Name: Why AI is the new BI
Uploaded: 2025-03-03T11:00:32.774Z
Duration: 36 min 52 s
Description: Why AI is the new BI

Transcript for "Why AI is the new BI": Hello, everyone, and, welcome to our generative AI week, a full week of brilliant content to cut through the noise of generative AI and give you actual frameworks and playbooks to help accelerate your Gen AI journey. Just a bit of housekeeping before we get going. If you do run into any issues, do let our team know. There's a chat section, and we'll be checking the messages, and updating. So feel free to ask any questions in the q and a. And for additional resources or to sign up for the free trial, take a look at the doc section, on your right hand side of your screen. Just to briefly introduce myself, my name is James Smith. I run the ThoughtSpot business in EMEA. And so that you can get to know me a little bit better, I thought I'd share a little fun data point with you. I'm a gin collector. I, so I have over 252 different bottles of gin at home. And, yes, they are all open, and, yes, I have tried all of them. I'm joined today by the wonderful Addie McNamara. So, Addie, do you wanna introduce yourself? Hi, guys. Very excited to welcome you to Gen AI Week. My name is Addie. I'm a senior sales engineer here at ThoughtSpot. And my data fun fact is that, I love the Garmin app, but I didn't think that its, visualizations were quite detailed enough. So when I was training for a marathon, I exported all of my Garmin data and built myself a personalized marathon training plan in Python. Highly recommend. So marathon running and gin drinking. That's what you've, that's what you've got today. Great to have you with us. Cool. Let's dive in, to the to today's session on why AI is the new BI. So, a little bit more instruction for myself. I've worked in analytics for the last fifteen years, pretty much all my career, and I've learned more in the last twelve months than I probably did in the first fourteen years. And that's because, AI is, disrupting the market, disrupting the world we live in. World and technology innovation is moving so fast, and we all need to step up, we all need to evolve and embrace this paradigm shift or maybe face being left behind. We believe that with AI, you can truly put data in the hands of everyone in your business, see those sparks fly when insights are found, and you can make better decisions. So bringing AI and BI together is what we're gonna be talking about today. I'm gonna talk about, the importance of that, and then Addie gonna show how you can do it, with with our platform. So this graph just gives a little bit of perspective on the challenge that most data teams and organizations are facing at the minute. For a long time, through the age of reporting and dashboards, awareness of what could be done with data and the adoption of data driven decision making kinda tracked each other. Adoption was fairly low. Most organizations haven't got past a point where more than 20 or 30% of their business are using data for decision makings. Lots of organizations invested heavily in data platforms, data literacy programs to make people aware of what was possible. But broadly, in most organizations, they've had limited effect. But two years ago, the world changed. ChatGPT was launched into public consciousness, and the generative AI wave started. Now almost every business professional is using some sort of large language model in order to ask questions and use AI to be more productive. So the curve of awareness, the power of data has increased dramatically, and suddenly nearly all business professionals want to search their data. Just in the last twelve months, for example, how many of you have had the request made of you or made the request yourselves? I just wanna chat GPT my data, and that is the new reality we're all facing. But the problems of the BI world still exist. AI is not a silver bullet that solves all problems. In order to fully leverage AI, we still need to overcome a lot of the problems. How do we get adoption from people with new platforms? How do we transform our operating model so that BI backlogs no longer exist and self-service becomes pervasive across the business? And a big one when using LLMs for analytical use cases, how do we get the accuracy we need from the answers that we're asking so that business users can trust the solution? And that's a real critical one when using large language models for analytical use cases. The answer, we think, is you've kinda got to meet people where they are. You need a platform that's a complete intelligence platform that caters for the needs of every single business user, caters for the needs of your business team, data data people, data scientists, also people who who would class themselves as an analyst in the business, you know, whether they work in finance or HR or wherever they sit really, being able to answer the questions that they need and also meet the business where they need, meet knowledge workers and people who, haven't naturally worked with data previously, helping them put data in the middle of their decision making process. And you you need a complete platform that fuses AI and BI together that gives you the trust, the accuracy, and the adoption required to transform your business. So Thoughtspot has this complete intelligence platform. How do we do it? Well, with our platform, we try and deliver AI and BI to every person in the business, whether it's delivering intelligence into your workflow, of a smart application, meeting end users where they are, or giving you your very own AI analyst to help answer every one of your analytical questions using natural language. Building on that complexity, automated insights being pushed to every user, showing you things like change analysis, alerts, and outliers, or allowing you to drill anywhere in your data and uncover predictive trends with our AI augmented dashboards, all built on a suite of tools for the analyst, helping them prepare and integrate data and deliver real time governed data that can be trusted. So we're going to show you some of this capability. I say we Addie going to show you some of this capability, and a lot of our cape the capability is gonna be shown today is focused on our autonomous AI agents. We call it Spotter, and this is an AI analyst for every single person in the business. So an an analyst that can sit on your shoulder and answer all the questions that you might have of your data. We think that Spotter has a little bit of secret sauce, so I just wanna take you through that before handing over Addie to to show you how it comes to life. So before we get get started, it's it's really important to go through this context because, you're about to see a lot of demonstrations where people ask a question when where we ask a question in natural language and get an answer. But unlike a lot of text to SQL solutions that are in the market at the minute that integrate with large language models, there's actually quite a lot of intelligence happening under the hood of Spotter, that it's important to recognize. Because it's through this intelligence that we can generate that we can guarantee the accuracy and trust that's required to get adoption from your business. Because without trust, people won't adopt the solution. And without adopting the solution, then they're not gonna get the outcomes that they need. So it's really important to get this trust right. So here, you know, you can see on the left we're asking a question, and you can see on the right we not get to an answer. But what happens in in between? Well, actually so when we first ask a question, ThoughtSpot doesn't immediately translate that into SQL in order to generate an answer. Actually, what happens is it goes through our consulting phase of our system, where similar to what an analyst, we're trying to understand why you're asking that question and what type of question you're asking. Are you asking a how question, a why question, a what question? And depending on what type of question you're asking, ThoughtSpot will pick an agent or a skill required to get the answer you need. Next, we're gonna layer on business context. Again, similar to an analyst that might understand your business. We're gonna layer on business context, metadata, knowledge graphs, business terminology, all of the business nuances that are needed to answer a question correctly, like understanding the difference between revenue and sales as a terminology or understanding the that when I say this week, I don't want to count Saturdays and Sundays. All of this business context that's specific to your business is in is in Spotter helping you get the right answer to your question. Next, we go into the execution phase, and this is this is where thoughts will turn your answer into a search token that will then optimize and execute the sequel to answer the question. Now this is really the bit of the secret sauce because it's through these tokens that we can ensure a % accuracy in the SQL that's being generated, and therefore the answer that you get back will be accurate and trusted. And that's built upon a decade of patented technology and a decade of ThoughtSpot innovating in search and helping people use search to get the answers of their data. Now the final step when the answer is generated, it's generated using visual best practice, but more importantly, it's not a static answer. You can take the answer and interact with it, drill down, ask the next question, spot an outlier, understand it, interact with it, fully conversational, all with human verification and trust. So that's a little bit about sort of our secret sauce, and I just wanted to give you that context so that when Adi is doing this great demonstration that you're about to see, you can understand what's happening under the hood and why that's important. So those are the things that we think a solution needs in order to make AI the new BI. But I've talked enough, so I'm gonna hand over to Adi, and she can take you through, bringing some of this to life. Addie, over to you. Fantastic. Thanks, James. During this demo, there are a few things that I wanna show you guys, and it'll walk through kind of three main components. One is, around Spotter, our AI agent that James just mentioned. What are the latest and greatest improvements today to that, AI agent within ThoughtSpot? The second will be how you can embed Spotter into your own application either as a stand alone or by leveraging other components of, your own LLM or your own agent to enhance what Spotter brings to the table. And then the last piece will be our kind of Spotter of tomorrow. Where are we going with this product to really answer your more interesting questions about why you're seeing certain trends in your data or how you can make changes in your business to adapt to those trends. So I'll go ahead and share my screen quickly. And first, we'll take a look at, the spotter of today. So the spotter of today, as James mentioned, is really focused around natural language conversational analytics. Not just one question and one answer, but really having a conversation with your data as if you were having that conversation with an analyst. And to that extent, we've added a few more capabilities recently that really takes ThoughtSpot to the next level, specifically around data fluency and around, visual storytelling. So when I say data fluency, I mean, on this homepage here, I've got, some mobile network analysis data selected. When you log in as a business user for the first time, maybe this is a little bit overwhelming and you're just not exactly sure what's within this dataset or what kinds of questions you can ask. So I could actually start asking ThoughtSpot a question like what kinds of questions can I ask here? And what it's gonna do for us is basically go through our underlying data model. It might have multiple different tables underneath the hood, but it's gonna bring back some different categories of questions that I might wanna ask things about. It'll bring back some relevant columns, some example questions of things that I might wanna ask. So I'll go ahead and take one of these questions. What is the average total bill amount per customer? And ask ThoughtSpot to go ahead and fetch some data for me on that. Now this is where you can see that it's translating with our trust layer underneath the hood. What that means is that it's taking this question and using an LLM to map it to the correct parts of our metadata. Here it's got our total average total bill amount and it's pulled out customer name as, the attribute that we're looking at. Behind the scenes, we sent that SQL query optimized off to your warehouse to execute and then pulled back this visualization as a result. I could save this for myself for later. But really, like we said, this is about having a conversation. It's about iterating on our initial question. So I could say something like show me just the top 10. And it will remember the context of that initial question and pull back, the top 10 customers based on that average bill amount. Now this is very dynamic as well. I can right click on any of these data points and drill down into it to understand, additional attributes for this data model. But also something that naturally happens when you're speaking with an analyst is that you have a really specific way that you want to see a chart. So Thoughtspot will pick this best fit visualization, but we've also now added in this element of visual storytelling. So, James, can I ask you what is your either favorite or least favorite chart type? Give me, like, a a chart type you might wanna see. So I don't really have a favorite, but I do have a least favorite. So my least favorite is a pie chart. Okay. Well, I will show you this data as a pie chart, even though it's your least favorite. I can just ask Spotter here, show me this as a pie chart. And this way, we're not just iterating on the specific search tokens or the specific data questions that you're asking. We're, iterating on the entire visual, story and journey that your customers might want to take. So now I can see that Ruth Ruth h Smith had the highest average total bill amount, probably talking a lot on the phone lately. This is where we are today with ThoughtSpot, answering complex data questions, as well as questions about that data itself and questions about how I can visualize that data, to tell a nice story. ThoughtSpot out of the box can be embedded into your own application. And by that, I mean, you could take this search bar and stick it directly into your web application as is. But a lot of our customers are leveraging ThoughtSpot embedded in, slightly more creative ways. So a couple of the ways that I see our customers leveraging ThoughtSpot and embedded applications is, one, through what we call bodyless embed, and, two, through what we call data augmented generation. So I'm gonna show you an example of each of those, and you can see how, the agent reacts to your question in slightly different ways depending on how you implement this embed. First, I'll bring up our, bodyless embed. And when we talk about bodyless embed, behind the scenes, basically, what's gonna happen is when I ask a question of the system, your the agent is going to decide, is this a data question or is this not a data question? Is it maybe a question for our documentation, for our product specialists? Is it something that should be answered outside of the ThoughtSpot system? If so, it's routed to the appropriate place. If it is a data question, then it's routed to ThoughtSpot, and ThoughtSpot gives us back that robust data result. So here, I'll select our movie rental data here, and I can go ahead and ask ThoughtSpot. I can go ahead and ask ThoughtSpot, maybe, like, a more general question here. So what are some strategies to improve our movie rentals? I know the movie rental business is maybe not flourishing, at the moment with all these streaming services. So we wanna understand how can really we really improve, our rentals here. And the agent has decided that this doesn't meet the threshold of a data question per se. This is more, around, like, strategies in general. And so it's pulled back some different, options for me, maybe investigate promotions and discounts, loyalty programs, customer engagement. And so then I can come in and say, okay. Customer engagement is interesting. Maybe I wanna understand specifically our movie titles by the total number of customers. And now it's gonna think about this and try to decide again, is this a data question or not a data question? And this, I think, is a bit more explicit. Like, that is a data question. There's an answer as to how many, customers viewed each of these titles. So the agent has decided, let's pass this off to Spotter and then send, our prompt and pull back those search tokens again and get us this visualization as a result. Again, it's fully flexible. I could drill down on this data. I could ask a follow-up question here, but we've basically segmented our agent into data questions and non data questions. An alternative would be, maybe that initial question that I asked around, you know, what are some strategies to improve movie rentals? Maybe that is in some senses also a data question. So the second approach to, embedding ThoughtSpot within your own agent is to use data augmented generation. So to that extent, here I've got, some, auto insurance data for us to look at, and I'm gonna ask a really similar question. I'm gonna say, what are some strategies to improve our auto insurance sales? And what's happening behind the scenes here is that it's taking our question and trying to break it down into multiple questions that are actually data related questions that might help me answer this in a more tangible way. So instead of just generic advice about advice about customer loyalty and promotions, I might get some really specific questions here. What's our average total claims amount for customers in different age groups? That's gonna maybe help me, answer my initial question and come up with some additional strategies. So, Spotter has gone off and, answered those questions with data. It's built out a full live board for me here to help, give additional context to the answers that we're getting. But also we're, pulling back those results and we're able to make some actionable recommendations off of them. So here, we can see that there's drivers with previous license, revocations have significantly higher claims. So an actionable strategy would be to implement a risk based pricing strategy, rewarding good driving behavior. This is a really, I would say, complete and holistic way of implementing the Spotter platform into your own potential LLM or agent. All of this, is something that you could potentially build today. So with the spotter that exists in the ThoughtSpot platform. But what does the spotter of tomorrow really look like? The spotter of tomorrow is here to help answer these kinds of questions directly within the ThoughtSpot platform. So without having to bring in your own agent at all, maybe we wanna start answering some of those why or how questions. Here in, the spotter of kind of tomorrow, I've selected some car sales data, and I'm gonna ask ThoughtSpot a really simple question. I want ThoughtSpot to show me, our quarterly, sales. Behind the scenes, this is a question that ThoughtSpot today could answer. It's a kind of a what or a when, question about our data. What happened? When did it happen? But now that ThoughtSpot has pulled back this graph, I'm seeing some trends here that might be interesting to investigate more so than just kind of a what or a when question. I'm curious, James. Is there a question that you might, ask of this data if you were looking at this this graph as a result? Well, yeah. I was just looking at that that drop off at the end. So I think I'd be asking a lot why have sales dropped between q three and q four. Fantastic. That is exactly the question that I was hoping you would ask. Why did my sales decrease from q three twenty twenty four to q four twenty twenty four? And now behind the scenes, we're analyzing this data, and we're trying to find different attributes that might contribute to that change. So here, if I scroll up to where I asked my question, you can see that it's actually broken this apart into multiple different underlying questions. Did our selling price drop across model, across make, across seller? And then it'll go ahead and do that analysis for you. Try to pull back any results that are statistically significant. So you can see we had a big decrease in our, f one fifty sales during the time period selected. Maybe Ford as the model, specific makes or sellers that are really contributing to this change. And I even get, like, a high level summary down below that's telling me here's how much our sales actually decreased and that relative decrease as well as drawing my eye to some of the, underlying changes. This way I can make, real actionable changes in the way that I'm running my business to, hopefully, you know, bring that those sales back up during our next quarter. And then the last example of some of those more kind of advanced analytics questions that you might want to ask your data, If I launch directly from a live board here, I might be looking at these sales trends and have, again, like, a really sort of general question about my data. I might wanna ask, you know, this is useful for all of our different, executives. But as a sales manager in California, what are some of the trends in sales that I should be aware of? And maybe what can I do to improve on my sales? This is a really general question, but it's the type of thing that people ask all the time about, about their underlying data. Oh, hang tight. I'm just gonna select the correct skills here to apply on ThoughtSpot, and I'm gonna go ahead and ask that question about a sales manager in California trying to improve their sales. Again, not everybody knows the specific question that they're looking for. They just know the general, data trends that they wanna be able to see and some actionable insights that they wanna make off of that. So behind the scenes, again, we're using that LLM to break this more complex question apart into really discrete questions that we can take some more actionable insights on. So here, it's broken it across, it's broken my question down into several different underlying questions, and it's giving me a nice high level summary of our monthly sales, of some different sales figures across car car body types. So it's given me a full report, both in natural language and with those interactive data points to help me make some, informed decisions about how I might improve my sales. Down below, it'll give me that full summary, and then it's gonna prompt me, okay. Well, if you have time, you might wanna explore these additional questions that we could further iterate on. So this is really, I think, where, ThoughtSpot is going in the future and how it truly differentiates. You're no longer just asking what questions or when questions. You're really asking the why, the how, and, how do I improve for for the future. This I think is really, hopefully a game changer. James, what do you think of all of this? Exciting. We're certainly not in the world of BI anymore, are we? That that is AI at its best, the cutting edge of what can be done today, and it's really sophisticated stuff. I think, you know, as as a nontechnical business user myself, I could see how that could help me every single day. So, yeah, really exciting. Thanks for sharing, Addie. Of course. So I'll, I'll try and bring us home. So, obviously, what Addie showed you there, hopefully, hopefully, you, you enjoyed doing those that that demo. We've kinda zoomed through a bunch of different use cases there and kinda just shown a sliver of what could be done. I think it's worth pointing out the variation of those use cases are really important because when you're trying to fuse together AI and BI, it's about meeting the users where they are. So whether they are interacting on in, you know, the ThoughtSpot application or embedded within a Salesforce or ServiceNow application because that's where they make their decisions or interacting with a portal or on a mobile device. Wherever the business users are, it's important to meet them where they are with AI and with the sophisticated capability that Addie just walked you through. Now our innovation doesn't stop there. Addie shown you a little bit of what's to come, but throughout this year, we'll be continuing to develop our agentic framework to create what we're terming the autonomous enterprise. So think about having a bunch of line of business agents tasked at helping you improve your outcomes in your business. So I want to increase my sales by 10%. I want to reduce my customer churn by 5%. I want to take £200,000,000 out of my business. Set the agent that task, and it will go and work, and help you solve those problems. In your business, you'll have multiple agents all working simultaneous simultaneously, all interoperating with each other, helping you achieve your goals. It's a really exciting future, and it is just around the corner. So that's what you can expect for us in 2025, so watch this space. Now for the rest of this week, we've got a a stellar lineup. So, thank you for taking the time today. We are, coming up to time today, but don't worry. It's just the beginning. We've got such an exciting week ahead. We've got what I will say a whopper of a lineup for tomorrow. We've got Cindy Howson, our very own chief data strategy officer, talking about how to get serious about data strategy with the industry legend that is Donald Farmer and the wonderful Ari Kaplan, head evangelist at Databricks. So that's an incredibly, brilliant session about how to get your data strategy ready for AI. We've then got a customer story for you about someone who's doing this today. EasyJet Holidays, Tom Cronin, and our very own Maria Schell will be taking you through how to reveal AI driven analytics, has transformed the decision making at EasyJet. So do not miss that because EasyJet are really pushing the boundaries of what can be done with AI today, and it's a really good story. And then we'll have, Ricky and Aman on Thursday unpacking that easyJet success. So what are the frameworks? What are the methodologies? What are the playbooks that easyJet have used? What are the lessons they've learned? And what are the key takeaways that you can take and implement in your business today. So, a packed week full of content. If you have enjoyed today, please give us a feedback. And if you haven't, please give us your feedback. The feedback, you should be able to see is open, on the right hand side. It should only take thirty minutes to fill in, but we'd we'd really welcome, whether, your feedback on on today's session. Hopefully, you've learned a little bit. Hopefully, we've got you a little bit excited, and hopefully, we'll see you, for the rest of the week. But thank you for for for coming today. Addie, I don't know if you wanna say your goodbyes. Yes. Thanks so much, everybody. And if you guys have questions or anything, just please leave them in the in the chat, and we'll respond. Thanks, everyone. Yeah. Thanks, all. Alright. See you. Bye bye.