Moving Enterprise AI From Hype to Accountable Results With Freshworks

[00:00:00] Do you struggle to keep your AI agents from acting outside of compliance? Well, Denodo provides an AI data layer connecting your data systems, keeping guardrails consistent across all of your data platforms. So start scaling your business carefree with Denodo. And you can do that by visiting denodo.com to learn more. But now, on with today's show.

[00:00:27] Has your AI investment moved beyond impressive demos and successfully achieved measurable return on investment? Well, enterprise buyers are now asking about reliability, governance, total cost, traceability, and whether autonomous service operations actually resolve problems from beginning to end. Well, my guest today is the CTO at Freshworks.

[00:00:55] He has over 30 years of experience building enterprise software. And today, he's going to explain why the AI hype cycle is giving away to an accountability cycle. A cycle where every investment must prove its business value. And also discuss why mid-market companies provide a demanding test for enterprise AI. And we'll also explore how configuration-led architecture actually supports faster deployment.

[00:01:23] And what autonomous operations really look like in practice. And we'll even throw in there data sovereignty, cloud infrastructure, and the effect of coding agents on engineering judgment. Enough from me. Let me introduce you to my guest now. So, thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do? Thank you, Neil, for having me as a guest. My name is Murali Swaminathan.

[00:01:51] I am a CTO at Freshworks, leading engineering and architecture teams here. We have 30 plus years of experience in industry. I started as an engineer. I grew up the ladder, working for many companies, building enterprise software at definitely an enterprise scale. At companies like ServiceNow, CA, and many other small and large companies. And I'm fortunate to be at Freshworks as part of the broad journey here.

[00:02:16] As you know, Freshworks is an AI-powered unified service operations platform that's faster deploy, intuitive use, and enables every employee to be more productive. And further to the cap, we just got recognized as Gartner's magic quadrant leader in IT service management platforms. So, that kind of demonstrates our product and the capabilities that we have and the testimony that we have from our customers. Thank you so much for sitting down with me today.

[00:02:45] I've got so much I want to talk about because we've seen so many big changes over the last three to four years. And at the moment, though, it almost feels like we're entering a new phase of enterprise AI. The conversation has shifted from can we use AI, can we build AI, to can we prove it delivers value? And also, in some cases, can we trust it? So, of course, from everything you're seeing and hearing, from all the conversations that you're having,

[00:03:10] what are business leaders asking today that maybe they weren't asking 12 months ago? It is definitely, I think, what you're seeing is absolutely right. AI is no longer a novelty. Everybody is no longer seeing it as, oh, I'm going to just use AI. They're saying, how can I make use of AI? And how can I get a return on investment, ROI, out of it? So, that is where the conversation is going. Can I trust the AI? Can I, whatever is being done, can it, does it, does it really work end to end?

[00:03:39] Is it worth the investment I'm making from a cost perspective, from an infrastructure perspective, from anything that you're building with AI? Right. People want to know what's the ROI and the total cost of ownership. And this is not something we heard 12 months back, it was all about everybody was experimenting with AI. But now it's all about, okay, I want to know how it's working. I want to know it's no longer a black box. They want to know why it did what it did.

[00:04:04] That means they really want traceability and observability of why AI made a decision. And they want people to audit it. They want to be able to override it. They want to be able to gather the feedback and make it better. And before you join me on the show today, I was doing a little research on you. One of the things that stood out to me was that you said that the hype cycle is over and the accountability cycle has begun. And love that line.

[00:04:30] So, when organizations evaluate their AI investments today, though, what metrics actually matter and how should they measure success beyond productivity claims? And the reason I ask that is, for as long as I can remember in the world of IT, the mantra has always been, you can only improve what you measure. But there seems to be some kind of confusion or unsure of what we should be measuring here. Of course. Yeah, it's very hard to manage something that you cannot measure.

[00:04:58] And AI is maturing like any other cool technology that we've seen over the years, right? So, now everybody's trying to figure out, okay, what am I getting out of making my investment in AI? Am I getting a productivity investment through AI? Am I getting to do more things for less? Or am I going to do more things with fewer people? What kind of operational improvements that I'm seeing with leveraging AI? Is it just replacing what I did before with AI? Or am I actually reimagining my operational processes with AI?

[00:05:26] So, that's what we are seeing some of it. So, those are the questions that we see our customers and our leaders and CIOs asking that question. So, some of the metrics that they're all looking at is like, okay, is it helping me reduce the time to resolve? In this case, it's because it's not AI is helping me give me a suggestion, but it's not just solution. It's helping me take it to the end line and then helping me resolve the issue that I started working with. And the other one is, did I involve the human in the loop? Or it was automatically done through AI?

[00:05:56] And how much of the repetitive work that a human or an employee was doing has been automated through AI? So, these are all some of those things that are top of mind. The other thing that is important that companies are looking forward to is, are looking at is not just SLAs, the service level agreements. Now, they're looking at something called experience level agreements, XLAs. We introduced a feature called XLAs. It's not just that you resolve the issue in time. Did you actually improve the quality of support for the employees?

[00:06:26] That means that did they really make it productive for them? That it just took them end to end. It's not just, oh, I'd resolve the ticket in two minutes. It actually fixed the problem for the man. So, those are questions that our customers ask. And Freshworks has traditionally been seen as a very strong solution for mid-market organisations. And again, when I was doing a little research on you, I was reading how you believe mid-market is becoming the real proving ground for enterprise AI.

[00:06:55] So, again, from everything you're seeing and hearing here, what lessons do you think larger enterprises can learn from how these businesses are really approaching adoption and enjoying such success too? I think mid-market enterprise, we consider them as agile enterprises because for them, speed is the matter, right? So, they want to get it done fast with a reasonable cost. They don't have the luxury of time to spend a lot of time doing experimentation and spending a lot of time in R&D cycles. They want anything they invest in.

[00:07:24] They want immediate kind of gratification of the investment. That means that whatever they're deploying has to work. So, that's why the AI adoption is important for them. They want to be able to do some money doing some of those things. So, that's the crux of what mid-market companies we are seeing. So, that means they are making sure that they are trying out something, making sure it's working. If it's repeatable, then they put it in operation and they go on to the next thing, right?

[00:07:51] So, that's how mid-market companies is not about deploying everything at scale on day one, but taking it, breaking it up into parts, looking at which ones need the most AI kind of power, looking at the operational processes that are powering those things, refining them, redefining them if they have to, and then applying AI to get to do that work, measuring it. If it's repeatable and it's working fine, then take that and then make it work for the next set of business processes.

[00:08:20] So, that's what I think large enterprises can also learn of how you can do it in smaller pieces and how you can then scale it up after you have experimented. After you have proven it in smaller groups and how you can scale it to the larger groups. Another point that particularly stood out to me was your belief that ease is actually an architectural decision, not simply a user experience decision. And as I say that line out loud, I can hear techies rejoicing around the world saying, at last, someone gets it.

[00:08:49] But can you expand on why this matters and why platforms built for configuration rather than just customization have much more of a long-term advantage? Yeah, this is something that's our primary ethos of Freshworks. Intuitive to use, easy to deploy, easy to manage, easy to configure. Because it's not about just putting a nice UI on a complex system and calling it simple.

[00:09:15] We want to be able to think about configuration and not as a customization. Because when you build a configuration-versed kind of approaches, that means you can now upgrade it pretty quickly. But some of the things that are challenges of doing a complex system with customizations is that you have, you made it work for you through enough customizations, but you're stuck with a certain release and you cannot upgrade anymore. Right? That means you're not able to run the greatest and latest releases.

[00:09:44] That means you're not taking advantage of newer features. That means it becomes a drag on your company's productivity. And as a man, the providers are using new capabilities, they are not able to keep up. And that becomes a challenge for them. And so that's why ease of use is a big part of it. And configuration-based, not customization-based approach is what we believe in. And we want to make sure that same principles apply to AI as well.

[00:10:11] We want to make sure that when your AI is deployed, it works for you with configuration and not you don't have to customize the AI to make it work for you. And that's what we're seeing with some of our customers, how they are developing our AI capabilities. So as AI inevitably becomes embedded into service operations and workflow operations, and I'm curious, what does an autonomous operation actually look like?

[00:10:37] And to help answer the return on investment question as well, where are you seeing organizations already seeing AI taking ownership of routine work and also keeping human focused on higher value decisions? It's always a big balancing act, but what are you seeing here and how are they getting that balance right? Because it's a massive talking point right now. You're right. You're right. I think the autonomous word is definitely people have different understandings of the word autonomous.

[00:11:05] Autonomous doesn't mean I'm putting a robot on T-team and the machine is taking care of everything and the humans are focusing on other tasks, complex tasks. It means that, okay, so what we mean by autonomous is there are certain stuff that can be automated to the point where the repetitive work is what we're seeing, that you trust and you know it's working, can be automated so that the employees can focus on problems that matter.

[00:11:32] So a simple use case, I would say is if somebody comes in, a ticket comes in to say a printer is down. Today, an agent, human agent has to read it, understand what it means and say, oh, it's a printer ticket. Now I need to understand it's an IT issue, it's a facilities issue, it's a physical location issue. Then they need to manually triage it and then route it to different people and then figure out if it works or not.

[00:11:57] But in the case of AI, if you know it's a repeatable issue, AI reads the ticket, classifies it, routes it to the right team, opens the ticket facilities before even somebody needs to act on it. That way it routes it to the right person and if it's something that can be fixed automatically, it fixes for you. If it's not, it gets it to the right person so that it gets done the right way.

[00:12:19] It's all about how do you want to experiment with this kind of approaches where take a small problem, make sure that you're trying to look at it and end-to-end automating it wherever you can. And I think all too often many leaders only see the shiny UI, the seamless UX as well. But if we were to take a look under the hood, AI is also placing new demands on cloud infrastructure and data centers.

[00:12:46] So from your perspective, what infrastructure foundations need to be in place before organizations can confidently scale AI across their business? Because this is something we don't talk about enough, the actual infrastructure that makes everything that we all love and enjoy and creates the opportunities. But it's critical, isn't it? It is very critical. I think going back to our original point of how do I make sure that AI works at enterprise scale? Yeah.

[00:13:12] And if you wanted to work at enterprise scale, then you need to respect the norms of how the enterprise works. That means the enterprise is working across multiple geos. That means you need to make sure that when they are using the product, they are working with the right geos. But there are data sovereignty rules between the United States, Europe, or is it Australia or any other region. So you need to make sure that you respect the data residency rules while you are working with AI.

[00:13:39] Like similar to anything else you do with your enterprise product. Right? So you want to make sure that you're not violating privacy laws. You need to make sure that the separation of data happens. And then whenever you are diagnosing, the observability also is looking at the norms that are in that particular region. So that means that as part of making sure AI works for you at enterprise scale, you do have to set up guardrails, routing rules, and things.

[00:14:06] That will make sure that the requests are being routed to the right location because we are kind of applying all the local data sovereignty rules, not storing data or encrypting data. Depending on what is required, the infrastructure has to support those data sovereignty needs.

[00:14:27] And I was also reading that you've raised an interesting concern and crucial point that AI makes execution easier, but it can also make deep intuition harder to develop, particularly for engineers early in their career. And then there's the other side of the coin where if AI was to take away some of those entry-level roles, where do the senior engineers of tomorrow come from?

[00:14:50] So on this point, which again is a critical point, how should organizations ensure that they are still developing expertise, critical thinking, and accountability alongside that AI-assisted development? Again, massive talking point at the moment. That is true.

[00:15:05] I think just because using AI to generate code doesn't mean that you don't follow your regular software development practices of writing clean code, reviewing it, getting it tested, getting it deployed, and testing it with customer use cases in mind, and then rolling it out to customers. None of those rules for shipping software changes. Just that, yes, you're providing generating code faster.

[00:15:29] And it's important for engineers, whether you are starting your career or in the middle of a career or you are a senior person, to understand what's being built. So writing the code has become the easiest thing right now. But we are seeing engineers spend a lot more time in the planning and design stage, understanding what needs to be built. Are we designing it the right way? Are we designing it with scale in mind? Are we designing it with usability in mind?

[00:15:57] So those are questions that are being asked. And that's the level of conversations going on. The writing the code has become, oh, anybody can write code right now because you can use a coding agent to generate the code for you. So how you write the code, how you think about the architecture principles, how you think about designing for scale is all the things that we are expecting engineers to look at. And then you also have to make sure that you have review in place.

[00:16:22] Right. So we have since we are generating a lot of code, not all code can be humanly reviewed. So you have to set up certain rules and thresholds that we have used AI coding tools. And we also have human in the loop for certain things, for things that are, we have made sure that, okay, we are expecting developers to follow a certain pattern. And of code and writing code, we have the rules in place and AI reviewer can take care of it.

[00:16:46] For certain things, we expect it to fall back to a human reviewer who helps review the code that has already been reviewed by AI to make sure that it does not spend anything and then push it along the line. And then as you build confidence, more and more stuff is being delegated back to AI for reviewing. But we're continuing to keep human in the loop because there has to be a checks and balances to make sure that you're not introducing something that should not have been included in the first place.

[00:17:15] Well, it's time to have a bit of fun with you now. I'm now going to pull out my virtual crystal ball and ask you to look ahead and envision a future. If we were having this conversation in what, let's say, three to five years, which I know is impossible considering the scale of change we've seen in the last three to five years. But what do you think will distinguish the organizations that did just simply deploy AI tools from those that genuinely began transforming the way they operate?

[00:17:42] Different workflows, different ways of working, different mindset. And what practical advice would you leave listeners that are trying to build that foundation today? Yeah, thank you for asking that question, because I think people just jump in to solve the problem without understanding what are they trying to solve. So we want them to start with the problems and not with the technology, right? So they don't want to use a square bug and round hole. So they want to use the right technology for the right problem.

[00:18:09] So I would say, and then other thing that's key is you can't manage that you can't measure, right? So it's important for them to first measure, to say, how is it working right now? Understand the current way of working and see, okay, if I'm applying with AI, what are they expecting? That means define some ROI metrics to say, okay, this is what I'm getting with my current processes and this is what I want to get with AI. So building on those data quality is very important.

[00:18:37] Looking at, do I have ability to trace back what's happening with AI? Who's doing what when? Because without that auditable embrace, you're going to rely on something and if it makes mistakes, you don't know who to go back to. So that's important for us to have that auditable embrace. And then it's important for the teams to be self-standing. We don't want the teams to be relying on other supporting teams like you or anybody else to get some things done.

[00:19:04] So you want to empower the teams to enable themselves to build these kind of tools, AI-based products themselves. So at the end of the day, the goal is you want to be able to see, are you able to automate repetitive work? And then if the repetitive work that you're able to automate is trustable, that means it's working 100% of the times, then you go on and see, okay, now I'll make it work for the next set of processes. Lovely.

[00:19:30] And so much to take away and think about there from scalable data centers, cloud infrastructure and workflow automation to truly achieve autonomous operations. And we've only scratched the surface, really. So anybody listening that would like to find out more about you, Freshworks, connect with you or your team or just learn more about anything we talked about, where would you like me to point everyone? Thank you for, again, giving me an opportunity here.

[00:19:57] So you can always find us on freshworks.com, which is where we publish all our information about our products. And then if you're interested in finding out how we build, I would recommend our listeners to look at our Freshworks engineering blogs on Medium, where we share deep dives on AI, architecture, infrastructure and operating at scale. If you're a developer, we also want developers to check out developer.freshworks.com. And if you want to look me up, you can look me up on LinkedIn.

[00:20:25] There's a lot of stuff that I do that is publicly available. So looking forward to hearing feedback from the listeners. Awesome. Well, I think the message is pretty clear today. AI has moved from exploration to execution. And that execution bar is higher than people expected. Reliability, governance, total cost and measurable ROI. These are all questions that buyers are asking right now. Hype cycle is over. The accountability cycle has indeed started.

[00:20:53] And I will have links to everything that you mentioned there. And no matter which part you play in an enterprise, whether you're a techie or a leader, please check out the links that you'll find in the show notes and find out more there. And also feedback to me as well. I'd love to hear from you. But more than anything, thank you for starting this today. Really appreciate your time. Thank you, Neil. Appreciate me bringing on the show.

[00:21:15] I think our guest left us with a practical framework today for moving enterprise AI from experimentation into accountable execution. Yeah, begin with a business problem. Record how the process performs right now. Draw that line in the sand. Define what that ROI is. What you expect. And then test AI on that very same repeatable workflow before expanding it.

[00:21:44] And autonomous service operations, they do need observability, auditability, data residency controls, and a clear point where the human review enters that process. Because yes, speed is valuable. But a fast answer that cannot be explained or indeed trusted, that creates a whole heap of other problems. So I think his warning for engineering leaders deserves equal attention. Yes, coding agents make execution easier.

[00:22:12] But architecture, design, review, and accountability, they all firmly remain human responsibilities. And over to you. Has your organization entered the AI accountability cycle yet? Or are you still measuring activity while hoping business value will eventually appear? Hopefully it's the former, not the latter. But techtalksnetwork.com if you need me for anything at all. And I will be back again real soon with another guest. Well, thanks for listening today. Bye for now.