Leadership Insights: Real-Time Analytics

Transcript

Mike Vizard: Hey folks, welcome to the latest Digital CxO Leadership Insights series video. I’m your host, Mike Vizard. Today we are with Shruti Bhat who is the chief product officer for Rockset. We’re talking about the transition to analytics in real-time. And that’s become more pressing because of the current economy. So we’ve tried to get some insights as to whether or not folks will make that transition; how far along they are. And my question to Shruti is, what do you see? I mean, our business executives are saying we need to have, you know, up to the second analytics – or what is the definition of real time? Because it all could be in the eye of the beholder, right?

Shruti Bhat: Yeah, that’s a great question. And thank you so much for having me here. So, rea- time analytics, really is about two things. One is real-time data. So how fresh is your data, how real-time is your data, are your decisions being made on the latest data? That applies to two things. One is operational analytics and user based analytics. I’ll come back to that in a second. But really, the one you want to remember is that it’s not just about your data, it’s also about your queries. And this is what I really want to call out, because this is what we’re seeing in the new economy, which is, it’s no longer enough. If your data is coming in real time, you also need queries that are sub seconds and fast. Why is that important? Because suddenly, you are building user facing analytics, you’re building apps that your consumers are using, even for businesses you’re building, and SaaS applications with embedded analytics inside. And for that, your queries have to be really fast. So what are people asking for in this economy? They’re asking for low latency analytics, they’re asking for really fast sub second queries. And they’re asking for that to be very compute efficient. This is the new thing. Going into the next six months, we’re seeing more and more that it’s not just about, oh, as a data team, look at me, I got a great performance. Not good enough; you have to say, as your data team, I got you great performance and I cut your cost. And that is how you make progress in this economy.

Mike Vizard: In some ways, it’s often an awkward conversation for the IT folks who have to go explain to the business leaders as well, and say the analytics and the data warehouses they’ve had all these years weren’t really in real-time. And so maybe some of the decisions are suboptimal as a result, and the business leader’s gonna look at them with one eye closed and go, “Come back again?” Exactly. But the point, though, is from a business person’s perspective, you’re kind of like, well, what’s so hard about this? What is the challenge in making that happen and getting stuff in real time? Because it might give consumer experience and real-time all the time? So what is the issue that business IT folks need to wrestle with here to get to this new era of real-time analytics?

Shruti Bhat: Yeah, that’s such a great question. And I just want to say, it’s not that all this time they’ve been making the wrong decisions. It’s just that they’ve been making decisions and operating in a very different fashion. So if you think about businesses, a few years ago, it was executives looking back at the quarter, and it was, you know, getting insights on your business from the last quarter, from the last year, quarter over quarter year over year – what’s been happening. But now, what’s really changed is the operating model has changed. So this is why it’s called operational analytics, or it’s called even user facing analytics, where people on the ground have access to data to make decisions every day. Think of logistics tracking, for example – its supply chain efficiencies. But this is people on the ground tracking in real-time. Where’s my truck right now? How do I read out what can I do? So it’s not that they’re making wrong decisions. It’s just that now they have the power to empower all the operational folks and people on the ground to make better decisions every day. In the past, that was just not possible. Why was it not possible when, like you said, Facebook and Uber and all these people have been giving us real time for a few years now? It was not possible because all these big companies did it with 1000s and 1000s of data engineers – you’d talk to Uber and they’d say I have this army of data engineers and data scientists just building it out for me, because that whole digital stream is like the lifeblood of my business. And most traditional businesses could not operate that way. They could not hire a massive data team. So then, what did they try to do? They tried to do it in their warehouse, that same trusty warehouse, which was great for reporting. It’s perfect for reports, isn’t it? However, When you tried to do Operational Analytics, you tried to do real time analytics, you tried to do user facing analytics on your warehouse – what happens is your cost goes through the roof. Suddenly, you’re burning compute credits like nobody’s business. And why does this happen? Two things happen. One is, if you think about a warehouse – it was built for the era of batch; it was built in an era where executives did weekly, monthly quarterly reports. So it was ideal to load your data in batches, maybe every week, maybe every night. That’s the world that the warehouse is built for. You start updating your warehouse frequently, and suddenly, your costs go through the roof, because it does this very expensive merge operation and every insert- every update. And that is one problem. The other big problem is, again, the queries, it always comes back to your data access patterns. Certainly, with operational analytics and user facing analytics, you need to do a lot of queries. So it’s very frequent queries, and different types of queries. You know, you’re no longer asking, “Tell me the average sales price quarter over quarter?” No, you’re saying, “Tell me everything you know about Mike. What has Mike purchased? What is Mike clicking on?” And those are very selective queries. So the access pattern is different. And you try to do it in a warehouse, which does brute force scanning – of course, your compute cost goes through the roof. So you asked me what has been holding people back – this cost problem is what’s been holding people back – that suddenly you have this new era of data apps, a new era of data products, and the cost equation on the old tools of the previous era, which is, you know, warehouses, the price performance, the compute efficiency, the cost model, this just doesn’t add up.

Mike Vizard: We hear a lot of organizations are struggling with digital transformation initiatives. And I wonder if a lot of that has to do with just the simple fact that they didn’t think through the analytics part of this equation, and how they were going to understand the data in real-time to create some sort of experience in real time. And maybe they put the cart before the horse, and they kind of built all these processes without thinking about what the back-end capabilities required.

Shruti Bhat: That is definitely something I’ve seen firsthand. In fact, you know, a FinTech company that we’re working with – one of the largest buy now and pay later companies, they’re processing millions of transactions across thousands of merchants, and how do they catch anomalies? How do they catch fraud? How do they catch when Apple Pay stopped working in West Africa? I mean, these are all very digitally native kind of problems. And trying to do it in a warehouse, they found that they were always six hours behind. So by the time they caught the anomaly and the time they could take action, it was six hours. And just the cost of running that operation, even bringing it down to six hours was very, very expensive. So you’re right, putting the cart before the horse is basically, well, the right use case, the right problem solving, but without the right tools. And that’s why not trying to do it in a warehouse is what was making it so expensive for them. In fact, now with Rockset, we’ve been able to bring it down to one to two seconds. And we cut their cost almost in half. Because when you move from the batch era to the real-time era, with the right tools, guess what, you can actually reduce costs and increase performance at the same time.

Mike Vizard: There’s also a lot of distrust in the data and business people look at sometimes the reports that it sends over and they just shake their head and go, “Well, that doesn’t reflect any reality that I know.” And some of that just has to do with the nature of batch system. So do you think as we move along towards real time analytics, that the trust gap will narrow because we’ll be dealing with data in the moment?

Shruti Bhat: So a great question. Making the trust gap narrower is more about observability, data quality, making sure that people can see what’s happening to the end. So yes, we do invest a lot in that. However, whenever you’re dealing with data, if you don’t think about the end-to-end pipeline, that risk is always there. The other thing that we’ve seen is the way to reduce that risk is something that developers have been doing for a really long time, right? How will developers build software with DevOps and all the processes that they have? For us, we don’t serve business executives and analysts as much. So this is one big distinction, Mike, with real time analytics, it’s no longer a dashboard, right? Think about this. Do you really want to have a human staring at a real time dashboard? No. What you want is to have real time data. You want it to trigger alerts, you want it to have a program that’s monitoring what’s happening, tapping you on the shoulder and saying, “Hey, Mike, you need to take a look at this right now.” So that’s where the trust increases. Because as it happens in the moment, it’s coming and tapping you on the shoulder and saying, “Hey, I’m noticing some anomaly, you should go look into this right now.” And it lets you catch things in the moment. So that’s how the trust increases, by having programs that are monitoring it and having programs that are saying (I’ll give you an example) if the temperature has gone up in the middle of the day at noon, that’s normal, because we didn’t see it every single day, but if it’s happening in the middle of the night, something is off here Mike, so you might want to take a look. So that’s how the trust goes up. Because it’s constantly monitoring on your behalf and tapping you on the shoulder.

Mike Vizard: How smart will these systems get? Of course, we hear about AI all the time, but I guess, I just want to walk into my office one day and maybe speak out loud, “What are the three things that are likely to get me fired today?” And then have the report come up and kind of walk me through that and share some suggestions to avoid that unpleasant outcome.

Shruti Bhat: That’s so funny. Yeah, I hope we get there soon enough, I can’t wait. But the reality is, if you think about the three phases of what’s happening in the real-time and machine learning world, the first phase was just people investing in collecting real time data. So you saw a lot of this where people were investing in all the infrastructure to collect data in real time. The second phase, which we’re in right now, is a lot of people just figuring out how to get value out of that real-time data. So I know all this information, but can I personalize my experience? So it may not tell you how not to get fired, but when you walk into your office, it can adjust your temperature for you. Because it kind of now knows you normally walk in at this time of the day and you like this temperature. And as soon as you walk in, it can go adjust your temperature for you. So it’s gotten that far. But it can personalize things. You go online to shop for something, and it can personalize your shopping experience for you, based on who you are, what you purchased in the past and what you’re browsing on right now. And we almost take that for granted, don’t we? So that is the current phase we are in; it can tell you how to track. Uber’s a great example. But it can tell you how to track your groceries, if you’re shopping for groceries online – it can tell you where exactly it is and when it’s getting delivered. But the last phase is something that we’re just entering into, which is a lot more intelligent applications, as you call them, more machine learning, more real-time ML features serving – where these applications are learning constantly taking real-time signals, and then taking actions on your behalf. So if you’ve done (just to take it back to your example), if you’ve gotten fired enough times, it probably learned why got you fired and tells you what not to do.

Mike Vizard: What do you wish that people knew, especially C-level executives going into this conversation upfront, versus the things that they kind of discover six months later, a year later, that would have made the whole process a lot faster in terms of the transition?

Shruti Bhat: That’s a great question. The thing that I find C-levels are struggling with, especially as we go into this economy, is in the past, it was very CTO driven, right? You could invest in a lot of innovation projects, you could say I’m going to make this big bet. We’re going to do an experiment, we’re going to do some growth hacking. Today, CFOs are calling more of the shots because budgets are getting caught. Everybody has to be more prudent. And suddenly, you have to continue to show growth and continue to do the innovation. But you have to pay attention to the budget. So the thing that I hope that all the C-levels pay more attention to is how do you get what you need out of your data? How do you get the performance you need, the real-time signals you need? But also pay attention to cutting cost at the same time and by cutting costs to actually mean on your data infrastructure? So it’s perfectly reasonable that today we’re seeing a lot more teams coming to us saying, I’m spending a million dollars on my warehouse today. And I’ve been asked to cut – my CFO just asked me to cut that budget in half. So I have to somehow cut my data stack budget in half. But my CTO just asked me to double the performance. Because you know, these are user facing analytics and to keep my revenue going, I need to increase the performance for my customers. So this is the reality for data teams – the CFO is asking you to cut budget and your CTO is asking you to double the performance. How do you do that? So I hope that C-level executives go into this knowing that the only way to achieve this is to use the right tools for the job. If you’re trying to brute force your way with an older stack, you’re just, you know, trying to get more and more and eke out more performance with an older stack. That’s not possible anymore. So go into digital transformation initiatives  – go into these real-time analytics initiatives, knowing that there is a more efficient way, use the right tools for the job, really asked the questions, if you’re building reports then a warehouse is the right thing to do. If you’re building user facing analytics, or operational analytics, go pick a real-time analytics platform, because that’s the one that gives you the right efficiencies. And when you set up your evaluations, really set it up in the sense of price performance. And ask your team’s hard questions about price performance. Because that is how, in this new economy, you can get the best of both worlds and you can cut costs and increase performance.

Mike Vizard: Great. Hey, Shruti, thanks for being on the show and sharing your insights.

Shruti Bhat: Thank you so much, Mike. Really appreciate you having me here.

Mike Vizard: All right. And who knows, maybe get rid of a few tools along the way as you shift to real-time, and we’ll see if that will help pay for the transition. But we’ll see how that goes. Because sometimes you just need more tools. And that’s all there is to it. But hey, folks, thank you for watching this latest episode of the Digital CxO Leadership Insights series. You can find this episode and all our other episodes on the Digital CxO website. We’ll see you all next time.

Shruti Bhat: Thank you so much.