3,000 Tools and Enterprises Only Use 15% of Them

Sean Weisbrot: What's the hardest thing about what you do?

Saurabh Gupta: The hardest thing, Sean, is where we are at the stage to convince a customer that we have a solution which is better for them, and it will do it will not only do get them results faster, but also will be a much more efficient bet for them in the long run.

Sean Weisbrot: How do you convince them of that?

Saurabh Gupta: And, so this has been a big challenge for us, and, over the course of last two two and a half years, we learned. So initially, we used to go and we used to try to sell hard, and we we were not very successful. But what is working out for us now is this whole AI buzz is under tremendous pressure. There's lack of outcomes, and our positioning and pitch to our customers is that if you have a problem which you are struggling with, we can solve it in four to six weeks. Give it as a shot. And, most of the time, our customers really, really get excited or they give they get intrigued and they give us a chance because, they have been struggling it with that problem for maybe a year, year and a half. So six weeks or four weeks is not a big deal for them. And they say, okay. Let's give it a shot. And, we have been mostly successful. And, where we, really, really come out strong is when we focus on outcomes rather than activities. And most of the challenge that industry is facing, especially on the consumer side, customers, is they just get drowned in the bunch of activities that people keep talking about. So So we just go after the outcomes. We focus on the outcomes that really, really matter to the business, and we try to beat our promise of four to six weeks. Most of the time, we have done it within three to four.

Sean Weisbrot: Do you offer money back? It's a 100% money back guarantee if you're not happy in four to six weeks.

Saurabh Gupta: So most of the time, like, the bet is very small. So if the customer is not happy, we really don't want nobody. So then we'll be more than happy to go back and give a better solution when we are ready if we fail than trying to penalize the customer for not liking and still have to pay for us.

Sean Weisbrot: There's a lot of people if they if they, either they offer, like, a thirty day money back guarantee these are, like, you know, certain kinds of businesses. They'll say, if you don't get these results in ninety days, then we'll work with you for free until we get you there. Obviously, I know in your business, it's not easy to do something like that because it gets very expensive for you, but have you considered that?

Saurabh Gupta: So, Sean, I really don't believe in that. If we are not able to show like, a six week engagement,

Sean Weisbrot: Yeah.

Saurabh Gupta: our intention is always to deliver it in three to four weeks. In two weeks, if we are not able to figure out if we are getting in the right direction or not, then there's a problem with us. And most of the places, the the first couple of weeks are very, very decisive. They will tell whether we are going to get to the results or not. And the the problem could be ours that we had we don't have, we misunderstood the problem. We didn't don't have the skills to do it. On the other side, also, it's possible where a very, very likelihood we didn't understand the complexity of the customer's landscape, and that could be slowing us down. So for me, first two weeks, very critical. That should be the decision making point. And no one in the company is allowed to drag to six weeks just to make a point that we are trying to do something. No. And we walk out. That's that's the most important thing. We want the most in inelastic asset that people have, all of us have, is time. So we don't want to waste anyone's time on this.

Sean Weisbrot: Why should people care more about their data?

Saurabh Gupta: And why should not people care about their data? It's the most, most, most important asset. Today, literally, if you don't protect your data, someone is going to abuse it. It. Your data is going to be like, your personal data is very important. Your company data is equally important. How you can leverage your data to get a bot better competitive edge, increase the top line, reduce make your whole operations more efficient. There's a lot can be done with data. And what is also interesting that's happening is, like, we talk a lot about our old internal data. There's a lot of third party data also that comes in and starts influencing. There are the time where I can give you an example from my past where an organization where I worked with for for many years, they could save 14% of electricity bill just because they put sensors in there. So suddenly, if you are, like, still and then not and working and thinking, suddenly, the lights will go up. So you have to move around a little to make sure the lights get on. So I they were able to figure out, like, which offices that the ACs go at a certain rate, which a offices the lights go off at a certain time where they couldn't even switch on the corridor lights if 90% of the people are not. So I think that a lot can be done, not just from top line, but also from bottom line point of view. And, access to data and managing this data gives us multiple ways of sort of taking advantage. And with that whole excitement around AI and how AI can be is being used, I think data becomes even more critical. Where I see an area which people have to really, really put in effort is making sure the data is high quality. Those volumes of data can increase as much as possible. The quality is bad, and you put analytics on it or AI on top of it, you're not going to get good results. And even if you hope to get good results, you'll end up spending a lot of money on compute. So the overall foundation of data has to be done. Right? And, I I feel like, we are just starting. Every five years, I thought, like, we have reached a maturity in data, and suddenly you see another uptake. So I I'll be surprised if this data revolution is gonna slow down.

Sean Weisbrot: So for the audience, an example of, like, bad data management is when you and your team are, you know, viewing the pages of your application or your website, and yet you allow your IP addresses to be tracked. And so you say, wow. We have so many people looking at our website, but, actually, it's just you. And so you've, polluted your data with the wrong people's information. And so you have to find a way to remove yourself from the analytics in order to make sure that you understand what people are actually doing when they're using your systems.

Saurabh Gupta: That's a good example. I can give you another example of, actually, what is the cost of bad data to enterprises?

Sean Weisbrot: K.

Saurabh Gupta: So if you think about it, like, so when I was in IMF, International Monetary Fund, we collected data from 180 countries. And this get data was across different domains, balance of payments, national accounts, etcetera. And collecting this data in different formats, different structures, and it goes through a processing, checking, transformations, lot of value add, and getting to a place where it is shared. And there were occasions where when people start consuming data, they realize the data has issues. And there could be small issues like fat finger, some number got deleted in Excel sheet, or a decimal got shifted.

Sean Weisbrot: Yes.

Saurabh Gupta: But getting it caught on the consumer side means imagine not only the exposure you get, but all the effort that goes back again to the collection side. So the reason I brought this up is there is a lot of talks about data quality. And if the same bad quality data goes, AI is not gonna fix it. No one can fix it. There is someone has to check it. So moving data quality to the left, to the source, was something which everyone talks about. But is it really happening? What does data quality mean actually? It's these are all very, very comp get these things are getting more and more complicated. One simple example could be is, like, a population you gave an we were talking about, Portugal having 10% immigrants or a significant percentage immigrant. If suddenly that number gets messed up, like someone types in a wrong number and it becomes 20%. There should be some checks and balances around it. And and and that so how do you make sure that checks are there? And maybe that number is also right. So is there sufficient better data around it? Is there sufficient context around it provided so that data that looks bad has supporting details that, no. This is not bad, and this is right because of that, because of this particular reason. Suddenly, if a country allows a lot more immigration happening, you will see unusual spike. Then is it supported by a metadata? And is that metadata traveling with the data across from validation to checks to transformation and all the way to consumption? And not limited to just your website. All aspects like, all the activations, layers are getting the same input, whether it is through API, through analytics, whether the AI is getting the same. So I think there's a bunch of, complexity that is going to come in with increased volumes of data. And, a lot of it isn't getting tackled, but I think we'll keep getting surprises, and we have to be prepared for that.

Sean Weisbrot: How can people get closer to the source of truth for the data to minimize the that pollution?

Saurabh Gupta: So the best way to do it is get closer to the source. So if you are connecting to, say, your Salesforce data, bringing it into a data lake, data warehouse, some basic checks and balances should be put in white before even consuming it, before even moving the data. And if there are challenges and issues with the data, probably a call should be taken. And, again, data quality is something which is very use case dependent. In some places, you are okay tolerant with errors. Some places, you don't have. So, getting as close as possible to the ingestion point, check and validate the data is gonna be the key.

Sean Weisbrot: So, for example, like, right now, I get information from posthog, and I get information from, GA four, right, Google Analytics four, and I get information from CloudFlare.

Saurabh Gupta: Mhmm.

Sean Weisbrot: I don't know which one to trust because as far as I know, they're all the sources of truth, but they're all giving me different numbers.

Saurabh Gupta: So so I'll I'll give you an example from a mesh point of view data mesh point of view. Like, getting these data from different sources and creating a single source of truth is a process. So creating these source aligned data products, one is coming from CloudFlare, one is coming from Google Analytics, and having a bunch of routines run out to to validate which data is making sense and which is not, and then bringing it together into one source. I'll give you an example from, customer data. Like, as like, in a bank, they have multiple customer datas. Sean can have a mortgage account. Sean can have a savings account. And both the Sean's are not connected. So many of the banks are struggling with bringing a a customer three sixty view. And most of them realize later on because, the business units are trying to figure out how can they upsell, cross sell services and products to Sean. But the same thing, if it's brought to the left before and a single source of truth is created with the with the customer data, then they exactly know there's no point selling mortgage to Sean because he already has one. So I I think it is needed, moving the data sources, or the checks and balances as close as possible to the left side is is going to be a critical part for, making our operations more agile. And, like, to some extent, some of the AI capabilities, agentic capabilities can help in identification that Sean in this data source and that's the data source I'll say. But it goes complex very fast because Sean, who has addressed in Wyoming, or says Sean who has addressed in Portugal, I don't think any AI will be able to connect that easily.

Sean Weisbrot: And there's not many people with my name, so I think my case is is unique.

Saurabh Gupta: But if think about it. If you in a place where your name is Sean w, And in another place, it is, Sean Weisbrot. How will you connect those? And there are a lot I I I'm sure there are plenty of Sean's.

Sean Weisbrot: Yeah. I I think it depends on the context. And then if you go to Asian countries, they insist on using my middle name, where in Western countries, they don't really care about it. It's not important.

Saurabh Gupta: Yeah.

Sean Weisbrot: So, like, when I do business in Singapore, I have to sign or in Vietnam and China, I had to sign my middle name on all documents or it wasn't considered complete. But if I sign my middle name on a Western on a document in the West, nobody cares.

Saurabh Gupta: Wow. Okay. Yeah. Yeah. Yeah.

Sean Weisbrot: But I've had issues where if I get a flight and I don't have my middle name on the flight booking, and then they see they go to check me in and they see my middle name on the passport, they'll say I'm not the same person. Of

Saurabh Gupta: Yeah. I mean, I think these challenges will continue, Sean. Like, and,

Sean Weisbrot: course.

Saurabh Gupta: they are just gonna get more complicated. Think about it like, temperature data coming from two different sources. One is in Fahrenheit, one is in century.

Sean Weisbrot: The one in the one in Celsius is obviously the correct one,

Saurabh Gupta: I know.

Sean Weisbrot: even though Americans would tell you that's not true. Because

Saurabh Gupta: Slow.

Sean Weisbrot: how can water freeze at, what, 32 degrees?

Saurabh Gupta: Move it up.

Sean Weisbrot: It's it should freeze at zero and boil at a 100.

Saurabh Gupta: Yeah. I agree with you.

Sean Weisbrot: Sorry, Americans,

Saurabh Gupta: I got

Sean Weisbrot: but you're wrong.

Saurabh Gupta: it took me a while to get used to the Fahrenheit. I grew up in India, so everything was Celsius.

Sean Weisbrot: Yeah. And the thing that that's funny to me is any American that goes into STEM learns the metric system.

Saurabh Gupta: Mhmm.

Sean Weisbrot: Why not just teach all Americans the metric system?

Saurabh Gupta: Yeah.

Sean Weisbrot: It's not a difficult thing to do. Like, a few million people every year learn the metric system. Like,

Saurabh Gupta: Yeah.

Sean Weisbrot: I don't think it's that complicated to do, folks.

Saurabh Gupta: Something to

Sean Weisbrot: And then people and and then people find it funny that as an American,

Saurabh Gupta: think about.

Sean Weisbrot: I understand the metric system because, of course, I'm living outside The US. So, like, why do you know? And, like, I I even even weirder and completely more off topic. So I was in Lisbon, and a friend of mine was going to Riga. And I was like, oh, where are you headed? She's like, oh, I'm going to Riga. I'm like, why are you going to Latvia? She's like, how do you know Latvia? And I'm like, how do I not know Latvia? She's like, you're American. I go, that doesn't mean I don't know a European country. She's like, yeah. But you're American. I'm like, so? Like, sure. Half of the Americans don't have a passport or probably could show you where The US is on a map, but I'm not one of those. I I don't live there.

Saurabh Gupta: I'd Yeah. Yeah. Yeah. I think that's a I mean, this is gonna change.

Sean Weisbrot: Is it though? I feel like American education is really bad.

Saurabh Gupta: The education is but the influence is also there now. Right? If you think about it, a lot a big part of America is made out of immigrants.

Sean Weisbrot: Okay.

Saurabh Gupta: And,

Sean Weisbrot: So well, that's going down a different path. So you're saying the the immigrants are naturally going to make America smarter?

Saurabh Gupta: yeah. Yeah.

Sean Weisbrot: No. Okay. No. No. Let's no. That's this isn't the right place to have this conversation, but I understand what you're trying to do. Yeah. We're we're gonna cut that, and we're gonna just keep moving forward. I don't wanna get I don't wanna get political.

Saurabh Gupta: I'm not saying smart yeah. Yeah.

Sean Weisbrot: Don't wanna get political. So Adan Adan is my editor. Adan, please cut the part about the political stuff because that gets really heated really fast. Like, I'm I'm on your side, but, like, those people, I don't want those people to know that I exist.

Saurabh Gupta: I know. And I'm also trying to be careful.

Sean Weisbrot: Yeah. Because you wanna work with you wanna work with governments. So you don't wanna piss off the white supremacist American government. So, again, this is being this is being cut. It's still being cut. We're gonna okay. We're cutting, and we're starting from here.

Saurabh Gupta: Yeah.

Sean Weisbrot: Okay. How do you see this industry evolving? You said you feel like it's gonna become more complex, but do you have any specific thesis or something that you're experiencing right now that's telling that's informing you of how it's evolving?

Saurabh Gupta: So I can give you multiple couple of reasons where I think, there's gonna be a huge change in the industry. So I wanna start with the modern data stack. If you go to modern data stack, there are over 3,000 tools that exist for data ingestion, for data quality, orchestration. And, like, there's no limit to it. BIAI, 3,000 plus fields. And you go to any large enterprise, midsize enterprise, they would have had about 10 to 15 of those to put together their data platform. Now not only they have done this, they have invested in people's skills, the compute that goes with it. And a small change that comes in, they have to go and redo multiple places, changes and adjustments. The cost that this comes with also is, like, someone like, they have also many of these enterprises or or several of them have partnership with SIs, whom they are dependent on learning their operations. And if you look into a little bit more detail, what you will realize is these 10 to 15 tools that every enterprise is paying for, actually, they are not using more than 15 to 20% of the tools capability. But these tools are very, very funk highly functional. They have deep cap vertical capability that not every enterprise uses them. So I think the first round of disruption and transformation that is going to happen is enterprises are going to look for platforms and tools and products which can give end to end capability. They don't need to go too deep, but it should be one. And that's going to be a big, big shift because it will reduce the cost of managing, cost of building. And at the end of the day, it'll simplify experimentation. You don't need to touch 15 places to try something. So I think that is one big shift going to happen. AI is just going to force it a lot more because AI needs access to data. And every time there's a new change, new data, new transformation, meaning you don't need to tag these 50. So that's that is coming. I I strongly believe that is gonna happen. The second shift churn, I feel, is if you think for the whole life cycle of any data, there's a source system where the data is getting produced. The engineering effort to bring data to a data lake data warehouse. Then it is taken over by analytical engineers who their job is analyzing, finding insights out of the data, turning into, dashboards, reports, and then the consumer side. The consumer side could be executives, line managers, general regulatory reporting. I strongly believe the middle layer, which is the analytical layer, is going to go up. The bunch of models that we are having, like, whether it is anthropic or OpenAI, they are going to eliminate the interpretation part, which is mostly the analytical engine. So the whole system will, like, be engineering, producing good controlled data, metadata, and context consumed by these models, and the consumers are going to interact with these models. And I think that shift is coming. It is going to also go through some changes in terms of of our pressure test, in terms of token consumption. Suddenly, you'll see a huge bill coming too. But that's a good place to be where the discussion goes away from whether a model is going to be used or not. I think it's given everyone is gonna use it. So in some form or fashion, how do you optimize it? And that change is going to bring a shift in the skills and the jobs where they are going to exist. A lot of jobs should shift towards the left hand side with the engineering side. The analytical side should become thinner and thinner.

Sean Weisbrot: And you think the models themselves are going to become specialized for that purpose? Or

Saurabh Gupta: So models specialize for interpretation. So the way, Sean, I see it is these models are, they are sending a lot of knowledge. And these knowledge these domain knowledges and the industry knowledge used to be typically with analyst. Those are getting abstracted. What the model is going to is going to help is interpret the data from a lens of knowledge they have or the domain understanding there. Whether it is right or not, it's still going to be on a human or who's consuming it. But that the amount of knowledge that each model can has accumulated can eliminate billions of with tons and tons of analysts. And I I feel we should look at models as interpreters rather than someone who's analyzing. Analysis is still going to be on the consumer side.

Sean Weisbrot: Why do you think that is?

Saurabh Gupta: Because it's limited to the knowledge these models have. Right? If someone wants to so let's take a simple example. I have all financial data created in a data product. And I could there's a report which talks about how quarter on quarter performance has been. Now the moment an executive looks into this and he he or she finds something strange, they go to the analyst and say, hey. Can you give me a little more details of this? Panelist will do go dig into the details of the data. Give another one. Then there's another question. There's a back and forth continuously going on. And the part of this whole back and forth is also getting to understand what the question is from the consumer side or the exec side and how it can be sort of responded to the data. Now all this knowledge is already with models. Specialized models exist, which you'll have. Now if I'm seeing a report, which I serve through a model, same quad quarterly performance report, and then I say, can you give me more details of this? I see something strange. The model should be able to engage, interact with the data and get me the answer. And it should be capable enough or sufficient guardrails should be put that it should not start hallucinating and try to answer for something which doesn't exist. It should say, okay. You're asking suddenly you're looking at sales data, and suddenly you're asking for salaries of employees. That doesn't exist going up to your engineering team. But these models are going to get more and more capable just because it is they have a lot of knowledge embedded in them. And the time for a consumer to interact with a model and get an answer is going to significantly shrink. This was that's a place where, like, the time sync was with the analyst, which is going to have to go.

Sean Weisbrot: You mean the people talking to the AI is going they're gonna get their answers faster, or they're gonna use it less often? Or

Saurabh Gupta: So that they'll get the answers faster, and they have to they have to spend time in looking into whether this the answer that we are they have got is right or not. Is it making sense or not? So so think about this, Sean. Like, if I have a tabulated report on quarter on quarter results, I want to go into the next level of details. And if that report doesn't exist, an analyst will take a couple of days to a couple of weeks to build it for. But if the data exists, the model should be able to extract it for me. So imagine that if three weeks can be shrunk into three minutes, it's a huge compression of time.

Sean Weisbrot: Sure. Yeah. I mean, I I do that with Claude right now.

Saurabh Gupta: Exactly.

Sean Weisbrot: Like, I've done 320

Saurabh Gupta: Mhmm.

Sean Weisbrot: interviews, and I've taken the time before I had AI to generate the transcripts. You know?

Saurabh Gupta: Yeah.

Sean Weisbrot: And now Claude just generates the transcript for me when I upload the video into a Google Drive folder for one of the guests. Like, this will happen for your interview.

Saurabh Gupta: Mhmm.

Sean Weisbrot: And from that, I can then upload a YouTube video automatically. It automatically takes the tags and the titles and the description, and it puts all of that in there. It generates a thumbnail for me, puts all that into the YouTube, and then it creates a blog post in the website using the transcript data and the YouTube URL data. It pulls out all in, generates a new page, and it puts the transcript on the the website. And so now Claude has access to every transcript from every interview I've ever done. And so I can ask Claude about all of the data from all of the transcripts. And it's generated ebooks. It's generated reports. It's generated articles. It's generated a lot of things for me based on the sum total knowledge of all the conversations I've had with guests.

Saurabh Gupta: So so your example is a perfect example. You eliminated or really, really compressed a lot of time that a developer, a website designer, a ebook creator would have spent on. You just eliminated that. And

Sean Weisbrot: Yeah.

Saurabh Gupta: not not only that, like, your effort has drastically shifted in place of getting things done, getting these books created, getting transcripts ready to start validating the finished product. I think it's a huge saving of time and shifting of energies where in place of spending time in building, you're just looking into analyzing and validating.

Sean Weisbrot: Yeah. And I used to do all of this stuff manually every every episode, you know, from people don't realize, but, you know, from the thirty minute intro call to the one hour recording, and then I have a person that does the editing. But the pushing of data around from different spreadsheets and different applications and all of this from end to end could be four hours or five hours of work per episode from me, not even including my editor.

Saurabh Gupta: Yeah. Yeah. Yeah. Yeah. Yes.

Sean Weisbrot: And I've been able, with AI's help, to automate 95% of that work into, like, automated workflow scripts that Claude created for me. Plus, Claude has access to my repository and my host, and so it's using my serverless host to create these functions and these scripts in an automated way. And so now I don't manage the pushing of data. I manage the workflows. Did the workflow break or did the workflow work? So for example, when this interview finishes, SquadCast is supposed to transfer the recordings to your Google Drive folder that was created when you you booked the interview.

Saurabh Gupta: Yeah. Yeah.

Sean Weisbrot: And if that doesn't happen, I have to go to cursor and tell Claude to check on why that didn't happen, and then Claude will push it'll trigger the event to happen, and then it cascades into the the rest of the workflow. And so instead of me putting, like, you know, an hour of work from downloading it to my computer and then uploading it into the Google Drive folder and then generating the thing and then, you know, telling my editor, it does all of it for me. So now I just manage workflows instead of managing the pushing of data.

Saurabh Gupta: Yeah. Yeah. Yeah. Yeah. And imagine. Right? I'll I'd actually, I should ask you this question. Are you able to do more than what you were able to do before cloud?

Sean Weisbrot: Yeah. I can do a million things I couldn't do before.

Saurabh Gupta: Yeah. I I think that's where the shift is happening. And, the shift is, get will be complex, with structured data because that requires a lot more domain understanding and understanding of sources of data. But it's it's gonna be very, very highly valuable. I remember, scenarios when we used to project, the growth of forecast growth. And, and they will circular dependencies where if one country is projecting a certain growth, it it impacts another country. So we have to keep doing these iterations. And, it used to take weeks to come to a like, for these models to resolve. Now it can be done much faster. Very, very fast.

Sean Weisbrot: What's the most important thing you've learned working with data so far in your career?

Saurabh Gupta: So the most important thing, Sean, is what I have learned about working with data is you have to be very disciplined. The moment you try to cut corners, it'll come and bite you. And the only reason why I'm saying this is the most important for now is because, the volumes of data is going to continuously increase. The type of data is increasing. Sources are increasing. So one corner that you've got, it'll come in with you in the future.