In The Arena by TechArena - Data Observability in the AI Era with Unravel Data's Kunal Argawal

Episode Date: January 18, 2024

TechArena host Allyson Klein chats with Unravel Data CEO Kunal Argawal about how his organization is tapping AI to disrupt the data observability arena.

Welcome to the Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Alison Klein. Now, let's step into the arena. Welcome to the Tech Arena. My name is Alison Klein. Today today I'm joined by Kunal Agrawal, co-founder and CEO of Unravel Data Systems. Welcome to the program, Kunal. How are you doing? I'm doing great, Alison. Thank you so much for having me here.

So Kunal, why don't we just start with an introduction of you and the company and why you founded this company. Yeah, absolutely. So my name is Kunal Agarwal, co-founded and currently CEO of Unravel Data. We started this company for simplifying data operations, really, as we're seeing companies and data engineers, data scientists inside those companies trying to achieve their amazing goals. What we also see is they are caught up in the not so exciting, but necessary evils of making sure that all these plumbings and pipelines and data models actually work properly.

And what we noticed is the teams were actually spending more than half their day firefighting issues rather than being productive with the data stack. So we figured, hey, can we boil down the problems they have and use data science to actually give them answers to get by their day faster? And that's really what. You know, it's so funny. I've been doing a lot of research into AI and data analytics on the tech arena. And one of the things that keeps coming up is this, for lack of a better word, the unsexy part of getting a data pipeline together so that we can actually get insight from that data. I guess the question that I have for you is, why is that so difficult?

And what did you see as the opportunity to do something different? Yeah. So while everybody wants the latest fraud prevention application, the recommendation engine, or even a chat GPT for their businesses, it takes a lot of steps and a lot of work to get all that data actually working for you. So a data pipeline is nothing but the series of steps that's taking raw data in whatever format it is, and then bringing it all together to start making sense of that data, and then running algorithms and models on top of that data to get to the kind of applications that you need. Now, while it sounds simple, the complicated parts happen because there's multiple different systems that you're putting together to, first of all, make these data pipelines work.

What I mean by that is you may have different systems for ingestion of data, ELT, ETL, whatever your company is doing, to then bringing that into some sort of a data lake, doing some transformations on top of that data. You may be cleaning that data up. So those are already three or four steps. And then you're trying to use this data for either creating a brand new model. So you have another team, a data scientist team usually, that's experimenting with these inferences that they're getting from these not work the way they're anticipated to work. They may work well one time and really degrade in performance or reliability another time.

And getting down to why something happened because it touches so many people, so many systems, and so many steps in your data pipeline becomes a head-scratching problem. And that's the part that's very frustrating, very cumbersome, to go and dig down into system logs and machine-level metrics and try to triangulate and triage what happened and what could be the potential root causes of what happened? And then using trial and error techniques to get out of that problem and say, huh, what if I did this a little bit differently?

Would that get me a better result? And that's the frustrating exercise that we usually see data teams undertake. And it's just a complex, ever-growing problem because the number of systems, the number of applications, and the number of stages keep creating extra connection. I was so intrigued by the fact that you are using data analytics to try to solve that data problem. How do you do that? And how did you and your team create a solution that can really help data scientists and those who are operating on the data stack across different industries. When we think about the word observability, what it really means is observing.

So understanding what's happening. And while that is important, what we learned from working with our customers is that's not enough. People also need to understand why something's happening and how they can actually go and fix that particular problem or that particular issue that's at hand. And it's the why something happening part that takes a lot of expertise and a lot of time. And both of them are at a premium in this market right now. There's a spectrum of people that attack

and are working on the data stack. You may have your hardcore data engineers and people who have run in these systems for multiple years on one end. And then people who are business users and analysts who don't know the intricacies of how a Snowflake or a Databricks or a BigQuery platform actually works. And what we usually see is the ratio of experts to everyday users is usually one is to 10 to one is to 50, depending on the organization.

For everyone expert, you've got 10 or 49 other people who are not experts, and you have hundreds of thousands of these applications or data pipelines running on these stacks every day. So a machine has to solve that problem where the software itself should be able to not only tell you what's going on, but guide you to a remediation so that you can get out of that answer, or get out of that problem much more faster. And with that, we envisioned using telemetry data in the way our customers were using, say, customer data or behavioral data in a similar fashion. And we said, if we can get information about your apps, about your data sets,

about the underlying infrastructure and resources that you're using, and then apply similar techniques to help you shortlist the number of problems that you have, try to connect the dots between the cause and the effect, and then lead you down the path where you can get either pinpointed recommendations about what to go and do to solve that particular problem,

or at least narrow down the focus of things where you should be looking at, then that would have a significant impact on your day-to-day, which means that what we typically see people spend about 8 to 10 hours to resolve each issue can be shrunk down to five minutes to 10 minutes. And that's significant. So that just means that you'll be able to do more productivity during your day. And that's great for any business.

But even on a personal level, you just don't have to go through those frustrations. Or you don't have to be a PhD in these data systems to go and crack those issues that you're faced with every day. What is the customer response been as you've talked about this? Yeah, so our customer base are companies that are advanced in their data analytics journey or companies whose business is data. And we firmly believe, first and foremost, that every company is a data company or going to become a data company very soon,

otherwise they're going to be left behind. And now you can almost say that every company is going to become an AI company. When customers start to run meaningful applications that actually impact business outcomes, then performance and reliability become a first-class citizen. That's the only way that they can actually depend on their data applications or their AI models

to actually run their business. So using a technology like Unravel gives them the peace of mind and the reliability that they need to bank on the outcomes of these data models. So that's been the number one reason why companies choose Unravel. But the second part, which goes more on the operational side, is we also allow customers to scale their data operations in a very efficient manner, which goes down the cloud cost and the budgeting side of the equation. What we've seen is too many times customers will over-provision resources

or their users would do mostly unknowingly, you know, things that would cost more money on the environment or would not be able to scale in an appropriate manner. And we've seen customers control or optimize their cloud spend from 20% all the way up to 40, 50% in some cases, which means that they can run more workloads with the same amount of budget for that particular year or be able to get more throughput out of their environment itself because the use cases for data is not stopping. And these customers, once they realize

that they're able to scale this in a much more reliable, confident manner, then actually they open up the floodgates and start to bring more workloads on because now they're starting to see good ROI against their data projects. So those are the two big areas that we see. And both of them are highly measurable.

We do pride ourselves in giving customers back the ROI, you know, multiple X over in a short amount of time. When you look at 2024 and you think about 2023 being such a big year for generative AI with chat GPT and other solutions coming on, what do you think are going to be the key topics of 2024 as more companies try to implement some of these models? And can companies continue to try to take advantage of their data? Where do you think the industry needs to focus? And how are you part of that solution?

It's a great question. Year one or year two, I would say, of the AI boom, if you may, led to a lot of experimentation. So the next year, you know, while this is moving super fast, my prediction is we will see some of these data-leading industries and companies actually launch more applications in production, number one, meaning they would have meaningful outcomes coming out of these AI applications and endeavors. And actually the market demands that from them.

It's no more sufficient enough to say that you're doing something with AI or you have a partnership with one of the big cloud providers, especially Microsoft, if you may, you actually have to deliver meaningful, tangible, measurable outcomes from AI. And we're going to see many more of those proven, running in production, things that actually move the news for the business kind of production level models being rolled out. And number two, which is actually, you know, the other side of that coin is AI would bring about a lot of expense to the customers and today what we're

seeing is just industry wide people having an open checkbook around any initiatives around AI. But next year, we're going to see this consciousness of ROI or return on investment on AI pickups team, where business leaders are going to be asking their teams about what kind of returns have we gotten on any models that we've built out to any AI infrastructure that we've put out to the AI teams that we've invested in, what kind of tangible business returns have we seen? And that's going to be important to then power the next wave of AI projects

that that particular team and industry undertakes. So I think they both go hand in hand where you'll be able to prove out some of the outcomes that you've gotten from your early AI projects, but then being able to track down to the dollar and cents in some cases of what is the running cost of running these AI models? Is there ways I can make this more efficient? Are there ways in which we can drive higher margins

when we're able to sell these AI driven outcomes and products in the market, which will then help us obviously determine pricing, but also how do we scale it out from this lavish kind of environment or this experimental customer segmentation to then taking it mainstream and globally as well. So it's going to go more from being this fad thing to being a very real, tangible thing next year. When you think about the opportunity, are there any particular verticals that you think are early adoption candidates for you? And have you seen any trends in the market where certain industries are moving faster in this space? Oh, absolutely. So the way we look at it is we usually see

the super large organizations just by the nature of the sheer size, having a lot of data in the first place. And so we're talking about the global 2000s and a majority of that, not all of them, have eventually started going down the path of advancing their data analytics and moving into machine learning, if not AI. And then we see companies that are not the global 2000, but their business is data. And these could be the gaming companies. These could be the music streaming companies of the world, the online ad companies, companies whose major asset is data and their business

revolves around that. Industries would be pharmaceutical, healthcare. Those industries have seen the benefits from the COVID days and even prior to the COVID days around how bringing data analytics can help them leapfrog innovation and really bring products to market much faster. And then when we see the banking and the financial industries, we've always seen them adopt technologies faster than a lot of other industries. And it's happening again with AI, whether that is trading,

whether that is commercial banking, whether that's wealth management, any and all parts of the financial industry is definitely putting data analytics to good use and creating meaningful products. What excites me the most is how companies are creating tangible outcomes from AI itself. And it's not just, when you think about how do you go and deploy and where would you go and deploy AI inside your company?

Everybody thinks about the innovation part, which is what new products and what new cool things we can do with AI. But we're starting to see companies take a more pragmatic approach and saying,

hey, can we deploy AI to help us improve our operations? If you look at our company ops and all the functions that go on inside it, are there areas in there which have room for significant improvement, maybe from delivering higher customer value, creating products faster,

or even driving our costs lower? And those itself can have significant impact on how the company is actually growing and innovating. So I'm excited to see a lot of operational inefficiencies being answered through AI. And almost every industry started to do that now. And while the innovation side is important, it does take more cycles to go and get done and then put in production and actually run reliable models around it. But we're starting to see the industries I mentioned around healthcare, finance, and just internet or lots of consumer-facing companies like gaming actually use that more than anybody else. I can't wait to see how this rolls out. And I think it's

really true about the operational aspects about the companies, right? Even running my own company, I see everything being re-evaluated by how are we going to implement AI to make this more efficient or more effective for our clients. And it's such an interesting thing to think about what's going on inside of organizations today. I'm sure, Kunal, that you're going to be a big part of the solutions for many of those companies. As we head into 2024, what do you think is the thing that we should be looking at from you and your team in terms of what comes next? Yeah, Alison, we feel privileged and honored every day to play a small part in these companies' ambitious goals towards AI.

And our place with any of these companies are twofold. Number one, making sure that your AI models, your machine learning algorithms, your data teams are actually efficiently being used and are productive. And you're not wasting any of those resources out there. And honestly, that's a big enough space for us to capture. Our promise to this market is that it doesn't matter what environment, which cloud, which technology they'll be using. We will support those technologies because we understand that each of these technologies have their own

positives or uniqueness about why they should be used in a particular environment. One of the things that we continue doing is providing support for newer stacks and technologies as they keep coming up. We support the likes of Databricks, Snowflake, BigQuery, but then a long list of other data technologies, and we'll continue to do so.

And the second area is we will keep improving our own AI to help you get to answers faster. So if an application's got a problem, if the manager of the data architecture needs to figure out how do you run 10% more workloads on the similar environment, if the head of product is making sure that the pipelines meet their SLAs, all of those would now be almost like a chat GPT interface where you could simply ask it questions and it would help you remediate those problems in the fastest way possible. In fact, one of our goals as always is,

which is kind of counterintuitive, is if people spend less time in a product, they're more successful. If they want to do a product and they're able to get their answers about why something happened and how can you fix that problem in a second, that's better than spending 10 hours inside our product. So we try to automate every answer as closely to remediation as possible, but at least give them a guided remedy for it. That's fantastic. I really enjoyed this conversation and I learned a lot about Unravel and what's driving the success of Unravel. I can't wait to hear more about what goes on in 2024 with the company. Kunal, we'd love to have you back on the program at some point when

you've got new news. So one final question for you, for those who have not heard about Unravel, where would you send them for more information and engage your team? Yeah, so for more information, you can always check out our website at unraveldata.com. Plenty of information there about how customers are using the product. And I think those speak volumes. So you would be able to see your type of teams, whether on the data engineering, data operations, data science, or data architecture side, to truly understand what the product could do for you. And thank you again, Allison, for the opportunity to speak with your audience here. Yeah. Looking forward to an amazing 2024.

Thanks so much for being here. Thanks for joining The Tech Arena. Subscribe and engage at our website, thetecharena.net. All content is copyright by The Tech Arena. Subscribe and engage at our website, thetecharena.net. All content is copyright by The Tech Arena.