The Hype Phase is Over: Time to Unlock Real AI ROI – Predict 2025

Good morning, good afternoon, wherever you are. I hope you are starting a really wonderful new year for everybody out there. Um, this is 2025 and one topic that is at the top of everybody's mind still is ai.

Now, the thing about AI is that you simply cannot ignore it. Whether you are a technology leader, whether you are a business leader or in any of the other functions like HR or legal or finance, AI is at the top of the mind. And you in your leadership position are asked multiple questions around ai.

You are asked to build a strategy around it. You are asked to build projects around it and then drive those projects to conclusion and to successful conclusion in spite of all the ambiguity that is around this technology that is around this topic. So the, in the next 30 minutes or so, we are gonna try and give you a path through this confusion, we are gonna share some of the ways in which we have accomplished success in this area.

I'm gonna share some of, uh, the things that worked out for us and also some of the things that did not work out for us, so you don't have to repeat those mistakes. So let's go. I think it's going to be fun.

30 minutes. So the thing about AI is that a lot of people talk about, and in the media, when you hear about it, they talk about the upside of it. Oh, it's going to be fantastic.

It is going to change everybody's life for the better. Uh, there's going to be a massive increase in our GDP or the GDP of the world and the productivity will go up and most of them are very possible and are actually true. There can be unimaginable amount of growth when it comes to ai, but at the same moment, and in probably the same measure, we also have to think about the change that it is going to drive.

And that change could be on the environment side, it could be on how businesses are done today. It could be the change that might impact your job, my job, everybody else's job. So it's very important that when we think about ai, we think about the potential of unimaginable growth, but we should also take care that we are ready for the indiscriminate change that it is going to drive.

With that out of the way, the first question you must be asking, who are you and why should I listen to you? So my name is the bukhari. I'm the Chief Technology and product officer for Extreme Networks.

Extreme Networks, if you are not familiar with it, is the largest pure play enterprise networking company on the planet. Um, and you should listen to me or to Extreme because we have now successfully productized multiple generations of ai. And the latest one we recently announced is our extreme platform.

One that really takes AI and narrates it with a platform, an enterprise wide platform, um, so that you can actually unlock the real potential of it. Now, AI as a product and AI as part of a platform, that is an entirely different topic, which we are not gonna talk about today, but something maybe, uh, for future. So this, so all the things that I'm gonna talk about today are really based on our experience as we productized and brought to market extreme platform one.

Uh, so I'm not gonna stand here, well, in my case, sit here and pontificate. I'm gonna share some real world experiences with you, some of them very positive, and some of them, uh, which will be gotchas, right? Okay.

So let's go. Each one of us wants to be successful in ai. And in order to wrap our heads around it, it's sometimes easy to put something as an equation.

So here's the equation that we have learned over our experience of productizing it. There are multiple components in it. The first and the foremost is user experience and the trust gap.

And they are conversely, um, or they're not directly proportional to it, uh, themselves, or they are inversely proportional as they say. So the better the user experience, the more success you can have, but the bigger the trust gap, you know, the less success you will have. And then there are a couple of other components, which is technology and also willingness to pay.

And willingness to pay is something that people usually don't talk about it. So I'll talk about that towards the end and make sure that we actually understand what we mean by that. If you notice in this technology is not the biggest factor in there.

Uh, and that is one of the biggest problems that happen with AI projects. The moment somebody asks you, Hey, uh, so and so you need to come up with a AI project for our company. And the first thing that people go, they look out and they go out and they start searching for the best LLM or the best model to use that.

That's just an absolute wrong position or wrong place to start with the first and the foremost thing is user experience and trust gap. So that's where we will start with. Now, one thing to remember when it comes to trust is if your people do not trust it, they are simply not going to use it.

And it doesn't matter how good or bad your AI product or technology or agent is, if people don't use it, it's not going to be successful. So think about trust upfront. Now, the thing about trust is that trust and value, they kind of go hand, hand in hand.

So the more trust people have in any AI product, the more valuable they consider it, or conversely, the more valuable it is that could drive more trust in it. But there's a thing called complexity in there. And typically the higher the value of the ai, um, you know, unless you are at the very, very start, uh, it's going to directly relate to the complexity of the project.

Okay? So when it comes to trust, there are two, three things that you need to consider. Number one is pick a use case that actually matters.

So the first thing that we are gonna talk about is how do you pick a use case to which you are going to apply ai? That's number one. Number two, how are people actually going to experience?

So what is the UI for that ai? And the last part is culture. So these are the three things that feed directly into this trust and value, um, you know, equation that we talked about.

So the first thing, how do you pick a framework or how do you pick, um, an experience that you're gonna apply AI to? How do you pick your AI use case? How do you pick your AI strategy?

Different companies call it differently, but a lot of the times what happens is that you get a call from your boss or from your board and they ask you like, Hey, we need a AI strategy, and off you go to the races. But I think the best way to do it is to take a step back and think about what are the use cases, current use cases inside your company to which you wanna apply it. And there's a very easy or simple way to wrap your head around it.

And we call it the ARC framework. Um, and this is accelerate, replace, and create. So simply speaking, it is like what are the experiences in your, um, domain, in your work environment or in your company that you want to accelerate?

What are the ones that you want to replace and which are the ones that you want to create afresh? And this is in your context. If you are doing AI for your team, you can run the ARC framework in the context of your team.

If you're doing it for your entire enterprise, you can do it in the context of the enterprise. And if you are productizing it and you are bringing it out to an industry, then you can do it in the context of that industry. But the framework nonetheless, stays the same.

What are you going to accelerate? What are you going to replace and what are you going to create? So here, let's just take a few examples.

These are simple, very simple examples that will make sense to everybody when it comes to acceleration. A lot of companies start with their customer support experience. Why?

Because nobody likes to sit on that call listening to that, you know, fun elevator music and waiting for a customer support agent. Or when you get to that agent, nobody likes to start from like, is the light blinking and have you turned it on? And is the power plugged in and stuff?

People want to get to their resolution very quickly. And a lot of companies, a lot of use cases on applying AI to that. And what are we doing here?

They are accelerating an experience that is already present there, right? And it is very powerful. That's an example of accelerating a use case.

Now, let's think about an example of perhaps replacing a use case. And we are in the networking technology domain, so I'm taking a lot of examples from there because these are things that we have already done in our portfolio. So this experience is the acquisition of knowledge.

Now, what do I mean by that? It could be finding information inside the documentation. It could be finding information in real time for a product that you're using or a problem that you're having, or even things like explaining something that the product is showing you.

You remember those error codes. Now, good thing we don't have error codes and SaaS applications, but sometimes information that is displayed, um, in those dashboards and in you in those, you know, my pretty, uh, graphs, you might wanna know what they actually mean. Now having a chat bot that is right there and giving you that information in real time, that is expo that is really replacing the experience of having to go out digging through documentation and websites.

So again, very simple example. Now in terms of creating a new experience, uh, this is something for example, right now at this point in time, um, especially in the networking world, if you want to pick up information from various different pluses in your network in real time and create this what if scenarios and stuff, it's something that is very, very cumbersome to do, to a point where I would say that they are not really something that everybody does in their daily life. So it's almost that that experience does not really exist.

So this is with the use of AI creating an experience where you can acquire the information, you can actually, uh, structure the information and you can build insights on top of it all within the same product within like minutes or so. So this is creating an experience that is brand new. Now this is just the example.

You have to apply it to your space, to your domain. But think in terms of what are you going to accelerate, what are you going to replace or what are you going to create? The chances are you're probably gonna start with the a part first.

Now, as you're picking this up, one example or one thing that I will share with you is do not pick a use case that nobody cares about because you are going to get some money to actually go this AI use case, or once you do it, you are only going to get more money if that use case were valuable. So go pick a use case that is valuable from day one. Don't go for the low hanging fruit, not when it comes to ai.

Okay? So this was the part of picking an experience. So once you have picked up the experience and you are trying to build something around it, the second thing that you have to think about is how the users of this AI application, what would be their experience?

Because as different experiences are developed during ai, the trust or adoption for that is not linear. You know, if somebody is used to a conversational AI and they will start adopting it, but when you actually, um, introduce them to an interactive or collaborative AI or to a ai, the adoption is gonna drop down before it comes back up. So consider these things as you build the UI for your program.

So how did we build the UI for the ai? Um, in platform one, we essentially went with three different ways of interacting with ai. So conversational, interactive and autonomous agents.

And my feedback, um, or my, um, sharing my experience with you, um, that don't go with just one because if you just put a chat bot in your product or in your project, uh, very soon you're gonna find that there are use cases that do not lend itself very well to a conversational interface. So always invest in and think about various different interfaces, very different user experiences. So here, let's just take a quick example.

This is something that is going to be very familiar to you, all of you. This, by the way, is extreme platform one. Um, of course this thing is gonna look a little bit different based on whatever product you are using, but the idea here is conversational.

So you have, think about that you have an expert sitting right next to you or another colleague sitting right next to you who happens to be an AI chat bot, and you are talking to that chat bot. And as you are talking, you are learning, as we talked about knowledge acquisition. Uh, you are doing things, but this is a conversational interface and this is something that people are a lot more familiar now in 2025.

So this is the base experience that you can provide in your project or in your product. But as we move forward, um, there is the second interface, which I think will become more and more valuable as we go past 2025. Um, and that is really the experience of creating real time dashboarding.

It's really a canvas. We call it an AI canvas. And the idea there is that there's a lot of information that is present in the products, but not always in a way, shape or form that is actually valuable.

And that is something that you can really, truly use. I'm talking about the users. So giving the ability to the users to interact with the system and create a new dashboard altogether.

So in this example, what happening is that the user is actually working with that conversational ai, and it is asking the AI to give it or give the user multiple different pieces of information. But while the user is doing that, the user has the ability to take that information and put it on a canvas. And as the conversation continues, it continues to put more things on the interface or on the canvas, and in the end it ends up with essentially a real-time dashboard, right, that they can now create and then now they can actually use it whenever they want.

This is an absolute powerful way of allowing people to use AI to create something that was not present there. And, uh, this has worked fantastically for us, and this is something that I highly encourage people to think about and use when you are doing your AI projects or your AI products, right? And we'll talk about the difference between a project and a product in a little bit.

So if you go to the next slide, um, you will see that the next big thing is obviously agent ai. Still, I briefly talked about, you know, the conversational as well as collaborative ai. Uh, but this is 2025 and everybody should be thinking about agent ai.

And if you are not, then I'm pretty sure you're getting peppered by everybody to think about it and come up with a strategy around it. Now, let's take a swing back on the agents, because before you go into the technology, it's kind of important and interesting to understand what an agent does because when we talk about agent AI in the industry, there are two connotations in which we talk about it. One is, um, you know, running AI agents behind the scenes do all these different things, right?

So there's a master agent, and then, you know, there's a bouncer agent and there's an agent that distributes the query and so on and so forth. But if you think about a curtain, and on this side of the curtain is the user, and on this side is the system. Agentic AI in the system is one thing, but then exposing those agents to the user side or user visible agents is something a little bit different.

And now here we are talking about the user visible agents. So this is agents that user would interface with. So some of the characteristics of it, the whole idea behind agent AI or AI agents is that they're autonomous in nature.

Now, autonomous doesn't mean that they cannot be a human in the loop. You can have both kind of agents out there. Um, but the key is that it also has memory.

So it has context memory in, in AI words is really context for the user. So it retains the context. Context, it is reactive or it can be proactive.

It is very specialized in doing a few things. Um, and remember the last part, it can interact with humans or it can interact with other agents. So this is kind of a rough definition of agents as they appear to the user.

Okay, so those are the agents, but how should I think about agents? And you should think about agents the same way as you think about humans. So I always like to give this example that, um, how do you create teams, right?

You find various people, there's definitions of their roles, they're expert in different things, and then they interface with each other, they interact with each other, and that creates a human team. Now, if you take it to the other extreme, what happens is that you have a bunch of agents. Now imagine that these functions are being done by AI agents.

And these AI agents need to be organized the same way as you organize human teams, and then they talk to each other. Now, honestly, these are two extremes of the spectrum. A fully human team, which is not only just the extreme, but is also the norm today, but in the far out world, we could or far out time, we can think of, you know, teams that are just a hundred percent based on AI agents.

But what's the reality? The reality is going to be that it'll be a combination of human and digital workforce. So when you are building an AI agent, when you are thinking about an AI agent, keep this context in mind that your AI agents should behave towards the human users as humanly as possible.

So with this context, uh, let's just give you an example. So this is how we have done agents in extreme platform one. Now the video will do a bunch of different things, and of course you can rewind it, you can pause it, you can look at it.

But generally what's happening is that when you have agents, you have a lot of AI agents, uh, you need to do the following things. Number one, there needs to be some sort of a catalog where you can find an AI agent that does a certain thing. This is pretty similar to where you have a job description and you go out and you try to find somebody that can actually do that job.

So you have to be able to find AI agents, then you need to be able to go and configure them. This is kind of, uh, teaching a new person how to do their jobs. This is like about, um, now on the AI side, you actually go and configure that AI agent.

Um, and then once you have configured it, then you need to do lifecycle management and you run this. And in that configuration, it, that agent might be interacting with other AI agents or with human people or the human workforce out there. So that's just the example of how to think about AI agents.

So when you think about AI agents, start with the context of that this is digital workforce and it has to work with humans, and then start thinking about how to organize them, how to make them discoverable, and how to make them configurable. Now, once you do that, um, then only think about the technology. So I can spend a lot of time about technology behind the scene, but, um, what I would tell you is that most of the times, um, what our experiences, and we have hundreds of thousands of, you know, users out there, like 50, 60,000 big customers all the way from midmarket to, you know, fortune 10 out there.

And what are experiences that as people go into building these agentic ai, most of them will probably fail, but that's not a bad thing, that's a good thing because that's experimentation. Uh, and that just helps with that AI literacy across the organization. But once you have done that experimentation, you are gonna be, um, dealt with a question, the age old question of are you gonna build it yourself or are you gonna buy that as a product?

And I'm not gonna tell you to lean one way or another. You can build it yourself, you can buy it. Uh, but that is a decision that you have to take before you dive into the technology.

Of course, we are a product company, so it's not like we can actually buy it, so we have to build it ourselves. And as we built it, these are the experiences, um, that we had as part of that. Now, there is so much technology, there is technology overload essentially when it comes to AI out there.

So it just to organize it, and I'm gonna build out this slide just to organize it. We think of it in three big buckets at the, at the bottom of it is the data hub. Now e you can call it data hub.

You can call it data pipeline, you can call it data fabric, whatever you call it. But the idea there is this is the layer in which you are collecting data, you are organizing data, you are indexing it, you know, and you are putting data compliance in it. This is where who can access what resides.

This is where, uh, you know, your data residency and data governance pool, that that is where all of that sit. Then on layer above that is what we consider as AI services. This is where, you know, your model management sits in.

This is where your multi, um, you know, your rag architecture sit in. This is where if you are multimodal, that's where it's set in. So how do you organize all of these AI services together?

So that's, think of that as a second layer. The third layer is perhaps the most important, uh, in the actual usage of ai, and that's the safety and guardrails. This is the fact that, you know, how do you deal with hallucinations?

How do you ground your models? What is your resiliency? How do you build a human in the loop kind of on architecture and stuff?

So lots and lots of technology around there. And the reality is that I'm not gonna go ahead and give you recommendations because by the time I've finished this presentation, there's gonna be new advancements in this area. But what I can do is share some of the experiences and some of the learnings that we have in this process.

So starting from the bottom right in the data hub, uh, don't go with like, I need all of the data in the world to be able to do this. The most important part in the data is the data governance. Start with the data sets that you already have, but make sure that you think through the data governance, um, as one of the most important thing in your data layer.

The second thing which is really interesting is that, um, where is that data coming from structured, um, and unstructured. And if you are producing that data, then push down to those teams that are producing that data the right way to do it. I'll give you one example.

What we realized is, um, that a lot of the documentation that is being written, the simpler the English, or I would say the language simpler, the language that is used in that documentation, the easier it is for an AI system as opposed to when you're writing it for a human consumer, in which case, you know, you want to add more, uh, context and perhaps, you know, a little bit more flowery language and stuff. So things like that become really important in your data layer, in your AI layer. Um, quite frankly, uh, rag, uh, or some variation of it, like metadata based rag and has rag whatever rag you're using, because that is what will really help you when it comes to, um, trust and stuff on top of it.

Um, so at this point in time, rag architectures are really, really, really, really critical. Now, when you come to safety and guardrails, quite frankly, I would tell you don't go on this journey on your own. Um, made sure that you have some really good strategic partner for us.

Microsoft and Amazon are two really big partners, and we have learned a lot from there. Um, and we have incorporated a lot of their technology. But one thing that I will point out is that fact checking mechanism, which are really that, you know, how do you deal with hallucination?

How do you deal with, how do you ground your models and stuff? So that fact checking mechanisms are really important because remember, it doesn't really matter what your AI does. If people don't trust it and don't use it, there is no value to it.

So these are some of the things that you should consider if you are building your own project. And these are also the things that you should consider if you are buying AI from somebody. And at that point, this becomes questions that you should ask, okay?

Now I said, we'll come towards willingness to pay. And you might be thinking to yourself, Hey, I am building an AI that is for my own team or my own organization, or my own company. Well, this willingness to pay doesn't really, um, you know, come into, into play here.

But what I will tell you is that it does, because if you are building a product, then willingness to pay is very simple to understand that is something that your customer is willing to pay you for it, but if you are building an internal project that that willingness to pay really shows up in the, in, in the form of investment in your project. So how much your organization or your company is willing to invest in it, that's really also willingness to pay as well. So with that context, how should you think about willingness to pay?

Now this is very simple graphs and these graphs, I've kept them super simple on purpose, uh, because these topics can be complicated, but they can be presented at least at the very start of it very simply. So on on the X axis, you have a done adoption and know the y axi, you have willingness to pay. So willingness to pay actually increases that adoption, but there is a tipping point unless you have reached that tipping point.

Um, your ability to charge for it, either to your company or to your customers is not gonna be that high. And then once that chasm of adoption, um, is jumped or, or your users on the other side of it, then the willingness to pay is directly tied to the clarity of the ROI. And that ROI is based on what is the value of that ai.

Now, if you apply it to the ARC framework, generally speaking, this is not a hard and fast rule. There could be exceptions, but the more you are, um, the higher up or the farther out you are in the ARC framework, the higher the willingness to pay attached to it. So if you're just accelerating in experience, which is really truly automation and efficiency game, a productivity game, there's a certain willingness to pay as opposed to when you are creating a brand new experience, which might actually start a new brand new revenue stream there, the willingness to pay would be very different.

So these are some of the kind of contours of this conversation that you should be, uh, familiar with or you should keep in mind. So now in what we have seen in our efforts to productize this is that conversational and interactive ai, as we described earlier, they lend themselves better to just being part of the subscription. So what do I mean by that?

They're just part of the product, they make the product better. Or if you're an internal project, then it makes your project better, you know, uh, cheaper to run, or people love it more and they adopt it more. But this is the willingness to pay is embedded into the product or the project.

But as you go towards autonomous agents, the autonomous AI agents, that is where each agent can have a very clear determinable, demonstrable, ROI. And at that point in time, your willingness to pay internal or external should be attached to that. So what is that?

What is that ROI? Now this is an area that I would say is at the very start of its progress and there, so I'm gonna leave you with a few thoughts around this, um, as you determine these out. So how do you price this?

And then remember, if you're building a product, it is pricing. And if you are building an internal project, and this is about how much investment you're gonna ask, so it's, it, it it works for both. Are you gonna base it on the current cost of delivering agents, right?

You say like, Hey, it takes me X number of dollars, you know, to build this AI agent, so I'm gonna price it that way, or that's the money that I'm gonna ask for this project. Um, well, you could do that. That's a very simple way of doing it.

Uh, but the reality is that the cost of delivering this agents is going to, uh, crash over time because of the economies of the scale and the cost of the AI is coming down, uh, pretty, uh, tremendously over time. Um, so that might not be a good way to do that, or would you do it on the future cost of delivering this agent? Well, you don't really know what the cost of that agent is, or would you do it at the current cost of the task that the agent is going to produce or to deliver.

So think about it this way, I said that think about AI agents as more like human agents or human workforce. So typically what you do is this, this is a job description and in the market, this is the pay associated with that, which is a generally accepted cost of doing the task that the job does. So you can take that model, um, you know, in, in view as well as you're thinking about it, um, or you can also think about that.

What is the value of the outcome that it is still ing. So, you know, what is the task that the AI agent is doing and how do you value that? I would tell you that there is no easy way to do that, you know, so these are just some of the questions that you have to think about, but what I would tell you is that do not leave it for later when you are going into the ai, you have to think about all of these things in one go, and we'll just do a quick summary of it.

In order for you to be successful in ai, you have to think about user experience and trust. These are the two main components underneath this user experience and trust is how do you select the use case for which you're gonna apply AI and use the art framework or any other framework that works for you? ARC has worked really well for us, it works very well for our customers out there.

Uh, pick use cases that you wanna accelerate, you want to replace or you want to create. Um, and then that's a good way to start getting started on the use cases. Uh, pick a use case that actually matters.

Don't go for, you know, low hanging fruit and then consider what are the, what is the UI or the user experience you have to deliver on those use cases to help people get over that trust gap. So you start with conversational, you know, you start with human in the loop, then you maybe take them towards that collaborative interactive where some of the functions are done by the humans, some of the functions are done by the ai, and then eventually taking them towards these ent AI that can be very, very independent. But the key is that this is a journey.

So think about user experience and journey. Now, the two other things, obviously technology, technology. I kind of shared with you some of the things that you need to consider.

Remember, how are you gonna manage your data is going to be really critical. And then how are you going to build those safeguards and guardrails on the top? And then only the technology.

Uh, I know I'm, I'm A CTO myself, but I'll tell you, uh, that when it comes to ai, technology is comparatively the least important part of the equation for the success in ai. It's important, but the least important. And then lastly, think about willingness to pay.

It could be pricing if you're building a product, or it could be investment if you're doing an internal thing. Look, uh, I hope this was a useful conversation, um, because this is based on years of years of experience in productizing ai, successful productization of ai. And we believe that the path to the future is going to be brighter and better with ai, but there are hurdles on the way.

Um, but if you're cognizant of that, and if you are willing to think about that upfront, then we can actually be very successful. I wish you all a great start to the new year, and may your AI projects be successful and may you be, um, you know, on at the end of this year thinking about all the great journeys that you have been on. With that, thank you very much.

This is Nabil Luri signing out. Enjoy the rest of the conference.