Which AI in IT Management Actually Works? | Rakesh Jayaprakash, ManageEngine | TFiR

Swapnil Bhartiya: When it comes to AI and of course enterprise AI is showing up almost everywhere in IT management tools. But buyers can’t tell which features actually do the real work or actually save real work and which are just either the hype or just say AI label on top of the box. We are here in Orlando, Florida. Manage Engine is hosting user content and managing also ship ZIA agents across their whole suite from service management and observability to security and endpoint ops. And it’s user conference. Where we are is where customers are actually giving feedback and also influencing what actually goes into the platform. And to unpack all of that customer feedback, what Managed Engine is building, we have with us Rakesh Jay Prakash, Director of product management at ManageEngine. First of all, Rakesh, it’s great to have you on the show.

Rakesh Jayaprakash: Absolutely. Pleasure to be here.

Swapnil Bhartiya: Of course I am going to talk about AI and the holistack. I want to hear from you talk a bit about this conference, the interactions that you have had with the customers.

Rakesh Jayaprakash: Yeah. So this is our annual user conference in the us. We do at least one in the US every year. It’s a great way for us to kind of share updates in terms of what we are doing with our platform, where it’s headed, which is what our customers want to see in terms of what is the direction that we’re taking and how it will benefit them down the line. And of course a big part of the conference is also interacting with a lot of the customers, understanding how they use our products. And it’s a great place to kind of get a lot of feedback. Right. Because I always like to say we sit there, develop the products, but the customers technically sit with the product for a lot longer than we do. So in some ways they know our platform a lot better than what we do.

Swapnil Bhartiya: Yeah, because the user always knows the flaws that you never. Or that thing that sometimes users surprise. Oh, it can do that also. I never thought of that.

Rakesh Jayaprakash: Sometimes it also goes the other way where customer says, oh, I use this particular feature this way. And I was like, I didn’t even know it could do.

Swapnil Bhartiya: Exactly.

Rakesh Jayaprakash: Yeah.

Swapnil Bhartiya: So they always surprise. Now another part of that equation is that every vendor has AI on their roadmap. How does your team decide whether it’s based on customer feedback or your own roadmap, that which AI feature should get into Manage Engine product versus the ones that that should just stay on the bench or the white bull for now. How do you decide that?

Rakesh Jayaprakash: Okay, so one of the biggest measuring sticks for us to decide where to put AI is first and foremost, the reason why people want to even invest in AI is to help them cut down on the time it takes for them to do a lot of the manual stuff. So what we look at is what are the things that are consuming a lot of time for our customers? And you know, management has a wide portfolio of products from IT Service Management, observability or Endpoint Security and so on. So if you look at each of these domains, the idea or our thought process is where is the customer spending a lot of manual time, which can be automated. So if you look at the IT service management side, nobody wants to deal with tickets. So how can you even try to resolve the end user’s issue if even before the ticket gets raised? And when you look at Endpoint security before a customer tells you that there is something wrong with their particular workstation, let’s say an application crashed. If we put AI in the endpoint agents and if they can kind of run and understand what’s happening on that particular endpoint proactively, IT admins will know that something is wrong. And so, yeah, so we are trying to do things which avoid issue getting reported in the first place. So that by itself is a huge advantage for customers in terms of cutting down the time it takes for them to sit and reply and respond to tickets.

Swapnil Bhartiya: These user constants, you know, as you said, in the US you do globally, but once a year in the us that cannot be enough, you know, to get that. So what kind of feedback loop you have? How do you actually, what you explain, how do you actually measure, you know, whether an AI feature is actually delivering the value that you talk about versus it looks great in the demo, it looks great in the keynote slide, but when it goes to the customer, nobody actually even used that feature.

Rakesh Jayaprakash: I mean, that’s a very true statement and something that I personally relate to a lot. Because the conversational assistant that we’ve been talking about as part of this conference, which is Ask Zia, has been before LLMs, we used NLP. It was working on NLP technology. We released it at least I would say, eight years ago, before the whole LLM thing landed, which was aimed at people going in, asking questions, getting answers immediately, but in reality, not a lot of people were even using it. Right? So what we do is of course anonymously custom. I mean, because majority of our applications are on the cloud, we can see the usage stats for the various applications that we are running in our own data centers. So that gives a sense of, you know, how much of these features are actually being used by our customers. And two, even in the on premise world, again, customers can choose to share anonymized data in terms of their usage. That is a big, you know, data point or data set that we can give us immediate feedback in terms of whether a particular feature is being used or not. And of course in addition to the larger user conference format, we also do a lot of smaller seminars which is probably for half a day or one day across the globe. Where it’s in smaller cities, we ensure that we kind of covered, for example, if you’re talking about us, we ensure that there’s one in the east coast, west coast and of course everything in the middle. So yeah, it’s a combination of telemetry data that’s coming in from our cloud usage as well as our interactions with our customers, not just at the conference, but also in the individual customer visits or the smaller format seminars.

Swapnil Bhartiya: Excellent, thank you. Now, when it comes to IT teams, they don’t buy a lot of this marketing fluff. They do often ask where will they see AI make kind of real difference first, it will change their day to day life. So which manage engine products from your perspective or based on the customer interaction, have the most visible impact on user experience? As soon as IT teams start using it, where they’re like, hey look, we are seeing that impact from AI versus you will see the impact in like 6, 7 months. So what is that?

Rakesh Jayaprakash: So I would say if I were to pick a few solutions where the impact is clearly visible very immediately, I would probably start with Service Desk because it’s one of those IT tools which pretty much every person in the organization touches. So everyone has had an issue with their computer which we needed assistance, we always went to the sysadmin and the self service portal. So just by the fact that in an organization a service desk is one tool that almost all employees touch. So there you can see a bigger impact because just from the fact that it has a larger footprint in an organization. And of course the next area I would say is maybe observability because there we’ve had the age old problem of too many events, too many alerts triggering and people have always struggled with trying to figure out what is the actual issue that I need to prioritize. Right. So that’s another area where AI is doing a lot of prioritization for our customers and seeing a big impact. Yeah.

Swapnil Bhartiya: And since you mentioned, of course, Service Desk, if I’m not wrong, Service Desk has been using Zia for a while including of course pre built agents like you know, the whole L1 service desk specialist. How does that compare with your competitors like ServiceNow?

Rakesh Jayaprakash: There is a lot of things that we’re doing is also something like ServiceNow does as well. And I think to an extent AI is even like leveling the playing field because now everyone has access to like the same set of technology that is powering the backend. So you have to really differentiate what you’re doing. And Manage Engine has traditionally stood for value in terms of the value that you can get as a customer from using our products. So our guiding principle is not like throw AI at everything just because we can, but rather focus on what is actually helping the customer, the real impact in terms of helping them reduce the amount of time that they spend doing the same manual thing again and again.

Swapnil Bhartiya: And how does the rest of the stack and the role of Zia looks like on top of just, you know, Service Desk plus.

Rakesh Jayaprakash: So the way Zia actually is developed in house is that Zia is an umbrella initiative where, you know, our labs team will develop something like sentiment analysis or ticket prioritization or data cataloging. And this is consumed by all our applications, the various 60 different products that we have in terms of what is necessary for that particular domain. So for example, sentiment analysis would be useful in ITSM to go through conversations and figure out if the particular customer is frustrated or if they are happy. But sentiment analysis might not have any role to play in observability. So Zia is an umbrella stack which has a lot of these AI features that are being developed and they are absorbed by the individual product portfolios depending on what is necessary for that particular domain.

Swapnil Bhartiya: And if I’m not wrong, and I was talking to Rajesh also before that, that ZIA agents ship at no additional cost. You don’t charge for that. While a lot of vendors, they kind of try to extract as much value as they can. How does that pricing decision shape which feature get built? And also it also, as you earlier mentioned, even playing field, democratize it and also make it more affordable.

Rakesh Jayaprakash: Yeah, see what I mean? Rajesh might have mentioned this. Also we like AI to be that invisible layer that is powering all of our applications. Now there is absolute value in using AI in terms of helping our customers get the most from the platform. So we don’t believe that it should be charged. So and the fact that ManageEngine is a company which builds our own stack from scratch, I mean most organizations don’t have a 60 product portfolio which was built from the ground up. So the strategy has been always to do our own R and D and LLM. We are also doing our own R and D. We are hosting it on our own data centers. So we are not paying a huge bill to your OpenAIs and Anthropics of the world running in our own data center. And it also helps make our LLMs better over a period of time so that we are on par with what you see outside. So it might look like it’s a step backward in terms of using a smaller model, but smaller models are doing absolutely great, especially when it comes to working with specific domain specific data such as IT. And that’s the thing. So we want to help our customers get most value out of the products which they’re already paying for. We don’t want to nickel and dime them for the token usage because at the end of the day they’re using our LLMs within our own products for which they already have a subscription for. And we don’t believe in having to like extract more value for them to just use these features in our own products. So we see AI as a feature. We don’t charge a customer based on a feature ever. So I think that’s more or less the logic behind why we are taking this kind of an approach.

Swapnil Bhartiya: And of course your agents support MCP. Today everybody supports MCP. So people can bring their own tool, but while you have Zia, but do you also let customers bring their own LLMs? And because how important is that openness when customers are. Because sometimes one model doesn’t fit all the needs, you know. So can you talk about this openness there?

Rakesh Jayaprakash: Yeah. So even today you can. By default the Zia platform will work using our own LLM, but customers have the option to switch to an OpenAI model. So you can bring your own key, plug that in. So it will work within the OpenAI and Claude and the other LLMs of the world. So you can bring in your own key. It’ll have the same functionality, but it will work on your model. Now more interestingly, we are trying to do the same thing in our on premises applications also. So we are seeing customers host their own LLM, especially in like highly regulated industries like banking or healthcare where they’re dealing with a lot of sensitive data. Now even our on premises applications will support plugging into on premises LLMs. So because customers need that flexibility, you cannot have them go to the cloud every single time, especially when you’re dealing with a lot of sensitive data, that’s just not an option. So we don’t want customers to be tied into one particular model and not use all of the functionality that is available at their disposal. And just because of the deployment model they choose, they’re stuck with one. So we don’t want to ever do that.

Swapnil Bhartiya: We’re almost nearing the end of this year and customers are already kind of planning their next year. They also want to be fully aware of what does your roadmap look like? Of course, roadmaps also change depending on the market. But if you can kind of talk a quick little, what does your roadmap look like? So it can also reflect on your AI strategy and what momentum you are seeing and then what customers should be prepared for. Because, you know, we are serving the customers, right?

Rakesh Jayaprakash: Yeah, yeah, for sure. So especially this year, a lot of our initial plan for the roadmap had to be accommodative of a lot of the AI functionality and the developments that have been coming in. So that’s been an interesting journey in terms of balancing whether we should build a feature or focus on the AI side of things. But I would like to say that we’ve done a decent job of balancing the both. And roadmap is something that obviously a lot of our customers are interested in. Like day one, right after the main keynote and the one or two sessions, what is going on? I mean, the sessions that are going on right now are roadmap sessions. So we’ll ensure, I mean, our focus always been not to be carried away by AI, ensure that we build the functionality and the modules that are necessary for the individual product portfolio. However, use AI when we feel that yes, it does do a complementing job and actually make a difference.

Swapnil Bhartiya: Now, this is going to be an interesting question. What is one or more than one AI capabilities that customers keep asking, but you believe that this is not yet ready for production or can be prioritized, but customers keep asking for that.

Rakesh Jayaprakash: I’m trying to think so. Okay, so I wouldn’t pick one specific feature, but rather what customers want to do. I’m hearing a lot is run AI on top of unstructured data, which requires a lot of processing power. Right now, the side effect of it’s not that we can’t enable that, but what happens is when you are running AI on top of a lot of unstructured data without having a very specific focus, then it takes a lot of time for that processing to happen, which again, negatively impacts the user experience. So that is one area that we are definitely looking into. But we are taking it slow in terms of pushing AI on top of unstructured data at this point in time. Because when the data underneath is clean and categorized and the reliability layer that Rajesh was talking about earlier is set in properly, then putting in AI on top of it gets your answers faster and also you get meaningful results. Whereas you have like a dump of data that you want, that you have, then you want to put AI on top of it. More often than not, it doesn’t really give you the results that you expect.

Swapnil Bhartiya: Now let’s talk about privacy a bit. Managing them. Zoho. Like a lot of companies that say that you don’t use customer data to train your model. Now most companies use a lot of synthetic data, but how much is it driven by customer demand versus your own policy at the same time? Just the way if you look at general purpose AI, you know, it learns a lot from user data. So because there’s a lot of valuable telemetry, you know, so what is happening is, and sometimes no offenses, people say that, you know, AI driven customer support or anything else is not great because it’s trained on synthetic data. And no matter what, how much it is still synthetic, it’s not organic data. You walk into a drive through McDonald’s, you order something, at the last moment you say, you know what, I don’t want a large fry, I want a small fry. And please change from a juice to a coffee. A human will change that. But AI is not used to it, so it gets confused. So talk about the pros and cons and what would you like to see in the ideal world? At the same time, how do you look at, you know that, how things can be improved?

Rakesh Jayaprakash: What I would say is, you know, of course AI gets better as it has a better understanding of your own infrastructure. Now the way I see it is all the ManageEngine products, they operate across different industries. So we don’t build products for one specific industry. What we do is we give you like a blank slate or kind of like a clay that you can mold it into your own industry and add your own workflows by configuring specifically what you want. Now I also kind of see AI like that, wherein we give you a blanket AI which is capable of executing the things that you want to. But of course it’s not trained on your data when you’re starting to use it. But as you start using it more and more and as it understands your processes and it’s within your own account, that addresses the privacy bit, because it’s not like you’re training the larger model, rather you are using the AI functionality around the data that is coming into these applications. So that way it over time it gets better. So what I would say is, yes, when you are taking a model that is not highly optimized for you, some of the answers that it might throw initially might be very broad and not very specific. But over time it gets better. And I think that is the approach that we would probably take. Though it might initially look like you’re not getting a lot of value, eventually you do as and when more data comes in and you start using the products more.

Swapnil Bhartiya: The interesting thing with AI there are two or three things. First of all, a lot of folks think it’s a silver bullet to all their problems. Second is they think of it as a magic pixel. And third is that bigger is better. So these are all myths. What is your advice to customers? That sometimes their expectations are also realistic. At the same time they are not burning all of their budget tokens because actually AI is consuming all the budget. You don’t have money left to write the software, so which is actually not sustainable. So what is your advice to customers? So they maintain the balance they get most out of AI while investing the least in AI.

Rakesh Jayaprakash: Now what my advice would be is don’t start with the question in terms of asking what AI can do for me. Rather ask a question, what are my top problems which I want to look at AI to solve. So each organization has specifically specific use cases, like their top two or three problems that they want to address and see if AI can make a meaningful impact there. So that way, one, you’re scoping yourself to something really important and you’re not spending your budget in two days. So start with those use cases that are really important, try to address them. Take one thing at a time, do a pilot, see what are the outcomes, and then put it into production and keep improving on that. So that would be my advice. Don’t try to go too fast at the same time, that doesn’t mean you have to lag behind. Everyone at this point is still at the stage of experimenting. It might look like we are missing out on a lot. But if you are at least at a mindset where you are open to exploring, I think that’s all you need at this point in time. And do small pilots, pick up very specific use cases. I think those are the two main things that I would focus on.

Swapnil Bhartiya: And also, what is your opinion on flipping the whole script? Instead of boasting about how many tokens you consume, it should be more about ROI, what return I got from the investment at the end of the year. Because you need to evaluate what you can get done with AI versus, I mean, AI is critical. You know, it has become as essential as software or cloud. So that cat is not going back in the bag. But it should be focused, should be more on ROI, what you’re getting out of AI versus what you’re investing. And stop boasting, it should be opposite that, hey, I can go one mile in one gallon versus I can go 200 miles in one drop of gasoline. So how would you frame that? Should that be the mindset?

Rakesh Jayaprakash: I mean, I’ll give you an example. So one of the areas where we see AI can do a lot is when it comes to data analysis. You know, I have a lot of data now I’m going to put AI on top of it and I’m going to ask a question and it’s going to do a lot of analysis in a fraction of a few minutes and it’s going to give me an answer. So let’s say to get that particular answer, I spend about 100 tokens. You got the answer. Now my colleague goes and asks the same question, it does the same analysis, gives him the same answer again, it consumed another 100 tokens. So now 100 people do the same task and you end up with a really massive bill. Whereas when you have a governed analytics layer, a governed data platform, where you have all of your metrics defined, all of your definitions configured, and you have a proper metrics catalog, that operation runs just once and you have the answer. So on one side you might feel like you’re getting answers very quickly, but you’re also doing it not very economically. You know, it’s like we see people on Instagram getting out of Ferraris and Bugattis, but you don’t really know if they can actually afford it. So that’s kind of what’s really happening with the whole AI story. So I would say don’t get carried away. Focus on the fundamentals. In the long run, that is what is going to really make a difference.

Swapnil Bhartiya: Rakesh, thank you so much for taking time out. And I mean, I think the approach that you have suggested is very practical. Approach and also what goes into product influenced by customers and where, how they should use AI. Thank you for all those great insights and I would love to have you back on the show. Thank you.

Rakesh Jayaprakash: Absolutely. It was a pleasure talking to you. Thank you so much.