Maximizing Data Value: The Rising Role of DataOps and Strategic Orchestration | BMC Connect 2024

Hello, I'm Mike Ard and we're back with six Five Media at the BMC Connect conference in what I like to call Viva Las Vegas. We're here with my friend Basil and we're talking about ops and how to manage this at scale. Basil, welcome to show.

Thank you. Uh, happy to be here. Yeah, We've been talking about this data ops subject all conference, and the issue is there's so much data, there's so many different data types.

It's located everywhere you can possibly imagine in formats that people can't handle, right? How do we attempt to pull all this together in some level of scale that can drive an AI outcome? Right?

So as you mentioned that, you know, the amount of data that the companies are, are now dealing with is definitely a challenge, but I think what's really bringing data ops into a sharp focus for companies is that the CEOs and the boards of many companies have realized that the value that can be driven out of data is crucial for their success. So this is no longer an IT initiative, this is a boardroom sponsored initiative and um, but the volume of the data, the amount of data practices, gen AI is bringing a whole new world of practices to the world is creating a challenge at the operational level, right? So you've got lots of investment in this space, lots of sponsorship, uh, for that investment, but the challenges with the amount of technologies that are coming out in this space and the amount of data that those technologies are supposed to be handling is a challenge at an operational level.

And data ops really is sort of an acknowledgement by the industry that we need an operational model to be successful if we're gonna scale. We've been tossing around this phrase of, you know, data's the new oil for as long as I can remember. It seems like what we've lacked is like the pipelines to get the oil to the refinery so we can actually make something useful.

Um, yeah, and, and this is exactly why data pipeline and data pipeline orchestration in particular, because inside the pipeline, if you look at it, it's, it's not one or two platforms. Uh, public cloud in particular offers a myriad of solutions, specifically focused on data extraction data, E-T-L-E-L-T, data warehouses, data lakes. And while there can be a lot of debate about should we be considering ETL versus ELT, should it be a data lake or a data warehouse, there's one thing that there's no debate about.

Everybody agrees that in production everything should run in an automated fashion. What, what the challenge is with so many different technologies from so many different vendors, you really need a single point of control and really a strategic approach to this. Quite, quite frankly, the reason data ops is a topic now versus before is because ops was always something people did at the end of a development life cycle.

Um, where now data ops, in fact, Gartner's recently released a report which was titled Five Ways of Enhancing Data Engineering Practices. And it was all about ops readiness. So, um, ops has really now become strategic, and hence data ops and variations of it are ML ops, LLM ops, um, the whole gen AI boom's gonna put this whole ops thing on steroids.

Yeah. And part of that means that it's gonna be occurring in, in some instances in real time. I, you know, we've been batch oriented processing of data as long as we can remember, we all know how to do that.

It seems like what's happened now is with gen ai, I need responses in sub-second and I gotta get the right data, the right spot to drive that. And that's not easy, right? So there's definitely a lot of focus on how to get, um, data from its raw form to an insight for the business at the speed of light, uh, which often means real time, but at the same time, it's not going to be the be all end all for all use cases.

You're going to need a strategy which can combine event driven, uh, data movement, some data arrives, data, something a sensor, census something, and then does, uh, take action in many cases where lots of data aggregation is required. So if you think about modern ML models, they can only be trained when you have a large set of data to train it with. So if you think about large scale data that needs to be fed down model for training that may not always be in real time.

So if we zoom out and look at it, what I said earlier about data practices, you have bulk extraction, you have time-based extraction, and then you have change data capture, combine it with streaming, and we're seeing new architectures like Kappa emerge, which are now like combining real time and batch into single streams. So, um, that's part of the challenge and this is why ops is into focus, is how do you not only take the, the technologies involved in here, but how do you align the data practices along with it in a cohesive manner? We hear a lot about the phrase retrieval, augmented generation as it relates to AI models, AKA rag.

Is that a kind of a subset of data ops as a whole and it's only one piece of the puzzle? Uh, I think it's definitely a piece of the puzzle because if we think about, um, AI or gen AI and the models, once you've got a well-trained model and it's deployed, it's really a part of a larger data pipeline. You've got data coming from a lot of different sources, that data has to be, uh, vetted, it has to be cleaned.

Sometimes that is transformed. It has to be aggregated and then maybe it's going to be fed to a model that's gonna analyze and give a recommendation, for example. And the recommendation is often not the last step.

What do you do with that recommendation? So let's say you've got an ML model that has predicted, um, customer churn. Well, if you wanna send a promotional offer to your customers, the AI model is not the one that's gonna send that, um, re um, promotion to the customer.

That has to go back to the application layer, perhaps in an ERP or a CRM. So when you zoom out and look at this picture, the model as important as as much focus it has, it's really part of a much larger ecosystem, right? We only got about a minute left, but once that one thing you're seeing customers doing who are successful, that you wish everybody else would do.

So I think the customers that are, are seeing good success with this are, it's a, it's a combination of people, process and technology as always. And from a process perspective is to think about ops in a strategic fashion versus something that only happens when something has gone into, into production. So ops readiness and thinking about orchestration as a strategy, uh, is, is key in, in sort of defining, uh, between success and and failure at this point.

Alright Folks, you heard it here. A wise man once said, failing the plan is planning to foul. It's no different here.

Hey Basil, thanks for being on the show. Thank You. Appreciate it.

Alright, thank you all for watching the latest episode of our adventure here in Las Vegas with, um, six five on the road and our friends at BMC. And please stay tuned for the next set episodes that are coming up right behind this one.