in Salesforce since two thousand nine, went all the way to two thousand nineteen, ten years, had a chance to build a lot of things, including Einstein Analytics, which now is known as the CRMA or something like that.

3:10Um but Francois was with me on my team back then when we built that.

3:13I see.

3:14Um so that's uh we we go back uh a long way.

3:17Uh but you know, and then uh we we are quite tableau and all those kindness things, but um

3:24Then I left Salesforce.

3:26I did uh a startup in the supply chain.

3:28We sold that, which was very good.

3:31And then I went back to Salesforce as a boomerang.

3:33Uh to run an yes, yes, because you know it was a fantastic place.

3:38Yeah.

3:39And uh I I was there and they gave me an opportunity to run Sales Cloud, the flagship project.

3:46Which was like incredibly exciting opportunity.

3:49And as I was building that, the the you know, I'd done a lot of machine learning, not AI, in Einstein analytics days, and then the the leap to generative happened

3:59And the whole world started changing.

4:01And I and I think so we could very quickly I could pick up that look, the enterprise stack of the future is gonna be completely different than of the past.

4:11You know the way cloud changed the entire enterprise stack.

4:15Uh what was in the enterprise in nineteen ninety-five was not in the enterprise in two thousand five.

4:20True.

4:20Completely changed.

4:22The companies didn't exist, categories didn't exist.

4:25Uh AI is going to do the same thing, but just ten times faster and ten times bigger.

4:31Uh we are in Internet 2.

4:330, Tim.

4:34We are in Internet 2.

4:350.

4:36Everything's gonna change.

4:37Um and uh it's a phenomenal opportunity to rethink what does the modern enterprise, what does the AI enterprise look like

4:45Yeah.

4:46It's a unique opportunity to think about.

4:48This is not just about analytics.

4:50This is about what does the AI enterprise look like

4:53Um and that opportunity was just on my head and I was just thinking about it and uh ThoughtSpot uh kinda blipped on my radar, which I knew about the company.

5:02It's as you said, it's been around and like all good things in life has had its journey.

5:08Um, which is fantastic.

5:10But um what I realized is we are on the cusp of an upgrade super cycle, a generational change.

5:19Yeah.

5:19This is not about the next version.

5:21It is about the next category.

5:23Yeah.

5:24And there's a little difference in that, right?

5:26This is not about I'm giving you a next version BI

5:29This is about this is the next category.

5:31It's like moving from flip phones to smartphones.

5:35Yeah.

5:35You know?

5:36It's like, yeah, I mean

5:38Calling is no longer a feature, it's a commodity.

5:41Um and then that brought me to ThoughtSpot because ThoughtSpot's always had the vision of a profound vision of why can't I just search my data like I searched Google

5:52And uh the company has been pushing that hard and I'll tell you as we get into the conversation, I'll tell you some super secret sauce we have, uh which is very exciting.

6:01Uh but um the the LLM happened.

6:05And now everybody wants to talk to their data, obviously.

6:08Right.

6:09Everybody is just putting their LLM on the data, and uh that's not gonna work

6:15And uh we have a unique opportunity with ThoughtSpot.

6:18We are uh uh to go build that.

6:21But more importantly

6:23I came here because I truly believed him.

6:26Uh I'm very grateful for my time at Salesforce.

6:29I'll tell you I learned a lot there.

6:30I have so many friends there.

6:31I'm so grateful.

6:33I'm really, really grateful for our time there

6:36But this was my chance to assemble a team to build an iconic company.

6:43And that's why I came to Thoughts Fot.

6:45And that's what we're gonna be doing

6:48Exciting.

6:50Very inspiring.

6:51Um, Francois, uh a big big cheese to fill there after that entry.

6:56I'm not going to try so it's okay.

6:58I'm fine.

6:58I'm used to anyway.

7:00No worries.

7:02No, I mean you know, for for me the data has been something s interesting since day one.

7:06I I

7:07I love the fact that when you are touching the data, you are touching every department in the company.

7:12You are not stuck, you know, in in one region, but you are really across the company.

7:16So you are really like

7:17It's it's uh fundamental positions, fundamental um functions, which I think is strategic, right

7:25And why now is because now is really like some we can do stuff that were not possible before.

7:32And now we have the technology to really like

7:36Like for you know like promise has been forever, like self-service, but self-service was you know more oriented like analyst at the point.

7:43Yeah.

7:43Now we are finally at the point where we can serve business and users with deep insight, with really answering their real questions.

7:50And their real question is not like, you know, what is my dimension and measure by this, you know, numbers and this formula field, right?

7:57They are question are very business question and now we can do that.

8:00So it's it's uh it's it's not uh there is no better time, I will say.

8:04Amazing.

8:05Amazing.

8:06So um yeah, I guess I guess we better get into it.

8:08Like what is Thoughtspot?

8:09We've ha I've heard about it for a while and I think Katana you touched on this concept of search.

8:14That has always been sort of top of my mind.

8:16No, Thoughtspot is a search focused

8:18sort of company.

8:19I also know that um you acquired a company called Mode um like a a few years back, right?

8:25And so that's sort of another touch point where I heard a ThoughtSpot.

8:28But more recently I've been hearing more about Thoughtspot.

8:30So I thought when I was talking to Louie, I thought meh, I need to rediscover what it is.

8:34So yeah, I guess how how would you describe ThoughtSpot to to my audience?

8:37And rediscover is the rediscover is the right word because it's a completely new change.

8:42Look we

8:42Fundamentally, CloudSpot is an enterprise data NAI company.

8:47And our passion and our mission is to help people

8:52grow their businesses with their yeah.

8:58We help companies grow.

8:59That's our primary focus.

9:01Um you know I would love to like maybe I can show you a couple of things to orient your great uh way to to to talk and speak at the same time.

9:12So let me know if you can see my slide, okay

9:15Yeah, absolutely.

9:16It's nice and clear.

9:17Okay, great.

9:18So hey look, listen, uh thank you.

9:20You know we've got uh everything we do, uh I say this every time, Tim, all my best ideas come from my customers too.

9:26They're just we're just so lucky, we're just so lucky we've got the best customers in the world.

9:31The best customers in the world.

9:32You know, I don't know if you would know

9:34But ThoughtSpot's been on a tremendous upswing in the last two years.

9:38It's like the LLM was the best thing that ever happened to ThoughtSpot.

9:41Because ThoughtSpot origination happened with the idea of why can't I search my data

9:47Now everybody wants to talk to their data.

9:49So we are like, yes, finally everybody wants to do what we've been trying to do, which is great.

9:54Eight out of the top ten healthcare are thought spot customers and

9:57Uh our our number of users have doubled in the last twelve months.

10:01Wow.

10:01I have three point seven five million users now.

10:04And uh

10:05Touchpot's one of the best stories out here right now.

10:08But I'll tell you the best part of my my my company, why I'm so proud of what we are doing together is my team.

10:15Uh we've got a fantastic global team.

10:18We've got a large team in the US, we've got large team in London, we've got a big presence in India, in Hyderabad, in Bangalore, in Tiruvantipuram.

10:26Uh we've got an office.

10:28Uh we've just uh got a beachhead in Sydney.

10:30We've got a beachhead in Tokyo.

10:32Uh we just opened we are moving into a new office in Chicago uh in a in the next couple of

10:38Um so the these these are the people who inspire me every day.

10:42This these are my team members all over the world, Tim.

10:44And I and I think so this is TopSpot.

10:47If you want to know what is TopSpot, it's a group of people.

10:50Who are just passionate about doing making data a growth lever for your company

10:59That's what all these people are passionate about.

11:01And Hotspot's also a place where people build careers.

11:04You want to know what Hotspot is?

11:05This is where ThoughtSpot is, you know, people

11:08Last 12, 18 months, a lot of babies.

11:10We've got a lot of new spotters in the company.

11:14I was gonna say what's the name for those spots is great, yeah.

11:17Yeah, and we even have onesie.

11:19Like you know, it's it's just a fun it's fun.

11:21We are trying to build something special.

11:23We are not perfect, Tim.

11:25I'll tell you one thing.

11:26We have a lot of things we need to do better.

11:29And I'll I'll be the first one to say that.

11:32But I'm so proud.

11:33If you want to know what is ThoughtSpot, this is Totspot.

11:35This is TouchSpot team.

11:37This is what Totspot is.

11:38And if you want to know what is ThoughtSpot, it's one team

11:42uh focused on this mission.

11:44Uh to make the world more fact driven.

11:46We are a mission-driven company.

11:48Uh uh this is our mission to make the world more fact driven.

11:51It's a little utopian mission.

11:53It's a little

11:54um aspirational, if you may.

11:57But that's it's fun to dream, no?

11:59I mean it's Yeah, no values a kid.

12:02Uh but we really think data has the power to do more to the world than what we think of

12:07Um and we are trying to build a company that's uh that's grounded in core values.

12:13Our core values are trust, which I learned from Salesforce and I should give them credit for that.

12:17I should give Mark Benioff I should give Mark Benyoff and Parker Harris credit for this.

12:22Uh trust me, uh I learned this.

12:24I learned the importance of trust from both of them.

12:27Uh and and and it's very important.

12:29Nothing is more important than the trust of our customers, uh our partners, our our ecosystem, our employees.

12:36And we are obsessed with our customers.

12:39We try our best to do the best.

12:41And when we make a mistake, we own up to it and we go fix it.

12:45Um we are we we are in a hyper fragmented industry.

12:49You know how much competition I have.

12:54But we out innovate.

12:55You know, we outinnovate and uh look we're a we are kind of an intense bunch.

13:00Tim?

13:00Yeah.

13:01Uh I would say we had a little bit of an acquired taste.

13:07Uh move really fast and uh we are taking the world into this new future of autonomous.

13:13Right

13:13We really feel uh data and AI needs to be not just about helping you do your job, but actually doing your job.

13:22Uh and and we think this is the future.

13:25We think the future is everyone will have AI agents.

13:28The future is not the future of BI.

13:29As you said, this is the next category.

13:32Is the real-time intelligence for everyone where workflows are autonomous

13:36Where every application has inbuilt data and AI flowing into it.

13:41So that you're not going to a dashboard on the side

13:44It's like this Tim.

13:46You know, if you're driving your car, do you want the GPS in a separate device in another car or do you want it in your dashboard?

13:54Well, should be in my dashboard.

13:56But think about it today.

13:58We put dashboards on the side somewhere else.

14:01Um so we we are rethinking all of that.

14:03We are rethinking uh what is called the boundaryless data

14:06We think the world of structured, unstructured needs to blend and Frank Ranashu show of this.

14:12And talk sport is these three things.

14:14We think data drives the AI enterprise.

14:17And our number one product is our agent tech analytics product and uh Francois will show you that.

14:22Our number two product is our embedded AI product, where we are part of other people's applications, Tim.

14:29And I'm telling you, Tim.

14:31This is the hottest part of my business right now.

14:33It's rowing gagbasters.

14:36And kind of funny, maybe we just got lucky, you know.

14:39But everybody has realized that if I have an application

14:44and I'm giving it to my customers, it absolutely needs to have an interface where they can talk to their data.

14:50Yeah.

14:51I mean you cannot ship an application today without a talk to your data interface.

14:56And that they just embed it with PodSpot.

14:58And on May 1st, we are not going to show you this today, Tim, because it's not ready.

15:02The team is still building it.

15:05On May 1st, we are launching our Agent Spot.

15:08So very quickly thirty seconds agent take analytics it's all about helping you.

15:12It's very simple.

15:13Just tell your agent to do it.

15:15That's what we think.

15:16Uh and got spotter agents that Francois is gonna show you for an agent analyst.

15:22This is all about augmenting you with agents

15:26To become a super.

15:28It's about giving superpowers to your team.

15:31Francois will show you Sparter.

15:33Sparter model, our data modeling agent.

15:35The hardest thing now is semantic modeling.

15:38And Francois will show you that, Tim.

15:44Visualizations.

15:45I think so a breathtaking.

15:47Tim, do you know we've built our own visualization platform

15:50Uh we don't open source.

15:51We we we use we've got our own, we call it Muse.

15:54Uh so we built our charting library.

15:57Um our pro Another Studio Anal Studio is a product that we have which allows you to prep and get your data AI ready.

16:04And auto semantics is the foundation at all.

16:08And then we've got our intelligent applications.

16:12You know, that is the embedded product.

16:14uh which is all about driving intelligence inside your application to build intelligent applications.

16:21So I'll give you a great example.

16:23You probably have heard of a company Guidewire.

16:25Guidewire is an insurance company

16:27Uh the the analytics inside of Guidewire is Topspot.

16:31Or if you go to uh uh Thrive, Thrive Learning, there in UK.

16:36Uh the handle inside it is Totspot uh and many, many, many more examples like these.

16:42Um and finally, uh, you know, this is our agentic uh layer, this is the enterprise agents that we are building

16:49Because Tim, what we realize is a lot of people are building agents out there, but most of the are just confident idiots because they don't have access to people

17:00Yeah.

17:01And I think so the data plane, the data plane is gonna be the critical plane.

17:06So uh we were what to show you uh in the problem

17:10Francois will find more demo this for you and then we'll continue the conversations.

17:24You have I just I just understood like the semantic components, the m the visualization components, but yeah, you've you've you've clearly been busy and I think that's part of the rediscovery, right?

17:34Just showing showing the breadth of it.

17:37Yeah, and Tim for the benefit because I kind of sped through it because I want you to see the real product and not the slides.

17:43Yeah.

17:44I'll give you the slides.

17:44You can link them for your viewers if you want, like up to you.

17:48Absolutely.

17:48Yeah

17:48They can go into that.

17:49Yeah, absolutely.

17:50Absolutely.

17:51And from a business standpoint, just to put a pin in it, from a business standpoint, look, uh we are

17:57Uh we have more than thirty-four customers now, million dollars of ARR and above.

18:02We are in every geo.

18:04Um and uh it's a fun year to be.

18:07I'm having a lot of fun.

18:10Good

18:11All right.

18:24Let me share my screen.

18:28So here we go.

18:29So you you um you can see the hotspot here.

18:33Yeah.

18:35Um so the first thing you know I'm going to start with is uh is the data, right?

18:39Because without data you have nothing.

18:42And so the first way there is basically two main way to to move data into such spots.

18:47The first one is

18:48Through what we call Analyze Studio.

18:50And Analyze Studio is basically the integration of the mod technology into SetSpot applications.

19:00I have a pre-created like a banking dataset.

19:05And I'm going to click edit here.

19:07And on this banking dataset, what you can see is I have one table

19:11Beautiful table, right?

19:13And this table comes from a Google Sheet actually.

19:16So very easily you can upload a Google sheet

19:19The Google Sheets, so the data will be bringing inside an studio and you can sync it.

19:23So every time you change the Google Sheet, the data will be reflected here.

19:27You can uh you can view it, you can edit it in spreadsheet, so very uh very useful features.

19:34If you want, you can you know complete and create, for example, uh calculated field if you want.

19:40Boom.

19:41Oops, you can see the preview.

19:43So very WYSIWYG uh things to bring data in.

19:47Uh if it's coming from a Google Sheet or CSV and so on.

19:51Then

19:52You have uh you know other things right here.

19:55I brought the data from actually a SQL query.

19:59I run a SQL query through my uh Snowflake connections

20:04And though I can bring the data in the same way I can see it.

20:07I can even create some some quick report if I want to.

20:11But what is really cool in this thing is because you know the data can come from different systems, right?

20:16It's not all the time in the same locations.

20:19So what I can do here and what I have done is I actually have merged uh the two together.

20:25So I have I'm creating some kind of new model, new dataset.

20:30Which include my data coming from my spreadsheet and the data coming from my Snowflake connections.

20:36And you can here I can just write a standard SQL and I have I'm going to have this uh beautiful table

20:43And when this is ready I can just push it and publish it to uh for analysis inside inside hotspot.

20:50So that's kind of like the first way to bring data in inside hotspot.

20:53All this data is going to be cache in our sortspot layer, and then every query, every uh conversation will be uh available through uh through these cache uh applications.

21:05Right

21:06And and does that update um let's say it's it's live data, does your cache sort of is it smart enough to to go back and check?

21:13Hey, should I query live or should I just use my cache?

21:16So if you do this way, it's more like because you have decided to use the cache, so you can create a special and you can say, you know what, every hour just refresh my cache every every day or so on.

21:27Yeah.

21:27Like an extract, yeah.

21:28Yeah.

21:29So that's kind of exactly like an extract.

21:31Exactly.

21:31Like that.

21:32So that's kind of like the first way to bring the da that is.

21:35The other way is

21:36you know uh the one that most of our customers are using at 80% is more like direct query basically right so in this case I go back to Sartspot I'm going to create uh what we call a semantic model

21:48So I can bring it by the way I can connect to DBT, I can connect to Snowflake Semantic View, I can connect to Databricks Unity Catalog, or I can just create it directly inside hotspots

21:58So here in this case you select the connection, super easy.

22:01I'm going to create my banking dataset.

22:03I love banking.

22:04Everybody likes banking, I guess.

22:07Especially when you have a lot of money there.

22:09But anyway, I can you know start to manually create my semantic model.

22:12So I can drag and drop my tables and I can do the work, but

22:16You know like we are living in a world where it's great to do manual stuff but you know, why not asking an agent to basically do the work for you

22:24And so here, uh using Spotter model, I'm just going to ask him to create a model uh that will allow me to understand the customers, the usage of credit, saving account and loans

22:37And based on that, uh is going to basically look at all the different table available through the connections, uh all the different color.

22:47You're connected live to Snowflake

22:49Yeah.

22:49And boom is going to select for me like the table I should bring inside my semantic model.

22:56Right.

22:57Boom, here we are.

22:58Yeah.

22:59We get some recommendation of some factables with some, you know, uh and I'm just going to add it to the canvas

23:08And then I'm going to like this very simply to create my my full semantic model through the process.

23:14So now I'm going to ask for the supporting tables

23:17Boom again is going to look at all the catalog, all the available things, and is going to kind of like do the work for me.

23:24So things that was taking like you know quite some time is uh

23:29Now is like kind of uh quite simple.

23:33My favorite step actually is the next step.

23:36Because you know.

23:37I was gonna say, does it connect them up?

23:40Yeah, exactly.

23:41Thanks you for us.

23:46Exactly create the join.

23:49Boom.

23:50And and you know the join is like simple but not simple.

23:54Um context.

23:56Exactly.

23:57Which key, which direction the joint should go, from you know table A to table B or table B to table A, then you have like one to many, many to one and and so on.

24:06So like all this work is basically completely like uh suggested by by Spotter Model.

24:12So it's become super super simple.

24:14What is really cool by the way also is you can use a UI to do the work, but I can do the same in cloud.

24:21In cloud we have also MCP server.

24:23And through that, I can just include say, hey, you know, connect to uh Snowflake, look at the table, now create my model in in in in SouthSpot, and you can do that.

24:32So

24:33We are you know every feature we are building really want to make it super open for people because you know the paradigm is changing.

24:40More more and more people are living now is their uh is their cloud or or open AI

24:45And they are using that.

24:46So we really want to to re basically let people decide where they want to do the work.

24:51So here you have all the joints, you can see like the keys, the found yeah, you can see the type of joints and so on.

24:57So I just can add to the canvas.

24:59What is cool also here is at any point I can just use a UI, right?

25:04So it's not like I have to either just use an AI or just a manual work.

25:09I can just you know do uh back and forth

25:12Oh wow.

25:19That's super interesting.

25:20I didn't expect it to um create multiple sort of connection points, you know, if that makes sense.

25:26Yeah, th th that that that inherently suggests that I can use different parts of this model to answer different types of questions depending on the way the links are done, right?

25:34And and the starting point.

25:35Yeah.

25:36Right.

25:36Yeah, and I think the key point of this semantic layer like that is also you can query one table individually.

25:42You can query two tables, three tables and

25:45And this is what we have built.

25:46We have built an abstraction layers on top of that that you don't have to think about the SQL how you want to create it.

25:52We have created what we call the search token that kind of like abstracts this complexity.

25:56But at the same time, because we have put in place like chasm trap, fan trap, people cannot make mistakes, which are super important.

26:04Yeah.

26:05Amazing.

26:06That's super exciting.

26:08Yeah, so just you know to finish on that, because semantic is more than just junction and table.

26:13It's also about the context, right?

26:15The how you can bring more context to the agent that you can really answer uh accurate uh accurate answer.

26:21And this so we have created multiple systems, for example, one of them is AI context

26:26So AI context is something that we're generating automatically, which is going to bring to Sparter more context about every column that you have in your model.

26:35Right.

26:36And we do that automatically because you know we find out that creating and generating like two hundred description humans are not really good at that.

26:44They are going to contradict each other at some point and so on.

26:47So we we really like uh just wanting to make simple and we are generating that automatically.

26:52Yeah.

26:53And uh of course you may have rules in your business so you can use instruction again adding more context to your model

27:00uh about the way that you should calculate things or what is the definition of a red account or you know like triple knowledge that you can add at the model level.

27:09And finally, the other thing that we have which is really uh important is the concept of memory.

27:14Because again, context, you need to be uh able to learn, to self-learn.

27:19And so we have multiple concepts of memory.

27:22One is learning from a dashboard, because a dashboard is a collection of queries, so we can learn from that.

27:28Learning from every conversation also that you may have with Polar and so you can manage all your memory concept here through the through this UI

27:36Right.

27:37Amazing.

27:38And I guess does that does that does that help analysis um kind of across multiple different types of sessions in context?

27:46And like so what's the what's the ideal use case for that?

27:48Yeah.

27:49Yeah, I mean the whole idea exactly is that like it's cross-conversation, uh cross users to be able to share common knowledge about your business.

27:56Right.

27:57Some of the knowledge is at the model level, some of the knowledge is at the agent level, some of the knowledge is at the user level.

28:04And you may want to apply different levels, and that's why we have these different layers of context per se.

28:09And and I guess uh different roles, different personalities, personas, sorry, can can contribute to that and then you're doing the job of organizing that and then surfacing it to the rest of the organization.

28:19Perfect.

28:20Okay.

28:20Yes.

28:20And leveraging as much as we can from what we learn to others.

28:24when it applied to other people basically.

28:26Yeah, yeah, yeah.

28:27Makes sense.

28:28Amazing.

28:29Yeah.

28:29So that's kind of like, you know, how you create a semantic model, it's how you load the data.

28:34So either like through caching or through direct query

28:37I should be clear, uh it's not really data modeling, it's more like connecting to data and and setting it up for the next step.

28:43The data modeling is a little bit more involved than that, but yeah.

28:46Yes, correct.

28:47It's like creating a representation of your all your tables that you have created in your cloud.

28:52It makes sense.

28:54So then of course in such spots you can create a lifeboard, right?

28:58Dashboard that's what we call them

28:59uh very you know unique standard like you know great feature that you know everybody is still like uh very attached to uh so you can see our native libraries that we call Muse where you can really like you know

29:12Do uh cool stuff.

29:13Um you you can uh dynamically, you know, you apply filters, you can uh

29:20drill uh you can drill down on every like bar chart and and go to the next question and so on.

29:26So every like feature that you have like alerting, like subscribing to it, so yeah all the things that people expect are there and available.

29:35Yeah.

29:35But what is you know of course if you want to create that again, you may go through a complex process, right?

29:40So if let's say I want to create a new lifeboard uh similar to this one, I'm going to give it a name here.

29:49And I'm going to be able to start and create uh create my my all the different visualization that you saw on the screen.

29:57For that, what we are using is what we call the the search token.

30:00So search token is a way to express SQL queries.

30:03So for example, I can say multi-saving balance by uh customer state, boom, it's going to create me a query.

30:11I'm going to see it, it's great.

30:13You can see behind the scene by the way the the sequels that we have generated.

30:17But you know it's kind of simpler to see that than to see the sequel that has been generated.

30:23And then you can treat

30:25To look to your labor.

30:27And this is great, but again, we are living in a world where people expect uh and want to have more help.

30:33So that's why we have spotted this now.

30:35So Squadrele is boom.

30:37Um I'm just going to uh basically prompt my dashboard.

30:42I'm going to tell him like what I want to do.

30:44I want to create uh a banking dashboard and

30:47with two tabs and some description of the tab and it can be very broad or very specific based on if you really know what you want to create or not.

30:56Yeah

30:57And so based on that, SporterViz is going to take your request, look at the model behind it, select the right semantic level behind it, and create your different visualizations.

31:06So here you can see

31:08You have created eighteen visualizations, you have uh two tabs, you can see description from it, and now it's basically just doing the doing the work of the building.

31:17Yeah.

31:19Yeah.

31:19And is it using your own AI agent to do this or is it um because one of the things people sometimes have a concern about is, you know, what is the token usage behind this?

31:28You mentioned search tokens, so we can dig into that later, but yeah, what's the what's the actual process by which it's doing that?

31:34Is the AI driving ThoughtSpot or is something else going on?

31:38Yeah

31:39So basically it's it's really like, you know, first we abstract all the complexity of LLM and selecting the model they want.

31:45We so when people use hotspots, we take care of all of that, all of the wiring of selecting the right model

31:52uh using all the best foundational models that we have on the market and and we are you know basically packaging all of that together to provide the best experience which is in this case it's really about creating uh a dashboard

32:05Yeah, and so we don't have to worry about token usage or thing like that.

32:09It's our problem, not their problem.

32:12I see.

32:13Good

32:15And uh and so now it's doing the work.

32:17What is cool also is you can call spot of this on an existing dashboard.

32:21So you know like if you have an existing dashboard.

32:23Yes, and you want to make it better, but you don't really know how.

32:26You want to change the organization of it, you want to change the colors, you can also do that on the on the an extreme dashboard

32:34Yeah.

32:34Understood.

32:35So here's taking some some building time and you can see.

32:40It's always happens in a demo.

32:46But if you think about if you think about it like the time for me to create this one manually

32:51It's true.

32:54Yeah, yeah.

32:55Even just setting up, making the right decisions, and then um

33:00Kind of sequencing that, iterating through that.

33:03Um yeah, it's yeah.

33:05So it's actually done it.

33:06It's think it's just

33:08About to show it to us.

33:09You have creating thirteen visualization and now you have five have failed, so it's now he's retrying.

33:18Like a human, sometimes.

33:21Exactly.

33:21You try again, uh you try different way, if it's not working the right way, and uh and boom.

33:27So it's um

33:29It's really like a different way of working, I I think.

33:33Yeah.

33:38The models that kind of

33:39uh talk to themselves as a way of sort of I I guess simulating thought, but yeah.

33:44Um this is sort of what's going on here.

33:46Yeah.

33:46No.

33:47And I can open, you know, if you want to see like exactly what's going on.

33:50Each of the steps, exactly, yeah.

33:52Yeah, yeah, yeah.

33:53And these generate answers these generate answer snippets.

33:57What is that?

33:57What is what is what is that little um is it like a a tangent you can go off on or oh here we go.

34:03Yeah.

34:04Yeah, the other answers like you know, creating the answers and after is cre is um sorry, grouping.

34:08Oh, I see.

34:09Like deciding how to group the different answers together.

34:12Then he's creating tabs, then he's doing the style uh on the filter, then he finds out some uh you know applying like t the uh data labels or other styles you can see here, then you have your summary and so on

34:25Sorry's documentation along the way.

34:27Yeah.

34:27And your version you can do.

34:30Yes.

34:30Then I can continue, I can go back to the previous version and so on, but you know, this dashboard has been created right away in front of me

34:36Yeah.

34:37Really cool.

34:38I can again manually change it if I want to.

34:40Uh very simply uh two tabs, yeah.

34:43Yeah.

34:44Yeah, exactly.

34:45And I I can uh I can save it and go.

34:47Go ahead and save it.

34:48Yeah and my work.

34:49As a starting point, that's incredible.

34:51Yeah.

34:51That's really good because again that would take a lot of time.

34:54Yes.

34:54I can chime to my boss and say I can spend so many time to building it.

35:00This is where you are maximizing the token usage, right?

35:03And uh and you know now I have this beautiful KPI, I can create alerting, I I can alert on on KPI

35:11Yeah.

35:11I can subscribe to sell it by PDF and so on.

35:14But also every question I can drill down, which is you know I can drill down by other different uh level by product.

35:21So very dynamic dashboard.

35:23But of course

35:24What is really like interesting is on any step I can call spotter to ask follow-up questions.

35:29So from a BI, from a dashboard I have, I can then start a full conversation at any point of time and start a full AI conversations

35:37So that's the that's really interesting.

35:40This is the key part, Jim.

35:42I don't know if she it's a subtle part, so I'm gonna like just jump in because Yeah.

35:47Nobody like have you ever seen a chart and you never had a follow-on question?

35:52It just isn't.

35:54You know, people have a follow-on and a f you ask a question, you get an answer, you got two more questions.

35:58You got two more questions, you got two more questions

36:01In the previous generation of products, every question would become either a filter on the top.

36:07You you know those dashboards with thirty-forty filters on the top that you that your business stakeholders make you put out there

36:13Right?

36:14Because doing each picture is is is a question.

36:17And or it would become a copy of the dashboard, dash APAC, dash E me a dash London type something.

36:25But what we have done is we are like, no, no, no.

36:28The idea of self-service doesn't mean creating more dashboards.

36:32The idea of self-service is giving your user the starting point, which Francois said.

36:36And then putting AI in it, putting an agent in it.

36:40So if they have a follow-on question on that particular chart, they can just continue the conversation like what he's about to show you.

36:48Yeah.

36:48Amazing

36:50Yeah.

36:50Go ahead.

36:51I have a couple of questions, but I'll let Francois do the demo so that I don't uh spoil it.

36:58I can say like hey, um

37:01Can you uh tell me more about California?

37:06And what is interesting is you know again, you know, back to my point, like it's not BI question, it's very

37:11natural questions.

37:12You don't try to provide like measure and dimension and time series and things like that.

37:18It's just like going uh

37:20Very naturally and then spoiler is going to understand the question, is going to ground it to the semantic layer, to the context that we have in the system.

37:29And it's going to generate uh an answer for you, which is usually most of the time a graphic, because still people like graphic to to uh to results.

37:40But also it's going to provide you like understanding of of it.

37:43So like you know.

37:44Yeah.

37:45And more context.

37:46Yeah.

37:47And so, you know, like uh more information and then follow-up question, like hey, how you should look at that for the

37:52Like if you have more questions, these are the more questions you may want to ask yourself.

37:57So it's really uh I guess.

37:59Yeah, and I guess it's it's got semantic understanding.

38:02So it doesn't just rely on the data from the chart, it can go into other

38:06parts of the model and bring it in.

38:08Yes, exactly.

38:09Have the full access to all the data that is connected to to SATP.

38:12But then you know, people also they they like to just have like conversation starting really with uh with a with a with a bar, right?

38:21And in this case we have the second mode of spotter.

38:24So here for example, let's say I'm a you know I'm an I'm an analyst and I want to you know know more what question should I help my branch manager with

38:34And again, a lot of the problem sometimes is like how do you know the what are the right questions to ask?

38:39You know?

38:40Sometimes it's really hard to formulate.

38:42Also you sometime you don't really know the data available in the model and things like that.

38:45So here again

38:47It's what we call a data literacy skills.

38:49You can ask questions on top of your semantic level.

38:53So here I I know I can run some analysis on financial performance

38:57And here are the type of questions I can ask.

38:59I can do that on customer analytics, I can do it on operational metrics and all the question here that you can ask.

39:05that are grounded again to all the model and all the data available into this uh semantic model.

39:12So they are not like random questions you should ask, they are really grounded to what is available.

39:17Mm-hmm.

39:18Amazing.

39:19It's good.

39:20And it's it's it's quite nice to have this interface.

39:22I think people find this less intimidating than sometimes like a

39:25A full-on analytics solution and it i it can be I'm assuming this is what we talk about when you're embedding, you can take this and put it elsewhere.

39:33Yeah, yeah, exactly.

39:34Yeah, yeah.

39:34I can show you that actually.

39:35I have a a small uh small example of of that.

39:39Yeah.

39:39But then I can you know ask uh uh a query like my credit card balance by months over the last twelve months

39:46And so again, it's going to do the same process, try to find what are the best way to answer your question, grant it to your semantic model, creating a graphic and an answer based on that

40:00And boom, you have the answer, you have like the different uh graph.

40:04And what I really like uh in in Spoiler is the fact that you know when you start sometime in AI conversation, sometimes you would like to go back to the BI world, right?

40:13Yeah.

40:14And here in one click, if I click edit, I'm back to my standard BI product.

40:19I can still select measure if I want to.

40:21I can uh change the visualization if I want to.

40:24By the way, I can see the query also again if I want to.

40:28By the way, the query, like can you tell me from this query what we are doing on the graph?

40:35There's someone out there in the world who can.

40:37I'm sure that's your engineers.

40:40But but you know now if I show you the search token, can you tell me what is the query doing

40:45You know?

40:47Yeah, yeah.

40:47It's I guess it depends on the on the um the person, right?

40:52Because exactly, yes.

40:54You know, i the different modes suit different types of people, right?

40:57Like um

40:58This this is sort of probably close to the analyst.

41:01The previous window was closer to the everyday business user.

41:04Exactly.

41:05Um this is maybe a little bit even this whole screen is more BI developer rather than analyst and then

41:10The queries like UK warehouse like you know optimizer um going why are there too many CTEs in here or something like that.

41:18Yeah, this is kind of you know easier to understand definitely, the SQL much harder

41:23So yeah this this is really what we want to put here is the fact that like people when they see a graph they need to be able to trust it and the best way to trust is to understand.

41:33And that's why we spend a lot of time and effort to explain the query to people.

41:37Like you know, we are doing the sum of months security card, we are doing it by months, and we are filtering by the last twelve months.

41:44So it's really really easy to understand

41:46You can put it back to a dashboard if you like it by the way.

41:49But this is really, really cool for us.

41:50And then you have like the explanation of it and then you have the key inside, so it's kind of cool.

41:55But

41:55You know when you have on a graph like this, the first I mean one of questions that people may have, like for me for example, I I see these numbers increasing, right?

42:03So you're like okay why is the number increasing and again why not just asking the questions?

42:09Yes, what is uh the you know the credit card balance increase between February and March?

42:15And here again

42:17What is the value also of what we are building is we are relying on LLM for what is good at, but we are really on everything we have built in softspot for really what we are good at.

42:27So for example

42:28Back to the you know the search token here.

42:31We don't use LLM to generate the sequel.

42:34We actually use LLM to generate this this expression, this abstraction.

42:39I see.

42:41From this abstraction on our side we are generating the SQL query.

42:45So you know 100% you're 100% sure that for the same search token you will have all the time the same search yeah

42:53Yeah.

42:53Which is you know very important to be deterministic in in the BI world.

42:57And the same thing is here.

42:58When I do a Y, actually what we are kicking out is Y analysis.

43:03And Y analysis is an algorithm that we have on our side

43:07to find driver analysis, right?

43:09So we are not relying on LLM to do that.

43:11We are relying actually on the feature we have built in our code.

43:14So I think that the best, you know

43:17The best product sometimes is where you are using the right tools for the right task and you don't try to put LLM everywhere because it may be surprisingly bad sometimes

43:28Yeah.

43:29So here's interesting.

43:31Yeah.

43:32Like you see like the driver for why is the number increase and so on.

43:35So again, uh pretty cool.

43:38But uh you know I um

43:41I like to complexify a little bit the things.

43:44So uh because what what we have here also is we are connected

43:48We are connected to your other system in your enterprise.

43:51I see.

43:51I see.

43:52So we have connectors and and I'm going to play a little bit.

43:57So I will just go ahead and ask Spoiler if he can check my backlog is Azana.

44:02And look if there is any task assigned to me and I'm going to ask him just to run the task.

44:08So this is what he's doing now.

44:09He's going to call Asana, going to find where I am, he's going to look for the task assigned to me, looking at the descriptions

44:16And it's just going to do to run the work and the job there.

44:19So you need to find the job and I can show you my Azana.

44:22Here I have a task.

44:24And you can see in my task is I'm meeting with my branch manager next week.

44:28Build me a weekly monitoring framework.

44:31What are the most important five questions should be asking every week and about the credit card portfolio answer each one with live data

44:37So this is my task on Antena.

44:39Now if I go back, you can see that you know Spotify found the banking project.

44:44You can see it here.

44:46He has answered the question Find automatically the five uh questions running through spotter with a live answer

44:53And boom, I can see every there everything there.

44:56Eve even asking me like if I want to mark the the task as complete in the scanner.

45:00So taking action essentially, yeah.

45:02Yes.

45:02I can generate a report in Stack with this report.

45:05I can create a Google Drive if I want to.

45:07So

45:08It's really like actionability at the core of uh of the conversation, not just like driving inside, which is which is kind of cool

45:16And the last thing I want to l show you is is uh you know another skill that we have which is sometimes your questions are a little bit more complex even.

45:26So in this example here

45:28I want to be able to cluster all my customers based on their behavior.

45:34And so usually if you want to do that, what you will have to do

45:38is you will have to create a machine learning model.

45:41You will have to create a K-min cluster algorithm.

45:46So you will have to extract the data, you will have to create a Python notebook, you'll have to load your

45:52la ML library, you'll have to write everything.

45:56Then once you have run it, you will have to apply a business understanding of the technical cluster that has been created by your Python notebook.

46:06And then you'll have to share it with with your business and users.

46:09And here basically Spotter is doing all this work by itself

46:14So it's going to uh load the data that is required to get to be able to run the analysis.

46:19So you can see here is it's fetching the data.

46:22Then is uh using Python code to create the segment and is writing the Python code right away right here in front of us

46:30And once everything is done, it's just going to share with you like the result of of the of this um of this.

46:37Analysis, yeah.

46:38Yeah.

46:39That's amazing.

46:41That's really deep work as well.

46:42I I I guess um yeah we can come back to to this question later, but that that's like a a chain of different events, right?

46:49There's this there's a lot more

46:51Um it's not as simple as the Asana task, which is, hey, you know, do these three things.

46:56It's it's got to figure out what the steps, then execute each of those steps.

47:01access different tools that I'm assuming sit a little bit outside of uh ThoughtSpot.

47:06Um I don't know if Python is something that's built into the platform or

47:09Something you're doing in the warehouse in Snowflake.

47:12Um yeah, like be it good to understand how it's evaluating where to do that work.

47:16Yeah, actually Python is on our side

47:19So because again, also what we try is to have a platform which is you know allow you to manage your Snowflake data, data breaks data, uh Google, uh BigQuery, uh Redshift, whatever cloud data warehouse you are using

47:31You are you can use Hotspot and all the feature of Hotspot, whatever the you know, so it's really like uh on top of any biases form.

47:39Yes.

47:39Yeah.

47:40So here you can see the clusters that has been created.

47:43You can see each customer's how many customers per cluster.

47:46You can see the different characteristics of each cluster.

47:49Now you have detailed analysis.

47:51Uh so it's it's really like super um super deep in a way of what you can do.

47:57And again, you can then share, you know, I can say you know great uh create um

48:04A slack canvas with a result.

48:06Oh nice.

48:08It's funny because this would have taken like a consultant a week about a decade ago, right?

48:12Mike?

48:12Yeah.

48:15And it would have been quite expensive task button.

48:24But this part the slack the slack integration has been a game changer for my customers.

48:30Because what managed to do is bring actionability to where the team is working.

48:34Yes.

48:35And I have a question about this, which is a good thing.

48:37Yeah, go ahead, great.

48:38Jump in.

48:39Um i th there is there's this theme that I don't think anyone's discussing, which is like what is the future surface of work?

48:47Like, where are people

48:49Going to be executing their the work because it it seems like workflows are changing and there's a little bit of fragmentation, but when the dust settles

48:58the tools will will will congregate in one place.

49:01This is why I think you've built an MCP.

49:03Because you want to make sure that Thorsport can go where the people are doing their work, right?

49:07And so I I I've always wondered like what is like what is what is maybe your view on on this idea of surface of work.

49:13Maybe after this demo bit, but yeah, yeah, that was a question.

49:17Yeah we can so here you know I've uh created the slack so you can hear the reports generated by spotter directly based on the data

49:24grounded to the data and and so on.

49:27Just natively in Slack.

49:28That's amazing.

49:30I can share it and comment and you know.

49:32You know the as you say like the the the boundary of the work are completely disappearing, right?

49:37Yeah.

49:38Yeah.

49:38And just to answer your question because I think it's better to answer it in context.

49:41Yeah.

49:42I said like personally my job is not to decide for people.

49:46is to let them options.

49:48So I want to make spotter the most efficient and the most powerful agent, but also everything you see here, you can uh we have an MCP server so

49:59In cloud you can uh connect to spotter and in cloud you can ask your question and without using our UI you will have exactly the same insight because cloud is going to call Sotspot and Sotspot is going to provide the same answer

50:12Yeah.

50:13So I don't care where the work is going to happen, wherever it is, we are we we can help people where they are.

50:19There's one thing I want to add to what Francois said because because of the strategy he has

50:25It's really important what impact does that have, right?

50:29Because what what what is happening if you f if you contrast the world

50:35Previously, you would go to quote unquote Asana for your task, CRM for your CRM, go to some other system, then you would go to an analytic system

50:46So your day was organized by tools and not jobs.

50:51Yeah.

50:52It's a subtle distinction, right?

50:54Yeah

50:55But what you were trying to do was create one task which is I'll help me understand this customer's data, add them to a campaign and put them on a better NPS path

51:07That is the outcome you're hoping to achieve.

51:11So previously our days were not organized by outcomes, they were organized by tool silos.

51:17My CRM is my this is here, my ERP is here, my manufacturing is here.

51:22Now what has happened is because of the agent tech world, those systems can now be connected in a seamless fashion by outcomes versus technology.

51:32So outlined what we showed you was three systems coming together without anybody knowing about it.

51:38Yeah.

51:39I think so the where that happens doesn't matter.

51:42What's going to happen is

51:44You're gonna see silo tool silos fade into the background.

51:48Yeah.

51:49AI is the new UI

51:53And that's where work will happen.

51:55But that AI is the new UI will be everywhere.

51:57So it's not like one place where you would go.

52:00As long as it's organized by outcomes is what will matter.

52:03So much ambient.

52:04Yeah.

52:04Correct.

52:05Correct.

52:06Correct.

52:08Um last things I just want to share for fun is also we have a research mode.

52:12So in this case I I ask a little bit more complex question.

52:16I asked can you draft a plan to increase my customer base by 10% in the next six months, taking inspiration from GP Morgan Chase current premium crowd?

52:24that you can find on the web and my existing product banker that you can find in Confluence and build me a growth strategy uh for that.

52:32So

52:33Here is going to go through like first looking at the web for GP Morgan Chase because Potter have access to it.

52:40Then he was able to go to Confluence to retrieve my document banker, which describe all my different products in Confluence.

52:48Then it proposed like an a plan to build.

52:51Phase one, phase two, phase three, expected deliverable.

52:55And he asked me if I like the plan or not.

52:57And I say of course yes, because I'm lazy and I don't want to change it

53:01Uh and then just based on that is creating the full plan using all the tools, querying, uh, you know, obviously there's a cloud data warehouse querying.

53:11the web querying everything to have and to provide a comparative plan of of all my strategy to increase by ten percent.

53:18Amazing

53:19So it's really pledge.

53:21And here's, you know, it's take more time because obviously there is more thinking, more like reasoning uh about it.

53:28Yeah, yeah.

53:29Go make a cup of tea for this one.

53:32Exactly.

53:33The last thing I wanted to share with you actually was how you embed all of that, right?

53:37So here we have an example.

53:39Here we have a bank website, imagine

53:42I I'm Acme Bank and I can manage my customers, my account, my transaction is great.

53:47But now I want to embed uh SATSWAT into it.

53:51So as a developer, what every good developer will do is they will bring VS Code.

53:56Here it is.

53:58I have my website.

53:59I have cloud code on the side because again, you know, I love to get help.

54:04Exactly.

54:05And here I'm just going to ask actually Cloud Code, right?

54:08I'm going to ask Cloud Code to create two new tabs using Spottercode because Spottercode is again another MCP server that we have created.

54:16that basically teach Cloud how to use Sotspot.

54:20How our SDK is working, how our API is working, how you what are the template, code example and everything like that.

54:29And so that's you know what you can see here is is um is checking with Podacode MCP, is retrieving our documentations, our SDK reference, so I have to click yes.

54:40And now looking at all this information is basically going to modify the website to add and to embed our technology to this this website

54:50I see.

54:51So it just to again make side by side.

54:54Yes.

54:54Back to a point, you know, every function needs to have an agent to make them more efficient.

54:59Yeah, and you could use this in um cursor or anti gravity, whatever whatever coding tool you use, yeah.

55:06Whatever coding tool you use, yes.

55:08We're on top of any IDE basically.

55:10I see, I see

55:12That's pretty good.

55:13So he's doing his work.

55:14This is the really hard stuff actually.

55:16Um you know, I I've seen a lot of my my peers who were not developers.

55:22suddenly brave into this world because because of this little side window.

55:27They call it vibe coding.

55:29I actually don't think it's vibe coding.

55:30I think it's unfair to call it vibe coding because the people still have

55:34the intuition of what the right questions to ask are.

55:39So I treat them a little bit more like advanced in a vibe cater who's just like, uh, you know

55:46But yeah, it's interesting.

55:47Yeah, I agree.

55:48Yeah.

55:48Yeah.

55:48You still need to know a little bit what you are doing and what you want to do.

55:51Exactly, exactly.

55:52Exactly.

55:52But then you agree with that.

55:53And here we are.

55:54So this is not embedded.

55:55Wow.

55:55Okay.

55:56Yeah.

55:56My new dashboard is here.

55:57It's embedded.

55:58And you know, Isa again is a big part of our business.

56:01You can really like style everything.

56:02You can really make it your own.

56:04Because you know, when customers embed hotspots, they don't want to, you know, they want they don't want to embed hotspots, they want to embed it to look exactly like they are and that's followers that are

56:14Same way spoiler is here and you can embed it and you can start which uh with speaking with it.

56:18So then it's port spot as well.

56:21Yes everything is available through this um this uh VS code

56:26Okay.

56:27That's a really good story.

56:28And I think embedding embedding has kind of been a slap on opportunity for a while.

56:32A lot of a lot of vendors have tried it.

56:35Um, you know, with varying success.

56:36I think the the the thing that most

56:40companies have lacked is a product mindset to kind of really make these things come to life.

56:45But then on the other side, um I've always thought analytics companies have made the pricing models a little bit difficult to to to really

56:52uh allow companies to run with like I guess iteration and I think where where we are today AI is allowing for that iteration

57:00to happen a lot faster, which has brought the cost down, which has I think opened up the use case again for more people to discover it.

57:06So that's that's super interesting.

57:12I think so the business model matters.

57:14Like when we sell to our embedded customers, we are not selling based on user uh like number of qu that no, we are selling by consumption

57:24So if you if you have five users who access it, then you can monetize that data whichever way you want.

57:30So we are emanating the business model of our customers and that's giving them that

57:37It's the the value to do it has completely collapsed now because of the business model we have adopted.

57:42Yeah, yeah.

57:43If they're successful, you're successful.

57:45That's it.

57:46Like my and honestly, most of the time they are successful and it's interesting.

57:51When they grow, my revenue automatically grows.

57:55Yeah.

57:55It's it's perfect.

57:57It's a good day.

57:58Yes.

57:59And then and back to the embedded story, right?

58:01A lot of more and more customers they want to embed SouthSpot into their own agent.

58:05Because they have their own agent obviously they want to provide to their customers and this is where our MCP connectors, like here is an example of our MCP within uh cloud

58:14Uh you know, I have a question and then he's going to uh you know use that spot and is going to provide my answer.

58:22And so

58:22The same experience if I have the same question within third spot, I will have the same answer as I expect uh I uh ask in the MCP server.

58:29And so

58:30I love the you know again it's all about API, SDK, uh MCP to to make your product embeddable in somebody else product very seamlessly.

58:40I love that

58:43Amazing.

58:44Um thank you so much.

58:45That's been that's been an incredible tour of the product.

58:48Um I've got so many more questions, but I'm also very mindful of your time.

58:53Um

58:54I think if I if I could ask one one question, I guess before be before we go away.

58:59I think um there's a lot of focus on semantic modeling, uh semantics generally speaking, and I know that you're one of the early sort of contributors to the

59:07open semantic interchange, right?

59:09And I think what I'd be keen to understand is how how does how does um initiatives like that

59:16sort of open up possibilities not just for yourself but for your customers because I think one of the one of the things I've always struggled with semantics is that no two businesses are the same, right?

59:26And no no two

59:28uh problems are the same.

59:29Even if you have the same business type you might have the same problem manifest in a completely different way.

59:34So how do you build a standard

59:36that tries to solve for this problem when I think every B Ital also has completely different context, different ways of working.

59:43You have Sparta

59:45Salesforce has Genie, uh, you know, Sigma has something else, and Omni has, you know, something else.

59:51Like how how do you solve for that problem?

59:53Yeah.

59:54Yeah, I mean I think the first you know OSI I think has been created uh really for customer mindsets

60:01The fact that customers have multiple tools and the semantic layer um unfortunately is hard to have it stuck in one application because you are augmenting your semantic layer

60:13at a different level, right?

60:14It's part of you may augment it through conversation that we have and we are grabbing very useful knowledge and you want to catch it

60:21In your cloud at a house, you may want to store obviously the lay the the joint and the table definition and thing like that.

60:27In another tool you may also generate something.

60:30So OSI, the whole point of OSI is coming through a common language

60:35where you can describe your semantic layers with that can be understood by every vendor.

60:42Which means that us, if we export it as an OSI format

60:47I can upload it to another tools and it's not like I have to work one by one, one to one, you know, I have to be an integration with these vendors or these vendors and these vendors.

60:56It's like a common language that we are all speaking, so it's making it super easy to basically

61:01transport the semantic layer from one tools to another tools.

61:04So I think that's really the target for SI and I I am super super excited for the format and all the working group we are part on to really define the uh the amazing format.

61:16And and the other part other part of there is also our approach.

61:20This is the b this is the reason, right?

61:23We are we are we are not tied to any one stack.

61:27Yeah.

61:28Unlike all the other data products in the market, if you know what I'm talking about, right?

61:32They're tied to a stack and they come with a lot of deep baggage, each stack.

61:37Like you have for for you to do this, you have to do this and then you have to do this

61:41We we we we f we liberate our customers from that.

61:44We are like, okay, great.

61:45If you want your semantic views in Snowflake because you want to use Snowflake.

61:50You know what?

61:50We have set up a bi-directional sync with Snowflake semantic views.

61:54We'll just inherit it from there and you don't have to do the work twice.

61:57If you're on DBT, DBT is another great, great platform out there, right?

62:02We are bi-directional with them.

62:03And you're like, no, you want to have it in TBT, but add further when you bring it to Thoughtspot because as Franz said every layer adds more context.

62:12Yes, you can do that too.

62:13So this is a

62:15This is an architectural choice we have made.

62:18Is we are going to give our customers the open ecosystem to drive intelligence across their platform.

62:25And that's where our semantic layer is uh is really helping us.

62:29But the other part which Francois alluded to is the analytical expressibility that we have built in our semantic layer.

62:36Uh uh I I'll forward you uh uh a nice uh blog we uh one of my product leaders just wrote.

62:42Uh I think so should you should take a look at it.

62:45That will tell you how deep our semantic layer is.

62:47Yeah, please do.

62:49Yeah, it's very exciting.

62:51Amazing.

62:52Well listen, um I I want to be really respectful of your time.

62:55Thank you both so much for

62:57For joining me.

62:57I hope this is the start of something exciting.

62:59I'm gonna work with uh Lewis, Lindsay, your colleagues to t to better understand the product.

63:03Hopefully you see some more videos on my channel about Thoughts Fot and um yeah.

63:07If you see something, a video you wanna challenge my thinking, come back.

63:10We'll have more chats and

63:11Likewise, yeah, I'm sure I'll be in touch to to ask more questions.

63:15You know, this was our first meeting and it was more introductory to kinda give you a harbor tour, a very rapid one.

63:22Uh but uh very sincerely we appreciate the opportunity.

63:26Uh and and just thank you for that.

63:28And Francois, what a great demo, huh?

63:30Like you saw that demo.

63:32Absolutely.

63:32That was just incredible what he does.

63:34Incredible.

63:35Uh I've got a great team and um I hope to I'll send you some of the slides to will be helpful.

63:40Yeah, please do and we'll get that on the on the video.

63:42Um amazing.

63:43Thank you, Tim.

63:44Let's call it there.

63:45Take care.

63:46No worries.

63:47Bye