Selector AI Conclusions on the Future of Network Automation

So, just a quick summary, right? This is a purpose-built Kubernetes native platform for hybrid cloud and networking. AIOps designed for today's hybrid cloud and networking ecosystem and tailored to meet the customer's needs.

I think the big differentiator between us and, and most other organizations trying this is that AI was a founding principle. We didn't have the solution and we're trying to bolt AI on at the end as some widget or some assistant that doesn't really fit the platform. The platform was purpose-driven.

The other thing is, I think from a year ago when Nitin presented here, it feels like 30 years have passed in the world of technology from a year ago. And I think a year ago there was still some, let's say, valid skepticism about artificial intelligence, generally speaking, not questioning our solution or that we were, you know, not able to do what we claim we can do. Just a general uncomfortable unease about AI.

A year has passed and I think we've earned more. It's, it's earned our trust, it's earned more comfort level. It's becoming something we do every day in our life.

Right? Is there anyone on the panel who's not used? Chad, GBT yet?

Anyone? Is there anyone in the room? No.

No. Everyone has used this stuff. Right?

The other thing is, again, that envy of software. They have copilot. He, it took him 10 minutes to write a Python script.

'cause he's got ai. Why is it taking me three days to write a network configuration file? Why is it taking me five hours to troubleshoot a network problem?

And why don't I have the tools to do it? Everyone else has the tools. Well now the tools are here, right?

A, a co-pilot for your infrastructure. Um, you saw me do it from my phone through Slack. It works through teams, it works through WebEx.

And in our roadmap, we are, we are really closing in on an MCP and adopting and embracing the next wave of, of artificial intelligence, which is model context protocol. So, um, for example, NetBox has an MCP selector is working on an MCP. Our customers have been asking for MCP, so that way they can start extending selector to connect with other solutions.

So I'm very ahead of schedule here, but I'd love to thank you for your time and I'd love to take more questions and comments. Yeah. Um, please.

So back on the ROI discussion. Yes. You focused very tangibly on 58 events, not 58 separate tickets, one ticket that summarizes everything that happened there.

Are you taking the ROI discussion to the human level? So like, how is that, do you have any good stories from your observation or customer reports on how that's freed up time? How that's enabled, um, op staff to do more?

Can you speak to that? Yeah, so typically, um, you know, as a vendor, we, we don't hear a lot of, Hey, you guys are doing an awesome job and we love what you're doing. There's a lot of, Hey, could you do this better or could you change this?

Sure. But to that point, um, some of our customers, um, speak about our solution publicly, right? Like we saw Jeremy at Auto Con speak about it and how selectors become the cornerstone of their automation, um, capability.

Um, I've spoken to other customers. So here's another example. We have a customer, we're in a proof of concept with to close the loop on the automation and they estimate it's gonna save $500,000 a year for one recurring problem.

They have, they did the numbers on this and they said every year we're spending about half a million in human and human and other capital to resolve some issue with a service provider. That just happens constantly. Sure.

It's just a routine thing. Those people, they, they've literally told their network staff, you do not have to deal with this problem anymore, and we're putting that half a million dollars back into the organization. Why?

Because selector can not only detect the root cause, but now we can kick off an automated workflow to shut the port and let the service provider know, right? It's not something they can do anything about. It's a service provider issue.

Mm-hmm. But now they can do something about it. They can shut the port, send the alert, open up a ticket, notify the provider through agents, and ideally when they receive the notification that's been resolved, re-enable the port and bring the service back up.

Yep. So actual truly tangible impact, right? Um, another customer had, it was funny, they had a pre and a post selector kind of, uh, story for me.

They had a firewall change that was approved, documented, approved, went into five pilot sites. Those five pilot sites were black hold from the network network team, no idea looking at network stuff, right? Someone clues in three days later that this was an approved firewall change and we've rolled it back and now the sites are back online post selector.

They had a smart alert. These five sites are offline because of this firewall change, right? I mean, light bulbs go off.

People go, whoa. Wow. That is powerful.

Right? So, um, I love hearing the success stories, so if you have success stories, please let me know, right? We're happy to hear the complaints and the, the improvement suggestions, but we'd love to hear your, um, success stories.

Is there a donna Sorry, go ahead. Is there a way to introduce your own teaching into the network learning model? Uh, for instance, I'm sure I'm not the only one who's done this, but I've created, um, uh, routing level coupled systems to solve a specific problem.

Okay. Uh, to something that may not be completely obvious to a traditional root cause analysis. Okay?

Uh, uh, but I'd like to be able to say, Hey, these are the events that we saw, um, AI apps, you gave me a a root cause that's wrong. This is what actually what happened. Is there a way to do that?

So yes, we, we do have a feedback loop in that you can thumbs down the answer, which feeds back to the model that the answer it provided was not valid or rejected by the user. So there is natural learning within the system. You can give it a thumbs down and it feeds back to itself.

And then if you follow up, it will give you kind of a me culpa, I was wrong. Here's the correct answer and try it again from its own learning. Part two of this is that we have a custom query dashboard.

So let's, let's start with the dashboard, right? I'll put a query in through SQL to create that view of what you wanted to see from there. There's a natural language button that I put my human phrase in.

This is John's custom unique situation or whatever, right? So then when you prompt it later, could you show me if there's anything going on with John's current John's custom solution? It will then bring you that dashboard and that context that you set up as a query.

Does that make sense? Yes. Okay.

So I, I like the, show me the work bit. Can you pick apart every element of that? Show me the work and say, Hey, you know, this is where you went wrong.

Do this instead. No. So yes, you can show the work and click on a a, so it's, um, the, by the show me the work, it's the knowledge graph.

So it shows you the correlations that were made and we have a sliding scale. Show me tighter coupling of correlations or looser coupling of correlations. So you'll get more or less lines in the correlation view.

And then you can click on an individual node and it will give you the payload from that node to help reinforce why it answered the way it did. But currently our feedback loop is through the natural language, a thumbs down, this answer was incorrect, right? And, and that's happening less frequently as the models improve.

So some of our customers who move from say, llama to Gemini Flash two, which has really big context windows very fast. We're seeing really remarkable results from the Gemini model In terms of selector AI's, uh, sales, uh, cycle. And I know who you're selling into.

Any observations about, are you seeing a shift in the personas? Are you selling to say AI teams more, you know, chief AI officers, right? And also, you know, how's the sales cycle coming along, uh, in terms of, you know, gaining more brand awareness and mind share out there in the ecosystem?

Yeah. So, um, I'll work backwards. We're, I think we've had a boom in the last six months or so in interest, I think partly because of mitten's presentation at this event, which led to real customers and real customer engagement.

But Steven and the marketing team have put a ton of work, Gianna and Steven and, um, Dallan and Satchin, um, a concerted effort to be more engaging on LinkedIn, to have more blogs, to have more webinars to get ourselves out there. And my role of trying to, you know, evangelize AI at a high level, right? Mm-hmm.

But also more focused on our product and our demonstrations and doing, you know, engagements like this. We're going to hit 10,000 LinkedIn followers this week, probably. Um, our engagement is up 60 or 70% over the past few weeks and months.

Steven deserves a lot of the credit for that. He's a really hardworking guy. Um, the other thing is there is customer to customer, right?

And there's a little bit of, um, fear of missing out going on, right? So I think as, as we grow, it's gonna, it's momentum, right? Yes.

It'll start to snowball and then eventually we'll become a, a, you know, a, a recognizable name in the industry. I, I'm not sure that we're there yet. We have a lot of work to do.

Um, we've secured a booth at Cisco Live to help, uh, with our brand awareness as well. Um, and I'm gonna be speaking at Wireshark and OG and a bunch of other events and auto con hopefully. Um, so no things are going well in terms of, um, the market penetration.

Um, healthcare is very interested, alright? And it kind of makes sense, right? They, they wanna put their money into new MRIs mm-hmm.

Right? Not more networking staff, right? So, so everyone's fighting for dollars and we can fit in nicely to be an affordable solution that offers them digital workers, digital coworkers empower your existing staff with this new member of your team almost.

Right? I dunno if you saw Jensen Wong from NVIDIA's, his keynote a few months ago, he believes that the IT department of the future is going to be an HR department for ai. Right.

That our jobs are gonna completely change and we're just gonna be policing our agents to make sure that John's agents talking to Bill's agent correctly. Right? Uh, it's an interesting perspective, but healthcare is interested, finance is interested.

Anyone with big data is interested. Uh, ISPs obviously more and more ISPs are coming on board. Um, customers with like multicast challenges, different unique segments.

It's, it's funny when you, when you offer a platform like this, the type of problems that come your way and some are completely unique, um, a cruise ship with two data centers that's out in the middle of the ocean and, you know, if there's a problem with their network, it's, they're out in the middle of the ocean, right? So, so it's a lot of interest. Right on.

And are you so like to networking teams mostly, uh, are, are there is any diversification of, you know, across all these organizations who you're talking to? So certain application teams are interested, but it primarily, right now it's, it's infrastructure teams and, and I would say cloud teams and data center teams. Um, and the interest tends to start with a network engineer.

So it tends to be grassroots and from the bottom up. We don't get a lot of, well, we get some, but it's not a lot of CEOs. I want selector and I want it tomorrow.

And it gets pushed down into the org. It tends to come from practitioners and engineers who happen to have seen a video or a post or heard about it. All right.

I'm thinking service provider, ISP type world, um, obvious is gonna be very useful to the network engineers and things like that. What about customers calling in? Uh, do they have actually had, or is there a simplified view for them so that, you know, when they start asking things, the context is only their in their VRF or, you know, in their, in their, uh, v VP n setup?

Yeah. That, that Provider, that's a really exciting opportunity for us. So one of our large ISPs is, um, looking to maybe monetize our solution as customer facing, which makes sense, right?

They, they pay so much for our solution and then they can turn around and maybe make a profit off of that solution, right? To offer a portal that customers can come in and say, here's my circuit id, what's going on, right? So we are not going to, are are not going, likely not going to do that, but our customers can use our platform as they will and can make it customer facing if they want with a few tweaks.

I, I mean, I just worry a little bit, you know, accidentally hallucinates tells me all about somebody else. Yes. Things of that nature.

It's like, well, yeah, it's an answer, but, um, it wasn't the right context and things of that nature. I think risk is the only thing like you to your exactly, to your point risk is what's holding that back a little bit and, and confidence and trust. But once we get there and have the appropriate guardrails, I think that you might start seeing some natural language interfaces from your service providers or, or other companies that is actually using selector as the backend.

Yeah, I could see that coming in the, in the very near future. I don't wanna name the ISP 'cause I don't know if they want me to tell, but, um, there is a major ISP looking at monetizing with a customer facing portal. Okay.

Back to the sales. Uh, is it mostly direct or do you have indirect? Uh, Uh, we do have a channel channel partners.

Okay. So we do, um, a certain amount of our business is done through channel partners and um, a lot of it is direct though. Okay.

Yeah. So about 80% direct. Is there a Way to kick?

Yeah, Steven, do you have maybe some numbers on our, Um, so we have a diversified go-to market approach. Um, you know, we work with channel partners, you know, like WWT, myriad 360, Caresoft Presidio, and you know, they are out there, you know, augmenting our sales team on the other side, you know, through our marketing business development and sales team. You know, we're working directly with the customers.

And so it's really a multi-pronged approach. I'd say our revenue split is like 70, 80% direct right now. And then the other, you know, is coming from channel customer referrals are also huge for us.

So like, you know, word of mouth is big, um, so a lot of our customers are telling their peers about it and that's led to a lot of proof of concepts for us. Uh, and also, you know, the biggest thing for us is we're focusing on showing the experience. Like at network field, day out at Auto Con out at og, um, a lot of the reactions we get right are, can you actually do this?

Right? And the written word only goes so far, but then when they see this actually happening, they're like, wow, this is real. So we're really focusing on showing the art of possible and what we can do.

And we're really focusing on getting in front of the customers and the prospects and being, you know, in their environment so they can see the reality of the solution. You know, we started to expand internationally. Now we have an office out in Japan.

You know, we're doing all these in, in person proof of value, proof of concepts. 'cause we wanna deliver a solution from day one that's working. So our focus is that POC is actually delivery ready at the end, right?

Like, you're ready to go on day one. So we've really invested a lot. Our engineers deserve all the credit for that.

Our customer facing teams, you know, they invest so much time into every, you know, proof of value, proof of concept because we're, we're really focused on that customized use case, right? We want it to work for them. Not every network's built the same.

Everyone's got different data layers, data lakes, you know, we can't normalize one fit all solutions. So we focus on making sure they get the value from day one and work with them to make their data work for them. Good to see that momentum and that springs to mind, uh, the question say over the next 12 months Yeah.

How well selector be defining success. I mean, what are some Sure guidelines, uh, uh, mileposts we can, uh, look at? Yeah, so, you know, we're on track to, again this year, you know, double our customer base up.

We've really been doubling every year. And our net retention is actually the thing we're most proud of. So we land a customer, they're expanding by 70 to a hundred percent after year one to from their initial use case to the next.

So the product becoming sticky and becoming valuable and expanding to other departments, other teams, other business units. That's how we're kind of defining success right now, that land and expand motion. We wanna be seen as that augmented team to our customers so they feel select as part of their DNA.

Yes. And, uh, I definitely see how it's, uh, helping improve, you know, network intelligence Yeah. And so forth.

And also, uh, saw, you know, business outcomes invoked, uh, uh, uh mm-hmm during the whole, uh, conversation by John. And I guess anecdotally or, you know, tangibly what are the business outcome tie-ins? Uh, are they saying, well, with selector, not only are we gaining these network, uh, benefits, but it's also improving these, you know, uh, specific business outcomes.

So let's think operational efficiency, right? Um, one of our customers, I can't name them obviously, but like we know the average cost of a ticket is $2. Right?

Before selector, they were generating 150,000 tickets a month, right? And that was costing 'em close of $4 million a year. Just in, just in the tickets.

Just in ticket costs. Not solving the problems or human cost just in tickets. We reduce that down to 40 KA year.

That's a simple, you know, operational impact There. And that's just the cost of the tickets. That doesn't include the ripple effect of how much more free time everyone has.

'cause they're not fighting 150,000 tickets a year, right? Yes. Alright.

And also think of, you know, your engineer's happiness, right? Who wants to spend all day going through all these dashboards, all these spreadsheets, stitching together the data. If you can knock that out in under an hour and then all of a sudden you can spend your time innovating, your engineers are much happier.

They're doing things that are more impactful in the business, growing the business. And it's also what's an outage worth to you? You know, we've talked to multiple people that like, hey, an hour down is $2 million in lost revenue.

Right? You know, if we can take away that hour and those maintenance windows and make 'em more visible and they're ready for them, it's just so much more impactful. And like a saying I think of in my head, right, is instead of worrying about that next incident, you're inviting it in a way because you're ready to deal with it.

You're not afraid that all hell's gonna break loose, right? And you have to drop everything, every team's panicking. You're gonna have all these angry customers coming.

You can get to resolution so much quicker because you have access to that data in one single point instead of trying to log in and everywhere and talking to every team across the globe. And then sometimes you have to have, you know, down detector yelling at you and you know, you got all these angry fans on X and everywhere else, you know, screaming for your names for 10 hours. That is very tangible to, you know, your business' reputation.

And that's why we really focus on companies where the network is the revenue driver. Right. You know, if that's their focus and that's how the business operates, there's so much value for selector there, right?

We're not a fit for everyone. We don't try to be, you know, a small network that, you know, has like a hundred devices and you know, they did a pretty good job setting it up. You know, they don't have a lot of downtimes.

It works for what they need. We're not, probably not the fit there, right? But if you have thousands of devices and or you have 500 to a thousand, right?

And that downtime is mission critical, we're a great fit there and we can help tell that story and we evangelize those teams. Yeah, it's definitely sounding like you're improving the workforce experience and, uh, playing a integral role in transforming the culture. You know, like making, coming here, you know, well, less arduous and well like simply more fun.

And you know, I've always found it rare. I've been, you know, in marketing for over 15 years now. Customers in enterprise are the hardest to get to talk about you.

Right? Whether it's behind closed doors out in the public, our customers are so happy with the outcomes they're talking about us, whether it's in public at conferences, or whether it's behind closed doors, referrals, whatever we need. Like they really believe in the value we're showing them.

And to me, that's something I've never seen before because usually they're very protective about what they're using, right? There's all these guardrails, they're very willing to work with us 'cause they're so happy with what we've provided them. Good to see.

Thank you. Um, on the mix of deployment models and the work that it takes to adapt it to a customer environment, how is that working in terms of, are you relying on those reseller resources to do implementation? Do you, do you do customer education for your customers?

Do you need resident engineers? What, what's the mix of resources there? Yeah, we, we try to do it as a, as a service data science as a service, right?

So there's certain things, CIS log, S-M-P-G-M-I, it, it's pretty standard stuff we can deal with. Sometimes there is, you're still dealing with XML or you're still dealing with some custom implementation or, uh, so the work happens on our end though, right now we have had customers delay until they got their data sorted out. Okay?

Right. So here's an example of something like a source of truth, a NetBox or an auto bot, right? We had a customer that didn't have one.

We, we would stand that up for you as part of our deployment, we will stand that up because it's so, one Of those third party Tools, one of those third party tools, we can help you get your source of truth in order. Because without that, right? We don't have the me we don't wanna release a solution that's missing a huge piece of metadata.

And all you get is IP and MAC addresses, right? Not locations or racks or whatever. Right?

So, so there is a level of customization that we can work together with the organization. Um, but our, and I don't think it's arrogance. Our data team says, bring whatever you want, bring the data to us.

Um, just on the point about selector, again, I've been, this is my first time with a startup and what I find remarkable is just how, how much we can achieve with such a small team there. There's less than 150 employees. Um, but, but compared to a giant barge in the ocean, a little kayak is nimble and can pivot and change directions very quickly, right?

So, you know, MCP came out, was announced in December three, four months ago. Our team's working on an implementation. That's how nimble we can be.

And we're all working together with the same goal. We all wanna make selector, you know, a billion dollar company with lots and lots of customers with global, you know, growth and become a name brand. We're all rowing in that same direction.

So it's been a remarkable journey just for culturally for me to join a startup and see how hard we work. And, um, you know, they buy me loud shirts if I ask them to, whatever they need, whatever I need. Um, so we've got about 15 minutes left.

If there's any other questions or comments that we're, we're happy to take them. I had another question on, on, uh, like tenancy, multi-tenancy. Um, so you showed, you know, each customer has their own, you know, bespoke environment where their data safe and separate from the others, right?

But is there other h is there additional hierarchies of multi-tenancy? For example, for like if I'm an MSP, right? I want have this and I want all of my customers in selector, but I want to make sure their data stays separate, right?

Or to, you know, like John mentioned, like exposing it from a, a larger service provider and they just wanna make sure that that customer's data stays separate. Is is in one, I guess logical instance, is there, is there a concept of, of another, another layer of package? We would likely have that similar view of three customers.

Okay. But within a customer. I see.

So we might have, even, even if it's the finance department doesn't want their finance data incorporated into the manufacturing data, we can deliver micro network language models just spinning up more pods. I see. And then from a, from a tenancy point of view, if, if, if I'm an engineer from tenant one and I'm asking how's the health of my network, it will only reflect my instance of the network language model.

Okay. Now does that, is there additional complexity on like the source of true site? You have to do, you have to have like different set source of truth for every single one?

Yeah. Ingestion to, to specific data lakes. Okay.

There's, that is another layer of complexity and challenge multi-tenancy, but we can deal with it. We have dealt with it. Um, the other point I wanted to make, Steven brought up a good point about tool consolidation.

Uh, you notice I didn't say here that we're here to replace anything, right? I'm not here to replace your SolarWinds or rip out that solution you've had for 15 years. Healthcare brought this up when I was thinking about it.

Um, we're gonna augment it, right? So if you're sending everything into SolarWinds, we can tap into SolarWinds API and bring that data over and over time we'll start to gradually phase that out or shift that out, right? It's an augmentation, just like AI's augmenting people.

Selector is augmenting your existing tools and your existing processes with, with, you know, with a goal in mind of being the only tool that you need. But we're not here to, you know, rip and replace some existing solution. And that could be John, I'm go ahead.

Yeah. Does selector, I don't know if you've covered it already. I apologize if you did, but does Selector have a discovery capability?

Uh, yes. Yes we do. We can go ahead and discover, uh, your network for sure.

Yes. Through our collectors and you Can map it. We To See the, the diagram of Hi Carol again.

You mapping the network. Um, So back to your operational twins. Yeah, we, we can map it in this sense.

It's a little high. Yeah. You want core distribution access.

I Would like you to map my data center and give me tools that I can put into Visio and look at it again. Okay. And then I wanna change something.

And can we do that yet? Not yet, not yet. Multimodal capabilities are coming though.

Okay. Yeah. Thank you.

No, it's a good question.