Erik DeGiorgi, Netspeek

The 16:9 PODCAST IS SPONSORED BY SCREENFEED – DIGITAL SIGNAGE CONTENT

The people who build and maintain very large networks of displays, PCs, servers and other devices tend to have more to do than time to do it, and when some technical shit hits the operating fan, trying to work out what's happening and what to do about it takes experience, brainpower and what can be punishing downtime.

So what if generative AI could be used by a network operations center team to comb through knowledge bases and trouble ticket archives to identify solutions in seconds, instead of minutes or hours? And what if a lot of meat and potato workflows done to deliver services and maintain uptimes could be automated, and handled by an AI bot?

That's the premise of Netspeek, a start-up that formally came out of stealth mode this week - with an AI-driven SaaS solution aimed at integrators, solutions providers and enterprise-level companies that use a lot of AV gear. The Boston-based company is focused more at launch on unified communications, because of the scale and need out there. But Netspeek's toolset is also applicable to digital signage, and can bolt on to existing device management solutions.

The guy driving this will be familiar in digital signage hardware circles. Erik DeGiorgi was running the specialty PC firm MediaVue, but sold that company about a year ago. Since then, he's been forehead-deep working with a small dev team on Netspeek. We caught up last week and he gave me the rundown.

Subscribe from wherever you pick up new podcasts.

TRANSCRIPT

Erik, nice to chat once again. You sold your company about a year ago, and I don't want to say disappeared, but kind of went off the grid in terms of digital signage, and now you are launching a new company called Netspeek. What is that? 

Erik DeGiorgi: Thanks for having me back, Dave. It's crazy. Time flies. I think it's well over two years at this point since our last conversation. 

We launched Netspeek at the beginning of the year. At the same time, we sold out MediaVue. Netspeek is bringing to market the first generative AI platform focused on supporting the day-to-day operations of mixed vendor estates of pro AV networks. Digital signage is certainly a component of that. We're really focused on the totality of pro AV technologies. So it includes a lot of UCC unified collaborations and communications technologies as well as signage, and really targeting office spaces. So think about meeting rooms and conference rooms. You might have a Zoom or a Teams environment in there as well as a signage system or classroom environments, and what we've developed is a generative AI solution that can be embedded into those networks, that can work alongside human operators, network administrators, technicians to help them support them in their daily workflows, and then also bring a large amount of automation.

So our platform can not only kind of observe what's going on in a network, kind of a 24/7-365 way, but then take action and use its own logic and reason and independent thinking to analyze situations the same way a human operator would and then structure and generate responses. So being able to directly address equipment and solve problems independently. We're pretty excited to bring that to market. We're launching to the industry here in a week, and then we'll be demoing at ISE at the beginning of February. 

You’ll have your own stand at ISE?

Erik DeGiorgi: Yes, and I did pull up the booth number ahead of the call, but of course now it's on a different tab. It's in the Innovation Park, and the booth number is CS820, and it's actually centrally located there in the Innovation Park. So actually right outside the digital signage area. 

Yeah, I think for people going to ISE, the Innovation Park is kind of along the main corridor in between halls. 

Erik DeGiorgi: Yep, it's the central hallway. 

Okay, so people should be able to find you there. 

Erik DeGiorgi: Hopefully, yep. 

Not a sprawling booth like a Samsung or LG or something, but… 

Erik DeGiorgi: We measure in single meters. I think it's a 2x3 meter booth. 

Startup life.

Erik DeGiorgi: The price was right. 

There are lots of device management platforms out there, either independent third-party platforms that you would subscribe to and bolt onto your system or a fair number of companies, whether they're integrators or CMS software companies in the context of digital signage have their own device management code written in, how is this different? 

Erik DeGiorgi: Yeah, absolutely. Netspeek is not another monitoring platform. Monitoring is a necessary component, right? You need to know what you have on the network and know what it's doing as a foundation. But our value really lies in the intelligence that we're bringing into that. So it's taking that monitoring and observation, but then actually doing something with it in doing that either again to assist a human by bringing kind of an encyclopedic knowledge and institutional knowledge or whether it's through the automation, and so we're going to market with a total solution.

We have a monitoring platform that we've developed as a necessary part of our total solution, but we actually are also partnering with existing remote monitoring and management platforms to essentially bolt on to them, and then bring that intelligence to their monitoring platform and actually at ISE, you'll be able to see that as well. 

So they should happily run in parallel using APIs or…? 

Erik DeGiorgi: Yep. So we hook into the existing monitoring platform and we essentially bolt on the, the reasoning and the intelligence, and then allow an existing user to leverage that front end, and that monitoring platform that they're already familiar with.

Who do you think you're primarily going to be selling this into? Is it like integrators and service providers who have network operation centers or would it be end users? 

Erik DeGiorgi: So it's a little bit of both, and candidly at an early stage, you tend to take a bit more of a scattershot approach, and test where the value emerges. It's a new technology, gen AI, everybody knows it's there and in a large part don't know what to do with it. But we've kind of honed in on three initial go to market opportunities.

One is like a total solution directed towards the end user. One is more of a channel centric focus, whether it's a system integrator managed service provider. We're actually already engaged with a few, of each, that are interested in leveraging the platform in that capacity. And then also, like I said, with, an existing management. You could be a manufacturer. So think about even an independent manufacturer, or a platform provider, like an existing monitoring platform. So an existing tool is specific to a manufacturer or a tool for more broad-based management. Like I said, we can kind of bolt into those and go to market that way as well. 

So in the scenario of a network operation center in the context of digital signage, an integrator that's doing the work to monitor a large QSR network for a restaurant chain that doesn't want to do that internally and they've got a whole bunch of screens up on a wall and they've got big curved desktop screens and the whole bit and they're watching what's going on. 

Is the idea here in part that. As a problem develops and it's kind of weird and not familiar that if you had to go into a whole bunch of manuals and archived information, it would take many minutes, maybe even hours to do it versus if this is all on a learned model that the solution or at least ideas on a resolution could come up in seconds?

Erik DeGiorgi: We really kind of lean into the personification of our platform. So our product is called Lena and Lena is an acronym that stands for Language Enabled Network Administrator. So we really have modeled the platform and the solution after the workflows that human operators perform every day. So imagine being in that knock and sitting there next to your colleague, Lena and Lena happens to be trained on every respective certification related to the deployment, and has been trained in every application software that's being used, has an encyclopedic knowledge of every technical document for every piece of equipment or technology that's in that deployment, has the ability to - at the speed of light - comb through any historical information like previous support tickets or anything like that that's been related. 

So being confronted with a situation, whether that's a critical situation or whether it's looking at something that's preventative, or maintenance-oriented, just imagine having this kind of superhuman user that can just as a human operator analyze the situation, develop a logic flow, think critically about that situation, pull in outside information to help diagnose a potential issue, construct a resolution, and then either autonomously or along with a human companion and approval, go ahead and execute that action.

One of the things that Lena can do out of the box is we've done all the integrations, and I say all we've done many, and we're continuing to do many more integrations with all the different devices and technologies that you see in these networks. So, as a generative AI, Lena can generate information for human consumption, but Lena can also generate structured information that translate down to device commands in various ways. So Lena can actually take action and do things on her own, and, we default to saying “her” because get used to personifying. Some people lean into that, some people don't. But you really kind of think about it as this if you had your next hire, your next employee that had all of this institutional knowledge and had the ability to take action in this way. 

What would be the ROI on something like that? I assume that if there's a problem emerging that seems kind of weird, that can take quite a bit of time theoretically to come to a resolution, unless you have somebody on staff who is almost like Lena and has that encyclopedic knowledge, otherwise it's going to take many minutes, right? 

Erik DeGiorgi: We quantify value in two ways, coming from two different directions. Again, think about the application. We're primarily focused on day one is kind of meeting spaces, conference rooms, classrooms, that type of stuff. So you have people, employees, workers going into those spaces, and your sales and marketing people having meetings every day and using those spaces. How much downtime is there in those rooms and what's the value of eliminating some of that downtime, right? So it's kind of a workforce efficiency quantification and we ran that as an exercise and based on our pricing models and some averages of salaries for typical people and took a stab at it and if we save one person in one room two minutes a day, it pays for itself. So imagine a meeting of four people. If you can shave 30 seconds off of that, it pays for itself. So that's kind of one way to look at it. 

The other way is what you were talking about is kind of the operational side. What is the cost to operate these networks and to develop that skilled labor, to deploy that skilled labor? And even with that skilled labor, there's kind of the human component. It takes time to process information to think through things. Lena operates at a different pace. So being able to not only just problem solve, troubleshoot, and come to a resolution quicker, but our real objective is actually to mitigate most of those problems from occurring to begin with. We're going to be showing a couple of things at ISE. I'll just give you a tangible example. A room check is something that we're going to be demonstrating. So for those unaware, at most corporate spaces, there are people that go around and check the rooms on a daily and weekly basis. It's a high frequency. It's a very manual process. I'm going room to room running a test call, making sure things are working before the start of the day. It's a high labor, high cost process. So we're actually demonstrating an automated room check where Lena being embedded in the network can go and perform that activity autonomously at really any frequency. 

So we're actually going to break the system and we'll be demoing Lena identifying that and actually resolving it. So something's logged out. I'm going to log back into that room's zoom account and run a test call and give it the green light. That's a very tangible thing. That's a real world thing that a lot of people do in these spaces that costs a lot of money and it's something that we're going to be able to help out with. 

Is this the way AI works, is it kind of a continuous learning thing, where the more that Lena is applied to a system, the more it's learning about its quirks and things that happen, or is it kind of a preset load of information and it's just working off of that?

Erik DeGiorgi: It's an interesting question. I would answer that by saying it's kind of a hybrid, because there's a couple things there. So first there's, I think, a data security and data integrity aspect, and that's something we take very seriously. So we were deploying and as we grow and many enterprises as we did in our previous company, of course, we're not going to be extracting any information or learning from that specific application and bringing that into the general knowledge. So none of our users' information related to this specific network gets brought into the general learning, right? So we take a very, data security approach.

I mean, AI is exciting. It's going to change things, but it's also scary, right? It's new, and we want to make sure that we have a very clear focus. I'm in a very clear message around that, that as we deploy this, any state specific or enterprise specific information is retained by that and it's not brought back into the general knowledge. But, in a different way of answering that question, our platform is built to think critically about situations and develop its own logic flows and reasoning flows. So what it is not is pre programmed, in a sense. If this, then that, right? That's not what we've done. We've created a body of knowledge and we've created a set of contexts and parameters, and then we allow Lena to kind of think on her own in order to address and identify and work essentially, again, the way a person would. So there's kind of a big question we could go in a couple different directions, but I don't know if that helps at all. 

Yeah, it does. I'm curious. The thing about AI, particularly in its early ages, and it's come a long way in the last year and a half, but one of the concerns was around hallucinations and AI models just making shit up.

How do you kind of wall that in so that if you have a resource archive that when Lena is trying to troubleshoot something and it comes up with a resolution, how confident can you be that this it's absolutely working off of what's there and not just kind of imagineering some other solution?

Erik DeGiorgi: Yeah, absolutely. So just as a human being would, if I ask you a question, you're going to want to have an answer, right? And if you don't really know the answer, you might kind of fudge it, right? These models are not too dissimilar in that respect. So it's a complicated answer.

There are several, if not many things that can be done. There's a lot of context that's provided, so Lena is never reaching out to the internet and looking for information. Lena's on rails in that sense, there's a very kind of tight context, we have built in mechanisms within the platform to self audit and self check so when Lena is presenting something, you know like, here's an issue. I'm going to construct a potential resolution to that. There's actually mechanisms that we've built into the platform that will take that output and fact check it, if that makes sense, at a high level. 

So you're absolutely right. It's absolutely a problem. But there are several mechanisms that we've put in place in order to mitigate hallucination, and quite seriously, we interact with this platform quite a bit as you can imagine, and it's really not an issue.

So with automation, are there kind of guardrails around that? Because if anybody who's watched Hollywood movies thinks about Skynet and Terminator and everything in between. So if there's a problem, would Lena decide, okay, we're having a problem with overheating, so I'll just turn off all the power in the building.

Erik DeGiorgi: No, I mean, again, Lena has very tight parameters, right? And it's actually not that hard to constrain a model in that way. We're not just letting this thing loose and it's multifold, right? So it's built into Lena's programming to not do that, but beyond that, automation is enabled as much as the human counterpart want to, right? So  are there non mission critical things that we're allowing when it's fully automated? Are there things that we need, human confirmation, it kind of just depends on the workflow. It depends on the application, how much you want to automate and how much you don't to be candid.

You know, we're learning what the temperature of the early adopters is and seeing it. It’s a brave new world for all of us. We're trying to figure out, it's all new, right. And it's not gonna be hell where it kills the crew to protect the mission. So it's far more benevolent, I think. I like to think about it more as the computer on the Starship Enterprise. Let's use, like, a good one, right? It's your friendly AI that's keeping track of all your critical systems and is there to help you work through problems.

I'm guessing among potential customers, there's at least a couple of lines of thinking. One being that Lena would allow us to do more with the staff that we have in our knocks or whatever their operation they're thinking about for it, but the other one would be Lena can take the place of like five staffers and we can save half a million bucks or something 

Erik DeGiorgi: Yeah, it's an interesting question. It's one we actually talk about quite a bit and you know, disruption disrupts in many ways, right? This technology is most certainly disruptive and that's not just in signage,it's going to be in every aspect of our life going forward. But I believe and it's just not my belief, but it's what I hear and observe, is that the teams, the humans that are actually tasked with doing the operational work every day are so overwhelmed, that it's going to be more of the former, right?

At least at the outset, certainly, there's so much we were talking about it the other day. You know, we have an immense amount of data. How much data comes out of these networks use your, example network, you were talking about the retail installation, how much data comes back from that network, whether it's telemetry or whether it’s telemetry coming back at the device level or whether it's information based on viewership. I mean, there's so much data that's coming from these systems. How much of it's actually used? You know, I don't know, if I take a stab at it, under 10%, 5%, I mean, probably very little of it's actually used. It's overload. So why don't you just put it on and now you have your data analyst and you can actually leverage that information. So there's a lot of work that's not being done in the spaces that Lena can step in and do that can support and work alongside human operators that are presently overtasked.

For those people who are listening and thinking well, this sounds interesting. I wonder if this can apply to my business and my operation they would then be wondering How do I do this? Am I buying an enterprise license? Am I having to install a local server? Is it a SaaS model? How does all this work?

Erik DeGiorgi: Yeah, so we are a pure SaaS model, and again there's a couple of different go to markets, whether it's your channel, but let's just say like the direct to end user. If I'm an enterprise and I want to adopt the platform, it's a SaaS model, it's largely a cloud based, all the magic goes on and in our cloud infrastructure. 

There's an edge client that gets deployed into each of the local networks in the enterprise, and between the cloud infrastructure, between the edge client, which is typically virtualized in the network, we securely and safely can communicate with all of the connected devices and operate the network as we've been talking about.

What kind of a learning curve is involved? I mean, I assume this is not the sort of thing that you just sign up for and get your activation code, plug it in and off you go to the races, like there would have to be quite a bit of onboarding, I would imagine. 

Erik DeGiorgi: Right. Well, so there is onboarding. I would say, in comparison to maybe some of the other universal multi vendor monitoring and management tools, it's less. I'll explain why. The burden to onboard our platform, so of course we work very closely with the client to do all of this, but we essentially do an asset dump, right? 

So every enterprise has a spectrum of asset management, let's just say, from enterprise to enterprise. So we bring that in, we clean it, we structure it so we know where everything is, there's a physical check to make sure that the right equipment is in the right place every. Again, we're focusing on rooms and meeting spaces and that kind of thing. So there's typically a specific vocabulary or nomenclature to those spaces that might be named after a sports team or something. I actually had an account of, believe it or not, that was named their rooms geographical places, but actually geographical places that they had other offices. So they had literally an office in Paris that had rooms called Milan and Berlin and everything. So imagine having to figure out how to train the model to not get screwed up with that. But nonetheless, that's our problem. 

So there's a bit of specific learning to the application. You onboard the devices and then really you're off and running because there's two main things that we do. We've done all the integrations with the devices so you really don't have to do that and unlike existing platforms where you have to do a lot of programming, like let's say you wanted a single touch to reboot a room or something like that and there's three or four devices or log into something, right? Typically, you'd have to program those subroutines and create those command structures, but all that's generated in our platform. So it's really saying, this is what I have, this is where it is, and that's it. So yes, there's onboarding, but compared to existing platforms, it's actually quite light. 

Is that onboarding part of the SaaS fee or am I paying like a number to get that part done?

Erik DeGiorgi: We're still figuring that out. What the objective and what we've done thus far is: you sign a contract with us. We do that work for you. There's no upfront fees and you're just paying your, your typical SaaS. 

But you're kind of learning on the fly and you may realize, Jesus, this is taking like two weeks of labor to onboard each our clients and need to do charge something to cover that 

Erik DeGiorgi: Well, that's why I chose my words carefully because I might we're figuring it out as we go. But that's the objective, right? The objective is that we just want to make it really clean, really simple. Hey, you're a customer you're locked into a year or two or three a contract, it's fine. We're going to get you up and running and we're going to support you along the way, and we're not going to nickel and dime you. 

So you talked to an integrator, I know you're kind of pre selling this to companies, at least making them aware. How are they responding? 

Erik DeGiorgi: Well, very well. We had a UC, industry veteran join us several months ago, he has about 30 years in the industry, Polycom, Crestron, and almost every day, we're having lots of really high level conversations with end users, integrators, partners at all levels.

We just kind of pinch ourselves. We're still batting a thousand, like every single conversation has led to more and everybody's very excited about it. You know, there's been this huge AI buzz. There really hasn't been a lot behind it, behind the buzz, and so we're super excited to show real world applications, bringing actual value to the industry and it's resonating, and we're very excited.

We had a pre-call and I asked about the level of activity among particularly larger companies in AI, and I was surprised when you said that it's largely just a roadmap item still for most companies. 

Erik DeGiorgi: Overwhelmingly, with a couple exceptions, the general feedback that we get is: we need to figure out how to do something with you. We know AI needs to be part of our strategy. You know, it's on paper and that's kind of about as far as we've gotten with it. 

If people want to know more about this, they could find you online at Netspeek.com, right? 

Erik DeGiorgi: That's right. Netspeek.com, and of course on LinkedIn, and for those going to the show we'd love to see you there. 

All right. Thank you, Erik.

Erik DeGiorgi: Great to catch up, Dave. Thanks so much.