AI Readiness and Data Management Challenges with Riverbed’s Dave Donatelli

Hello and welcome to the latest edition of the Techstrong AI video series. I'm your host, Mike Vizard. Today we're with Dave Donatelli, CEO of Riverbed, and we're talking about a new report that they put together.

I kinda, well, organizations are not nearly as ready for Gen AI as many would expect or hope, and there's lots of challenges and issues. Dave, welcome to show Mike, great to see you again. How are you doing?

I'm good. Give us the high points of this survey that you guys did in terms of what did you see? I mean, if I think back into, uh, last year we had this kind of sense of, uh, irrational exuberance about all things gen ai and maybe, uh, you know, something happened on the way to the forum, as they say.

Yeah, I think what we're seeing here is a very typical technology adoption, adoption cycle, right? Everybody gets all excited, everybody wants to progress. And then kind of the reality of sets in a, you know, this is more challenging to do maybe than we thought.

And that's kinda what the survey reflects. What we still see is, you know, incredibly high c-suite attention to this. I mean, I always mention, you know, most CEOs are smart.

They saw what happened when the internet really became commercialized. And, you know, some companies really advanced their business. Some companies kind of stayed neutral with it, and some companies went out of business because of it.

And every board and CEO uh, meeting, they still say, don't let, you know, get caught short with ai. So 94% of 'em in the survey say it's a top priority while only 37% really say that they're fully prepared to implement it. Mm-Hmm.

So what are the issues that they're running into in terms of implementing it? Is there, is it the data that's not prepared, the infrastructure or is it a combination of all kinds of things and what, what's at the top of that list? Yeah, I think, you know, as you can imagine, there's skillset issues since it's still, you know, a fairly new technology in the sense of generative, although, you know, machine learning's been around for some time.

So AI is all in, you know, which part of AI are we talking about implementing? But data, you know, the world always seems to come back to data. It's been that way my entire career.

And what you see people say in the survey is only four in 10 companies believe that their data is in such a state that they can actually use it to effectively for ai. And as you can imagine, that falls on a whole host of different issues. As you're aware in big organizations, uh, it can be legal, you know, worry about sharing data across borders, even within an individual company.

Um, it can be political. One department likes doesn't like the other department and doesn't trust them with their data or a different division. And then just really the technical issues, it's, you know, it is still very challenging and has always been very challenging to get significant amounts of data in a usable state on a consistent basis for most organizations around the world.

And AI has not yet solved that problem. I think there's this thing called, uh, data and physics and the laws of physics that don't seem to get suspended no matter how much we wish. But, um, how much of this goes back to we we're, we're moving towards, we need to bring the compute back to where the data already exists.

And sometimes it's on premise and sometimes it's in the cloud, but we need to understand exactly what resources are required where, and then organize that data in a way that makes it consumable by these LLMs. Is that kind of like the mission and have we maybe come a little full circle on our understanding of data management? I Think we've completely come full circle and, you know, this might sound, you know, a a little, um, you know, in our own self-interest, but it's, it's turned out to be true.

What we've seen is our acceleration business has seen its best growth in years. And this is the idea of just what you said is people need to move data very quickly and there is just huge amounts of data associated with AI that is in the wrong place and they need to get it to the right place in order to be successful there. The other thing that we've seen is in, in our implementation of ai, which is principally around the observability space.

So either network performance management or end user observability. Um, we've built a data store that I'm, I'm very proud of. 'cause what the data store allows us to do is scale data to great heights, you know, very, very significant scale.

So we can deal with lots of different data and we only deal with actual customer data, so not synthetic data. And we believe by solving that problem with real customer data and being able to scale very high allows us to have the most accurate AI out there for that segment of the market that we play in. But you're absolutely right.

There's gonna be a lot of work going in in change of architecture to make sure people can get the data at the exact place they need it in order to have effective ai. People are sometimes sensitive with what other folks are doing with their data. So, um, when you say customer data, how do you do that in a way that, uh, puts the guardrails in place?

That's a great question, and it's, um, one of the biggest challenges out there. Just as you said, organizations of all types do not wanna mix their data with open source data. And again, typically, you gotta remember we we're playing with large enterprises, you know, security is utmost concern for all host of, host of different reasons.

And so there's lots of restrictions on mixing with open source for, as you can imagine, a lot of restrictions with mixing with outside data. In our particular system, we only use in internal data, so only the customer's data, and it's not end user data or financial data, it's observability data. You know, how fast is the network moving, what's transpiring is something broken here, not broken here?

So it's really metadata about how their infrastructure works and applications work and cloud works versus, you know, looking at real live, you know, user data, their financials, things like that. So we think that makes it very safe for customers. They have complete control over what data is getting used in their own ai.

So again, complete control, end to end real data only doesn't, interm doesn't intermingle with anybody else's data. And in my mind, that kind of brings us back full circle because I think part of the issue is if I look at these LLMs that are out there, they are huge and they're probabilistic and, but I'm typically trying to insert them into a business process that needs to be deterministic. And I wonder if the next wave of gen AI is gonna be more compelling, where maybe the LMS are smaller, but they're trained on a narrow set of data, they're domain specific, and then we have agents that are trained for various tasks.

So, uh, are we a little ahead of ourselves with, in terms of where the tech is to actually operationalize ai? I, I think you're, you're completely accurate there. And I've said this really now for more than a year when I've been asked about this, is that if you think about it, it's, and again, in the, in the enterprise space, you know, com uh, consumer's completely different.

So consumer, large language model, generative, they're gonna do all kinds of things 'cause there's just not as many barriers around security and, and other things that, that, you know, large organizations have to deal with. But in the larger organizations, the real idea that there's gonna be one Uber data pool that handles all issues across the organization, I think is completely unrealistic tech technically, as you said, it's very difficult to do that. And again, there's a lot of other challenges around who gets to see whose data in order to do that as well.

So I think most practically what you're gonna see is more focused, more, you know, focusing on specific issues with a less of a data need, although still large, but less because it's just focused on that particular issue and more of a practical way for people to start to see a return on ai. And then after that, once we have all these different sizes of LLMs, you could argue they're gonna come in t-shirt sizes, small, medium and large. Um, and then there'll be agents that are trained for dairy things.

But we have to figure out how to orchestrate all that into something that feels like a workflow. So is that also part of the next challenge? Yeah, I mean, we see that in our products today, a a again, in our space, which is largely the observability space, uh, we have built what, what we call automations today, and then we're up to over 200, we just released it in May.

So they're going very, very quickly. And what we're seeing people successfully do, to your point earlier, Mike, is if they focus on solving specific organizational or business problems, they can automate that through AI and start to really get to full autonomy where you can have no humans involved in solving technical problems, which we've seen, we've seen customers eliminate literally thousands of problems per month at very large scale, um, by using these types of automations. But I think the key to that is really the practical approach of focusing in on things like help desk or service desk and what are the problems that desk sees and what's the data pool you need to help them resolve those problems through automation.

And that, I believe is really the state of the market today. And over time, as you said, some of these other larger, more, you know, enterprise wide type solutions, um, will become viable. But I don't think they're gonna be viable in the near term, Right?

As one wag one's put it, it's one thing to be wrong, it's another thing to be wrong at scale. Do we need some way to observe all these AI interactions and models and agents so that we know that the thing that's being automated is actually the thing that we hoped is automated? In, in our survey we did, you know, if you harken back to that survey, 84% of the customers said that's a requirement that, that they have to be able to instrument and understand what's really happening.

As you can imagine, there's always kind of the fear of, you know, when you truly automate something, people still need to understand what's happening in during that automation, you know, where, where it's the real state of their technology at any given time. And that's why observability is so important in this state. And that becomes more of a challenge all the time because of the fact that, again, we're now gonna integrate more data models, whether it's data from cloud or data from work from home, or data from your data center, um, SaaS based applications, homegrown applications.

So when you're, you know, the person responsible that if something goes wrong and you have to be able to explain what went wrong, where is it and how to fix it, um, having that precision and truly understanding, you know, what has happened, where it's happened and what the current state is, you could see, you can imagine why that's such a big requirement and, and overwhelmingly the customers say it is. And that's, you know, where we're very focused on solving that Problem for our customers. Do you think that as regulations evolve and maybe the people writing those regulations don't necessarily understand what it is that they're asking for, but ultimately won't those regulations require a level of observability that maybe we don't even have today, but as we go forward, there's gonna be an absolute, uh, mandate?

Um, well, very much so. I think, you know, as you said, I think it's been challenging to regulate tech 'cause the tech's moving very quickly. I think the customers actually do a pretty good job through their own requirements, meaning they won't put a product in production unless it meets, you know, their specific needs around observability around, um, you know, traceability of what's taking place so they can, you know, again, be able to audit it and report on that.

Mm-Hmm. So I think customers are gonna get there first with their own requirements or they won't implement the products. And you already see this today around, um, you know, a lot of large language models or, you know, you have to have committees within a company to even approve use.

And in fact, what we saw in our survey is, you know, over 57% of companies have an AI committee and that committee is doing their own internal, not only what, you know, what should our AI architecture be, but what are the, what are their technical requirements are gonna be for implementation and permission for folks within the company to actually implement. So we think that's a big deal. We also see almost an equal amount of people have have formed enterprise wide observability committees for the same idea, putting standards in place around observability so they can understand exactly what's happening in their enterprise.

What do you think the impact of all this is gonna be on our traditional networks? I scratch my head sometimes and I go, there's a lot of data, maybe orders of magnitude more data than ever. And we can cache some of it and we can process it, some of it locally, but it just seems like there's gonna be a lot more data than ever flying around networks that may not be able to handle that kind of bandwidth requirement.

Yeah, I mean the, the way I, I don't almost call it cat and mouse and, and as I mentioned, it's been a cat and mouse thing that's since technology has started, which is sometimes the networks get slightly ahead of the data and everybody thinks they're in great shape and then all of a sudden there's a new use of data that explodes data again. And, you know, they get back into what I call chase mode, right? How do I, how do I make my network more effective and efficient in order to deal with these tidal waves of data Very clearly With ai, we are back into chase mode.

As I mentioned in our case, you know, we are, we're in the, you know, the acceleration business. We see whole new customer demand happening because of this. And, and as you mentioned, people are moving data to different places than maybe they have had it traditionally.

And that's also really, you really are, it's forcing customers to rethink their architectures. And again, if you think about all these different architectures, right? Data collected in cloud SaaS, traditional at home, um, is very challenging right now.

And it's only gonna get more challenging again as AI continues to grow. So given all the issues we've discussed, is there a disconnect between what the CEOs think is possible with Gen ai? Because, you know, they see what the Microsoft, Googles and Amazons of the world are talking about, and the reality of what the IT team, the data scientists and the data engineers and the developers and the rest of the village it takes to put the AI together are seeing and the timetables are misaligned because the CEO thinks this is happening tomorrow, and the rest of the crew says, you know, we're 12 months or more away.

Yeah, it's, it's, it's exactly what you say is, you know, as I mentioned just slightly more than a third think they're fully prepared for ai. Now, if you talk to the IT departments, 72% say it's been really challenging to implement. Um, and their belief is it's about three years out.

So 86% of the people, and it's kind of interesting 'cause three years out is far enough out, you really don't have to do anything this second. But 80, 86% of the respondents say in three years we're gonna be fined. But, you know, until then it's, it's gonna take time.

And, and principally again, the, the challenges are the ones we've spoke spoken about, which is really data, getting the right data available in order to really implement ai. So definitely the c-suites ahead. Another interesting thing that came out in the survey though, was people were very high on their own capabilities.

So, um, over 82% said they were ahead of their competition. And as you can know, you can only have 51% ahead of the other half. So people were really, really, uh, up on that and only 5% thought they were either even or lagging.

So a lot of positive feelings, but I, again, I think we're still in early stages here to really see the results of it all. I'll tell you in my own experience, 'cause it goes both ways. Sometimes I feel like it folks are too pessimistic about change and the rate at which things will happen.

And yet on the other end of the thing, I can find just as many but who are probably overly optimistic about their capabilities. So if, if the bulker saying that they are three years out, what's your bet? Are you, or, or short, are you gonna short that or are you gonna bet long on that?

I I, I would short it on, on the practical side, I, I think, you know, if you, if you also look at it, most believe that the first use case is gonna be operational efficiency, IE cost savings. And there's always, you know, take it from a CEO, there's always pressure to get more efficient and reduce costs in any organization and whether they're public or private. And once people start to see that you can really save money, and we're seeing that with our customers today.

They are, they are, you know, I was talking to a customer the other day, they've reduced their amount of, um, help tickets, uh, 15% of their head into 20%. And that's real, you know, in a large organization that's significant amount of savings. So once people start to see that they're gonna be able to save money like that, there's gonna be a lot of enthusiasm to get that done as quickly as possible.

So I think it's, I I'm shorting the three years, but I'm shorting the three years under the guise of what we spoke about before, practical solutions that focus on a segment of your business, you know, the, the true what I call intergalactic, everything connected together, everything, ai, that, that's that. I'm, I'm, you know, I'm betting on the long end of that, plus that's gonna take a lot longer. All right, folks, I don't know if there's an AI sports book in Las Vegas, but maybe there should be based on this conversation.

We'll see where it goes. But hey Dave, thanks for being on the show, Mike, great to see you again. And thank you all for watching the latest episode of the Techstrong AI video series.

You can find this episode and others on our website. We invite you to check them all out. Until then, we'll see you next time.