In this edition of Digital CxO Insights Leadership series, Mike Vizard talks to ThoughtSpot CEO Sudheesh Nair about how ChatGPT and other generative artificial intelligence (AI) platforms will transform data analytics.
Mike Vizard: Hello, and welcome to the latest edition of the Digital CxO Leadership Insights series. I’m your host, Mike Vizard. Today we’re with Sudheesh Nair, CEO for ThoughtSpot. And we’re talking about the impact ChatGPT and other generative AI technologies are going to have on analytics, and basically, how we might all get a little smarter as a result. Sudheesh, welcome to the show.
Sudheesh Nair: Thank you, Michael. Thanks for having me.
Mike Vizard: We’ve had analytics for a while. And we’ve been playing around with various machine learning and deep learning algorithms to help enhance that whole experience and make it easier for as many folks as we can. But how will these new generative AI platforms augment that? And what impact is that going to have? Ultimately?
Sudheesh Nair: I think it all depends on what people do with it. Because if you talk to people about, you know, about ChatGPT or GPT3, specifically, there are two camps. One camp that will say, “Well, it’s a fad, it’s just a glorified autocomplete. It’s a confident liar. It just keeps building statement based on what next?” And there’s another camp that says, “Well, it’s going to disrupt everything, it’s going to kill all companies and everything will be replaced with that.” I believe that both camps have some points right and some point wrong, right? My view is that it is an extremely impactful moment in the history of computing, just like, you know, internet to browsers to iPhone, even to Amazon is probably the best example. For the first time we are going to have AI capabilities on tap. It’s available through API’s. What do you do with it? It’s up to you, in a sense that, like electricity is a utility, you know, sometimes I can use it to charge my phone or other times it’s used for light, what we build will make all the difference. In our point of view, there is tremendous amount of value we can derive for analytics, when AI is used to close the last mile. So that’s what ThoughtSpot is focused on. And the opportunities abound.
Mike Vizard: It seems to me, we spend a lot of time looking for the right data just to analyze it. So do you think maybe generative AI is going to help us at least identify what data sets may be most relevant? And then we’re going to have to of course check the veracity of them. Because as you noted, that platform might be a confident liar. But at the end of the day, you know, can we just accelerate and reduce the toil that today, we encounter every time we try to build something interesting from an analytics perspective?
Sudheesh Nair: Yeah, you dug deep into the exact point where complexities lie. So for example, analytics is about crunching numbers and coming up with a precise answer. Right? If I’m asking how many of my viewers returned from last episode, or this episode – let’s say you are trying to figure out that and let’s say you’re a content creator and you’re trying to figure out which episode performed well and which one did not. And now you want to know why. Because you want to always constantly learn and improve. And if you’re asking a platform how many people churned between my last episode to this episode, I cannot give you 10 answers as a platform; I should be able to give you one precise answer. And it has to be believable, right? That is something to keep in mind that analytics, particularly with respect to business data, precision, accuracy, trust, these things are paramount. Whereas if you think about searching general, unstructured data opinions – Google in general, it’s not that you have to have a precise answer, you can throw a billion answers as long as your the answer that you’re looking for is there in the first or second, you’re good. That’s a huge hassle. So ThoughtSpot, for example, we started early on with the idea that we have to make sure that business users, you know, someone like you is a content creator. You don’t have to go talk to someone else to get the data on your episode’s performance. Because the people who are building that dashboard for you, they have no clue. What is it that you do? And how is it that you are getting people to come back; what sort of message are people trying to hear a lesson from you? They don’t anyway; you are the domain expert. The problem in this case, the creator, you may or may not be a data expert, you may not speak SQL. What changed in the last three, four years, you can now get a boatload more data than before, which means there is a lot more data that is streaming it from your mobile. It’s not like radio right? We are sending this through Spotify for example. You can see how many people watch, how many people skip, how many people watch 1.5x speed versus 2x speed, and all that data is now available. If you can take that data, derive those insights, pack it in consumable fashion and give it to you so that you can action on it, that will be magic. However, that is a big gap because the people who speak data and the people who speak business – there is this gap. And that has to be solved, and there, GPT3, and in general, generative AI, you know, when built properly with tools, like ThoughtSpot can play a huge part.
Mike Vizard: Do you think that this is just the beginning of this whole process, because there are other large language models, and they might be more attuned to specific use cases and specific scenarios. Open AI seems to have built a platform that just grabbed all the data from a certain date before then and kind of created a model around that. But will it become more useful as we maybe have datasets that we’re building the models from that are more precise?
Sudheesh Nair: So there are two angles here that both are important to talk about. The first one, one of our co-founders, Ahmed, he was at Microsoft, Bing for a while and then went to Google, and then founded ThoughtSpot. And we have a lot of former Google people. One thing you have to understand is that Google has been the pioneer in large language models for a long time. They have at least five language models, four of them are publicly available, almost all of them could be equal or better in terms of capabilities, compared to what OpenAI did deliver. The difference is Google cannot afford to take a huge hit in reputation if you know, the model calls a journalist, Hitler, for example, right? You can’t have that kind of risk. So they have a much higher bar to cross with respect to what is available, what they can put their brand behind. Now, are they being too conservative – maybe? And that’s part of the reason why the CEO essentially created what he called a red alert within the organization to amp up the effort. The second thing, which is an even more important thing, is why large language model trained on general internet and general data is useful. Businesses, enterprise businesses would want specific domain training, if a financial industry probably want to have models that are trained within it. Life sciences, clinical research, pharmaceutical trials, things like that there are lots of nuanced conversations and you know, language models that need to be built. ThoughtSpot – we are building a platform for data analysts to use ThoughtSpot user experience to train the model for data definition, building data or worksheet building, which means that they can move away from building these dashboards that no one really reads, No one gets credit for – it is just sitting there in 10s of 1000s of them, instead of simply improving the chat type, they can stop doing that and focus on building language models and training the AI so that their business can derive immediate value. So for me, those are the sorts of things that I’m excited about. And you’re absolutely right, this is just the beginning.
Mike Vizard: Do you think we’ll see a significant bump in overall productivity, because we’ve been talking about productivity and it for two or three decades, and that number doesn’t seem to improve all that much. So as we kind of look at these new gender of AI platforms, might we actually see an actual increase in the productivity for data analysts and things that you know, that they no longer have to do, and they can do something more interesting?
Sudheesh Nair: I don’t necessarily agree that the productivity hasn’t increased. What I will tell you is that the things that are standing in the way of increased productivity are often people and organizational politics. For example, I’m just giving you one example, GitHub and Copilot, you know, Copilot has become a really good way of helping code developers and developers trying to write a code, and almost all the time 80% of the code they write can be routine, things that they will return 100 different times – sub routines that they will actually cut and paste. What they do is they look for code that they wrote before Ctrl C Ctrl. V, that’s the life of a developer, almost all, it’s not that glorious. There are, you know, logic building, there is pseudocode writing, there are things that they do, but oftentimes it is boring, but GitHub and with the AI driven Copilot can take all of that and build a few beautifully written spaces or type whatever that you like, a code snippet for you and then you can jump into it and modify that. It is an exponentially better way of being productive. Another example that is VC constantly within hotspots domain. So we do two things: One is think of it like – let’s say you’re hungry and you’re in a new city. Okay? And you don’t know the restaurant scene. Ten years ago, you would go down to the lobby, talk to the concierge and find out, what should I eat? And what is the constraint? Do they actually have a book of maps and they will peel one off, draw a few circles on some Italian restaurants, and then say, “Hey, here is where you should eat.” We don’t do that anymore, right? We actually open Yelp and say, show your restaurants, I haven’t had sushi for a while, let me see sushi restaurants, I want to go – and then you make a decision by interacting with the data. I want to walk, I want to look at the price, I want to look at the menu. It’s a bespoke experience that you create for yourself, right? We don’t do that conscious thing anymore in our personal life. But when it comes to business life, people still go down to the data teams and ask for that dashboard. What ThoughtSpot does, we give that Yelp experience – you’re trying to solve a business problem, you interact with the data through natural language, get the answers, make decisions and go about your life. In this context, it is important to remember that data – people have a job, which is to curate the data, and then step back so that business users can freely interact with it. That’s where AI can make significant impact. But you know, what is standing in the way is sometimes, you know, in particularly in some old school companies, and even in all leaders, they think that complexity is job security. You know – if I make things too simple, I will lose budget, I lose people. So when people and organizations are standing in the way of progress, you will absolutely not see productivity improvement. But it is not the fault of AI. It’s not the fault of technology not being there.
Mike Vizard: Do you think that C-level, execs will trust the data more if it’s coming from some sort of AI fuse system, because one of the issues we’ve had over the years is people are dubious about how the data was collected or entered in the first place? Because they know how sloppy things can be. And then they’re reluctant to kind of take the advice or the recommendations being surfaced. So do you think we can close that confidence gap, and the data currently exists?
Sudheesh Nair: No, and no execs should trust the data blindly that is coming out of any AI system anytime soon. Because if you do that, you are going to be in a world of trouble. I can tell you how we are doing – so we have built a platform that uses – so for example, in ThoughtSpot’s case, it’s important to tell you the history of ThoughtSpot a little bit to make sure that I’ll explain how. So we wanted people to have precise answers by using natural language. The problem is natural language is difficult – there is this lot of different variants, right? So for example, I want to know what the temperature is outside, I could ask how hot is outside? How cold is outside? What’s the weather outside? What’s the temperature outside? The answer to all of these questions could be the same exact number, it has to be one number. So we tried to build our own language model four years ago, and we decided that it is if you try to piece all the different combinations, the answers will not be precise. So we decided believability trust is more than more important than anything else. So we compromised on a we did not follow through on the large language model and give everyone free language. And we decided to focus on accuracy. And then what we did is we went from SQL to what we call relational search. So think of like Amazon search – when you go to Amazon, you don’t do free natural language. You search, red shoes, white tie, like whatever. But when you go to Google, you obviously write the full sentence, which was who is the best rock singer of all time, right? When you do that. But the Google, like I said, can give you a billion answers. People always wanted the Google type free and natural language for a precise answer. So the way we have approached this, it’s an extremely interesting one. Now that we have this, figured out how to deliver keyword based search that creates exact SQL and executes them on Snowflake and data bricks, and gives you the precise answer. We nailed that. When GPT three came out, we immediately used that to close that last mile. But when you type a question to impart spot, we will take that question, but underneath it, we will transparently show you first the keywords they used. So if you say what’s the temperature outside, underneath that we’ll show you here are the keywords we used. If you’re interested, you can click on those keywords and it will show you the SQL queries that we use. And then when you click on that, it’ll show you the schema visually which columns which tables we pulled it from. So what we are doing is that when you have this it is a fundamental principle that people sometimes forget when you abstract things but get simpler however, they will also become more opaque. So if you want to make believability possible, you have to show show basically, you have to shine a light to everything. So every time you ask a question, you can inspect – so it is going back to Ronald Reagan – trust but verify. That’s the model that we are taking, you should not be blindly trusting the answers coming from any models anytime.
Mike Vizard: Do you think we’ll see a lot more regulation that requires that level of transparency? Because there are people asking questions about black boxes and models and analytics and how were decisions actually arrived?
Sudheesh Nair: I wish I could, I’m not an expert in that. I have an opinion, which is probably not as good as yours. So what I do think, is that this tectonic shift that, you know, open AI just unleashed on the system, people are absorbing it. And there will be significant amount of tools and hype and dollars flying into it, which means that there will be bad actors, there will be scams, there will be a lot of, you know, mishaps happening here, which will prompt government to act. Will the government have the right kind of regulations and rules? I don’t know. What I do think is that it is important not to panic, not to go back and forth on this because just like everything else that exists in our culture is absorbed, we are going to absorb and overarching positive things will come out of it. I’m very optimistic. You know, last week I was in Japan, you know, here’s a country that is aging so fast that they don’t have children, you know, people are not having kids. How are you going to take care of the aging population? I mean, imagine self driving cars and drone based delivery; these things are really critical when your population is really aging. And you know, they don’t want to drive or they can’t drive to even go to the 7Eleven nearby to buy a cup of coffee or sugar. So AI, improvements in AI, whether it is helping you code or helping fly and deliver a pound of sugar from your neighborhood store. These things have real world ramifications. So my hope is that regulations won’t be too draconian. But at the same time, it will create enough gap for experimentation while holding back the bad actors. Am I optimistic that will happen really fast? Actually, no. But I do think that there will be course correction both ways, and we will find the right equilibrium.
Mike Vizard: So what’s your best advice to folks? We’ve seen banks kind of ban GPT. And we’ve seen other organizations aggressively embrace it. Is there a middle? Or what should people be doing right now?
Sudheesh Nair: It is absolutely one of the dumbest things to do. I mean, the speed with which things are changing, you could be the world’s largest bank. But you have absolutely no hope unless you’re willing to embrace change that is coming your way. But this is also moving fast – like ThoughtSpot, we are not the tableaus of the world or the industry standard. We’re like I said they keep printing maps, and saying here is your data dashboard, data is changing. Are you going to select say you get one dashboard a day from Tableau? If data is changing constantly, are you going to make like 150 dashboards, that’s not working, right? So that model won’t work. So banks or insurance companies, they have to change. ThoughtSpot – our experience is that largest of large banks are embracing this change, responsibility, and really fast. Like we have customers like Bank of America, Wells Fargo, Capital One, PNC Bank, you know, these are all customers who have deployed for sport at massive scale – Nikes and Canadian Tire CVS, Verizon, why are they doing this? These are not new companies. These companies have lasted decades after decade; the reason why they last is because they are really good at figuring out changes and embracing them. It may not feel like it because of their size – it might feel like it is low. But at the top of these companies, they have some tremendously talented and forward looking people who are looking at these products and seeing strength and saying you know what, we can’t be an old school company. The rate of change, the cycle of creative construction is getting so short. If you wait for things to settle, the change would have taken something else from us. So at the top of it, the companies are really built to change and adapt and progress. The culture inside the company sometimes usually in the middle management. There are people who lost touch with customers and the mission of the company. And they are just becoming a viscous, you know, thing that is holding back progress. My advice is get with the program or let them go. There is no place for people who are resisting change and progress. When the whole world is moving forward. And consumers users are looking for better services from your business.
Mike Vizard: Alright, folks, you heard it here. The future is here. Whether you like it or not, so you might as well jump in with both feet. Hey, Sudheesh, thanks for being on the show.
Sudheesh Nair: Thank you so much, Michael. I really appreciate it.
Mike Vizard: And thank you all for watching this latest edition of the Digital CxO Leadership Insights series. I’m your host Mike Vizard. You can find this and other episodes on the digitalcxo.com website. Once again, thanks for spending time with us.