How AI will change enterprise software with Akshaya Murthy

As the business world races toward AI-native solutions, enterprises are reevaluating their reliance on traditional, monolithic systems. Klarna’s decision to pivot from Salesforce to develop its own AI-first platform underscores a broader trend reshaping the technology landscape. But transitioning to AI-native systems is no small feat—legacy companies often face bureaucratic inertia, fragmented processes, and the challenge of retraining their workforce.

In this episode of Simulated Reality, we are joined by Akshaya Murthy, a seasoned technology and business leader based in the San Francisco Bay Area. Currently serving as the Director of AI Transformation at Zendesk, Akshaya spearheads the company’s global non-product AI strategies, focusing on leveraging AI to drive measurable business impact. With a deep commitment to responsible AI practices, Akshaya advocates for ethical and strategic implementation, ensuring AI serves as an enabler for progress rather than disruption.

With a career spanning software engineering, quality assurance, consulting, and technology risk management, Akshaya brings a rare blend of technical acumen and strategic insight to the table. A graduate in Electronics and Communication Engineering from India and an MBA in Strategy from the University of Pittsburgh, his expertise equips him to tackle the most pressing challenges in AI transformation.

Key Highlights

Transition to AI-First Approaches: The discussion emphasizes the inevitable shift for large enterprises towards AI-native or AI-first approaches, highlighting the necessity and challenges of this transition.

Executive Decisions and Consolidation: Insights into executive-level decisions, such as consolidation efforts, and the move towards AI-first methodologies in companies like Zendesk.

Challenges in Legacy Systems: Exploration of the hurdles legacy companies face, including bureaucracy, process documentation gaps, and the need for reskilling the workforce.

Build vs. Buy Strategy: Examination of the strategic decisions around building versus buying technology solutions, with a focus on the risks and benefits of each approach.

Vendor Selection and AI Operating Systems: Guidance on selecting the right vendors, the potential shift towards AI operating systems, and the concept of vendor lock-in.

Kashyap: Welcome everyone to the next episode of AIM Media House podcast Simulated Reality. Today we have with us Data and AI leader Akshaya Murthy. How are you doing today?

Akshaya: Wonderful to be on the podcast. Thanks for having me.

Kashyap: As I mentioned in our previous call, this topic is critical and vital, especially in light of Klarna moving away from Salesforce to build something AI-first. What are some of the conversations you, as a leader, are having with your clients? Was there any follow-up dialogue after the big Klarna announcement?

Akshaya: Not yet. But I think it’s good timing to talk about the whole AI native landscape and how it’s going to impact enterprise technology as a whole. If you take a step back and look at what is essentially AI native, it is any technology that relies on live data pipes, some sort of intelligence that can make less brittle decisions compared to previous automation efforts, and get your solution 80% of the way there in a matter of seconds rather than hours or days. That’s if we define that as the scope of an AI native solution. However it’s powered, LLMs, whatever it is, doesn’t matter. The current application stack in the enterprise space is terrible to say the least because it requires a lot of people to configure, manage, maintain and update.

Every small thing means you’re either building a new piece of code or you’re throwing people at the problem or you’re engineering the project or the process to make it work. Now in the AI native realm, what happens is there is end-to-end automation. What does it mean? Let me give you an example. You take digital marketing. It begins with some sort of a concept of, “I want to market my product,” and this is my concept of who it should be marketed to, what messaging I send out. Then they create the campaign design, and it usually gets to an agency to design the visuals, video, audio, whatever and then it gets vetted by humans.Then you have ad campaigns created. They get measured and then monetized. So you have here multiple human stakeholders, both internal and external. 

Take a look at the AI native realm. What can I do? I can create hundreds of concepts of what I want to market to be very, very precise to exact the persona or even the individual. So for example, say you like dogs. I can make sure my marketing campaign always has dogs in the stuff that I send out to you. So you can get really creative. I don’t have to go to an agency. I can go from concept to design to actual end product in 5 seconds using Sora or Adobe Firefly or whatever one of those tools are and I can create a brand campaign based on live data on big data I have captured from data exchanges and our own data gathering efforts as a marketing company. Automatically convert you into a user by giving you very personalized discounts based on your income level, your zip code, and keep you for life. So you see the shift. It’s not just one or two enterprise applications that got disrupted. It’s the entire value chain. That’s the shift in mindset. 

So I think Klarna is doing the right thing by saying these big ERP, CRM, HRM systems are very clunky, monolithic, and require a lot of people to manage and maintain. We are not that kind of a company. Our operating model does not align with this. So let’s build something that aligns with our operating model, which in turn aligns with our business strategy and our product strategy. So that’s the context you take into account. And my prediction is most enterprises won’t make that leap. They’re too conservative.

Kashyap: One question that arises is about the tech industry’s focus on learning and unlearning. Workday and Salesforce have scaled significantly because there was a market need, and these products are not cheap. Why did we reach a stage where companies kept adding more applications to cater to similar needs? For example, a slight difference in one process or a company doing something differently inspired whole new products instead of just add-ons.

Can you walk me through this journey a little bit? The world of enterprise software and the thought process behind its design is somewhat of a black box to the outside world.

Akshaya: The way IT infrastructure applications and ecosystems grow and balloons, for example, at Zendesk we use 40 applications to run finance. The reason why it ballooned to that is because if you want a consolidated application like an ERP, it’s really, really expensive and it’s not available to every company at every growth stage. So what happens is say a company starts off and they’re growing at 40% per year. They realize they need a billing system. They realize they need a GL system. They add those because they have to continue growing and say that is at a  10 million revenue level and then they lose focus.They go to 100 million. Now they haven’t changed the system to adapt to that particular mindset and growth rate and that level of criticality and it sort of calcifies.

Now a CIO comes along at a 500 million stage and says,”This is terrible. Let′s rip this out. Let′s put an ERP in place,”and they look at the price tag saying 500 million dollars per year  “No, we’re not going to do it. We are going to stick to this. Let’s upgrade this platform and augment this with something else where we are missing features.” 

That’s how it evolves over time and that calcification leads to a jumbled set of applications and a really bloated IT team. So that’s what has gotten us to this. And it’s always happened. From a SaaS perspective the promise was don’t have Onprem applications it’s going to reduce your cost. So SaaS, it is every Tom, Dick, and Harry added a SaaS application. Anything new that could be done was taken into consideration and added to the stack. So there was not much consideration. It’s also about the maturity of the organization at that level.

Usually, when the maturity comes to the billion and a half dollar stage of revenue, then they start thinking about how we need to consolidate this into an ERP because we have now 40 different data models that we have to make work with each other. So there’s a bunch of interface development that goes into it. So let’s get rid of all that. Let’s transform going to an ERP system, one data model, one workflow, all in one, and at that stage again, it creates its own issues because an ERP is not easy to maintain. You have to overhaul people and infrastructure, etc. It’s very expensive and time-consuming, and also you shouldn’t forget the people. 

When transformations happen, people are not going to stick around because they know their jobs are at stake. That’s why transformations fail because people with legacy knowledge of how the products work leave because they know they don’t have a play 2 years down the line. So all these factors lead to a jumbled mess, and I think AI native can save that. Let’s hope that’s the promise, at least.

Kashyap: You mentioned that as a company grows, it needs to set processes and hire ERP tools, but when it grows too much, it can become bureaucratic. Is that a correct simplification? Technology, in general, has always played a leveling role, not just in terms of affordability. This applies to enterprise sales and any field, even sports. I follow sports closely, and I believe that the democratization of technology can make an even playing field for teams that cannot afford really big players. Similarly, in a business context, if you are trying to grow and hire the services of ERP tools, technology will always try to level the playing field. Is that correct?

Akshaya: AI has been around for a very long time. Companies have applied it a lot, and many use data science and machine learning, etc. So that was a domain of operation just two years ago. What generative AI made possible is for anyone to tap into that power pretty quickly and very easily without much training. It was very intuitive. That’s why ChatGPT took off the way it did, becoming the fastest adopted technology in humankind’s history. If you look at what AI brings to the table, it’s not just about incremental changes; it’s paradigm changes. It’s quantum changes that you can make with your product, with your offering, with your service, etc. For example, there’s this company, I may be getting it wrong, Viscom for industrial design. You can go from sketch to concept in 20 seconds. Try that with legacy products.

Given that kind of power and power disparity between legacy versus AI, what enterprises quickly realized they had to put AI in all their applications. So let’s call that a business model change, where we were selling seats. Now we won’t sell seats. We’ll probably sell per concept you create for the value we have added to your ecosystem, and you pay us. That’s a business model change, let’s call it value creation. What is the value you’re providing to your customers? Now that necessitates how we capture value. The shift was from per seat to per outcome. So what that meant is on the back end, your operating model has changed because now you can scale your products infinitely potentially. You can add new features on a daily basis because you can prototype and do this really easily. And if you’re going for scale, what happens is the scope of your operations changes. The way you recognize revenue changes, the way you do your operations changes, the way you handle customer support changes because it’s a completely new set of ways of doing things. And AI applications quite necessarily give back a lot of data in terms of insights, in terms of usage, etc.

So you have to transform your organizations to learn from that data. So this is the context. There’s a business model change and an operating model change, and in a very big way. If you take ERPs into consideration, if you take HRM systems, they cannot handle this in the state it is today, forget 5 years from now, who knows what it’s going to look like. Legacy players, in a hurry to get into the AI game, have added features like summarization here and resume review there, which is not a business model shift. So customers, enterprises who consume applications like cloud, are realizing this will not set us up for success. We will be behind 2, 3, 5 years if we do this, so we need to go the AI native route.

So that’s the shift in mindset. Again, not all CIOs are thinking that far ahead. These are really hard to pull off because these are complete strategies. In the intermediary term, what they’re doing is adding an intelligent automated integrated AI layer on top of everything. For example, enterprise search, bringing all data together, putting all their data lakes like Snowflake or Databricks, that will allow for all your people to become analysts and look at the data in a new light and analyze and make better product and operations decisions.

So these are the two paradigms: you could do AI native, which is very hard and needs the board and the C suite to say we are sticking with this, this is a bet we are taking, all in, no questions asked, we live or die by AI. The second set is saying we can’t live or die by AI; we have legacy customers who rely on us to run their business, so let’s see what we can do. It’s very similar to electric vehicles versus ICE vehicle manufacturers and how they have to maintain both petrol, diesel, and electric vehicles, whereas electric vehicles just go one way. But eventually, I think there’ll be a massive consolidation of legacy players because they won’t be able to compete. And those who do come out the other side will be completely different companies compared to what they are today.

Kashyap: You mentioned several thoughts, and I have many in my head. My first question to you is about legacy companies like Salesforce and Workday. They have survived the test of time and scaled significantly, so they are not blindsided by the current changes. Where do you see these businesses moving in today’s world? What are some of the executive decisions you anticipate being taken at their level? For instance, you mentioned consolidation efforts at Zendesk. Do you see them adopting a completely AI-first approach, building work processes around it, or something else?

Akshaya: I think the really smart players are looking at rearchitecting their solutions, starting from a blank sheet of paper and building the solution brand new. That’s what they’re looking at. Oracle is doing that with their end-to-end AI, but for most legacy players, that is very hard. For example, financial industry players cannot rearchitect because there are so many regulations to get through, etc. So maybe those will survive for a while. But those enterprises that do not rearchitect face a pretty difficult future. I’m suggesting that a lot of the smaller players in the ERP space will probably be more nimble. It’s the larger players who have to deal with 50,000 customers that they have to offer a brand new solution to, saying, “Hey, throw everything you have out the window and use this.” That’s not an easy sell, so they have to balance the capex associated with this new development versus the opex associated with managing, migrating, and sunsetting their existing products.

I grew up with the cloud at Oracle, where we were transitioning from on-prem to OCI, Oracle Cloud Infrastructure, and I think they did a masterful job at transitioning customers along with their technology. Not everyone’s going to be able to do that. So, to answer the question of what strategic decisions have to be taken at the board level, there has to be a commitment saying no matter what, this is the minimum investment we’ll have in AI going forward, and we will stand by you in your transition. One good example is what Citi is doing. The CTO and the CEO are fully committed. It was initially going to cost  2 billion. It′s going to cost more than 2 billion. Beyond that, fine, they’re in it for the long term. So that kind of steady approach is one and 100% necessary.

Second, you need the right leadership and the right leadership that understands this new technology. What is data science? What is machine learning? What does it hold for me and my company in terms of values and possibilities for the future? Unless the leadership layer and one level down fully understand it, even if the manager and senior director layers understand and are at the forefront because they usually tend to be younger, there’s not going to be any movement because there’s a disconnect of buy-in between the executive layer and the execution layer. So I think that is step two: you have a C suite and an executive layer that is fully immersed in what AI is and what the technology is, and get technical with it. If not, get people to do that for you or replace that layer to make it go in the direction.

So those are two strategic decisions. Third, for the operational layer, you need to reskill your workforce as well. The operational layer is doing the work. That means you have to set expectations that your performance objectives change in the AI realm to what is relevant to AI. You’re no longer configuration experts or button pushers, but you all are analysts and  programmers. You have the tools to do this. And once you make that happen, I think the enterprise will be primed to switch. Till then, it’s going to be really, really hard because there’s going to be massive miscommunication and a misalignment of values and objectives.

Kashyap: I want to talk about this in two parts. First, let’s discuss legacy companies and large enterprises with massive operations. From the conversations I’m hearing, I personally believe this transition to AI-native or AI-first approaches is inevitable. What are some of the transitionary challenges and uphill battles these companies will face during this transition?

Second, let’s consider a smaller company, say one with around 5,000 employees. It’s much more accessible for them to reach out to function heads and set new processes. Despite this, there will still be challenges. The world is changing drastically, and companies must adapt or risk falling behind. What are some of the different challenges you see for these two sets of companies?

Akshaya: From an internal challenge perspective, there’s always this pressure to maintain the status quo. Let’s take two scenarios. One, a very large enterprise with layers and layers of management, as you rightly call it, bureaucracy. How do you overcome that bureaucracy and drive change? It’s really hard because you have to massage the messaging to go from one layer to another, and the messaging is lost at the end of the day from a value proposition because there’s no space for radical thought in a very large enterprise. You’re being a squeaky wheel at that point.

But there are some enterprises with, say, 5,000 employees that are already cloud-native, meaning they run on SaaS and don’t have any on-prem baggage that these larger enterprises have. They could potentially transition quicker because they’ve already done it once; they’ve already moved from on-prem to cloud and know what the transition and transformation look like. The challenges still lie in, for example, let me give an example. We consolidated all our payroll providers globally. We had about 14. We came down to one. That itself is challenging because you have to convince people that this is not something radical that will destroy their jobs. It is something that they’ll have to adapt to, and it’s going to make processes better.

But it’s a payroll system, so you are not only paying for the SaaS platform while it’s being implemented, but you also have to run it in parallel for a year or a year and a half because you’ve got to pay people the right money. So, this is a double cost associated with any transformation. One, budget is one constraint. Two, as I mentioned, people don’t stick around the moment they hear transformation and think their jobs are shaky. A lot of people, it’s organizational inertia; they don’t want to learn new things. They hope to get by doing their job. For at least a quarter of the population within a company, if you don’t want to learn new things, you’re going to move to a place where you’re comfortable with the skill sets you have. The remaining company employees will move on with newer skill sets. So that churn eliminates a lot of institutional knowledge.

One thing you’ll see very frequently in all types and sizes of enterprises is that process documentation does not exist. So when you’re doing the transformation, you have no idea how the process works. You have to invest more into consultants and process mining and mapping tools to first understand what is going on, which you have to transition to the new technology layer. Okay, those are three. Then four, there’s the external pressure. All the changes that you’re making aren’t free of charge in terms of disrupting the customer experience. You should not ideally disrupt the customer experience because you want to change your customer base; they want to continue using the software or the platform without any glitches.

So that’s the fourth pressure, especially if your company has a lot of debt or if you’re a publicly listed company: financial results. That’s one of the biggest downers to any transformation effort because you want to move to an ERP solution. Even a company as big as Zendesk, it’s going to cost $10 million. That’s not chump change. Plus, you’re paying for all the SaaS applications while the transition is happening. So, it’s an additional $10 million. Add a 25% extra to it as a buffer. So, that’s $12-13 million, and it’s going to take 2 years. Are people interested in doing that day job and doing this work? Probably not. You can hire an external consultant. Are they going to do a good job? Maybe. So, all these uncertainties are the bottlenecks from a transition perspective.

Now, talking about how many truly understand AI and the possibilities with it? The vendors selling you the stuff are selling the old stuff. Plus, they have a vested interest in selling the old stuff because they get kickbacks. If you go to a partner, if you go to a service provider, they’re going to sell particular platforms. They’re not going to say, “There’s this new shiny thing. Take this.” Because they don’t want to take a risk. They know they can sell the old one. So all these challenges, when they come through, make it really, really hard for decision-makers. That drags along the status quo.

Kashyap: Let’s discuss how larger companies have always managed transitions, whether in technology, processes, or business operations. What’s a good way of handling these transitions effectively? You’ve mentioned working closely on consolidating your enterprise suite into a unified system. However, as you pointed out, even Zendesk’s salespeople now have to convince newer clients, indicating that the mindset from a vendor perspective also needs to change. That might be a topic for another podcast. But focusing on large enterprises, let’s take a specific example instead of giving an overview. Let’s consider CRM, or a few work processes together, like CRM, ERPs, and ticketing systems.

Can you walk us through a rough framework you have in mind? What are the first steps, and what should be step three? Your insights would help large enterprises understand how to approach these transitions.

Akshaya: Large enterprises have made it past the internet revolution and multiple disruptions. I agree, but it’s come at a great cost financially. Very few have survived very well. Just look at IBM, for example, not doing as well as it was 20 years ago with the technology they were building.

To answer your question on what framework we can use, I think the first and foremost piece, as I mentioned, is the executive layer and the board have to be apprised of the possibilities of the situation, and they have to be educated about this new technology because it’s drastically new. In any company, what will happen is you will have 10% of the talent that is so interested in exploring new technologies and making it happen. Bring those people together because that’s where the new ideas are going to come from. We’ve done that pretty well at Zendesk. We are constantly scouting for folks who are interested in AI and asking for ideas. What can we do here? What can we do there?

One of the initiatives we’ve launched is called Zenovate, where anyone can give feedback and ideas on what to do with AI. It can be on product, operations, anything. No holds barred, and you get an opportunity to take a shot, demonstrating that this can be done. So that gives confidence to the employees and leadership that we are moving in the right direction, and you start to build that momentum and enthusiasm towards a new technology. That’s the first thing: you have to build that enthusiasm.

Second, when you start building out these new proofs of concepts, keep a constant flow of storytelling going at all levels and all hierarchies. Of course, you’ll have to have buy-in from your boss and your boss’s boss, etc. But we have been fortunate that there’s been a push top-down as well as bottom-up to explore AI both on the product side and on the operations side. So that’s a commitment they have to make: we are going to listen to voices that talk about this new technology.

And when you do the proof of concept, you have to promise that if this POC works, you’re going to scale it to the entire company. That’ll give the confidence that this is not just lip service, but the company’s serious about moving in this new direction. One thing we have done is there are these pockets of innovation in every function. Every quarter, we are trying to bring together those stories, called the innovation update, and it goes to the executive staff, saying, “Look what marketing is doing, look what finance is doing, look what HR is doing.” It gives them a glimpse, a view into how operations are changing and how it’s going to impact our revenue and cost model, so let’s act appropriately.

Some changes on the wholesale technology front: there has to be a foot down on what we are going to change. It’s not a matter of are we going to do it? It’s a matter of when, and given the paradigm that everything has to change on the operations front because it’s going to change the entire value chain, the question is how do we sequence it? What’s the most important? It’s not about priority; everything is a priority because you have to move forward in that direction very quickly. Sequencing should be squarely aligned with your business strategy, squarely aligned with what your company wants to do with respect to the value capture part of it as well as value creation.

So if you put that together, I think over a 2-year period, you can flip a company around to be AI-first. But that number one piece is commitment. Once you have that, I think everything else will fall into place.

Kashyap: My last question to you would be about the build versus buy decision. I don’t think companies can build everything entirely on their own; they will need to buy some components. Silicon Valley companies like DoorDash and Target prefer to build everything internally and have structured their teams accordingly. However, legacy companies might struggle with this approach. What is your guidance on reaching out to the right vendors? Who is the right vendor, and what qualities should you look for? Should companies prioritize moving from product to services? Any advice on this would be really helpful.

Akshaya: Choosing vendors is a hard thing to do. Legacy software or not—the procurement function is probably the most underrated function in any organization because they help move this stuff along. But from a guidance perspective, every company is going to be different. From a business model standpoint, if you want to go the services route, your technology stack is going to be completely different versus the product route. 

Building is an art form. What has been proven again and again in business studies and research is that whatever you’re building, if it’s not part of your core strategy, don’t do it because you’re going to do a half-assed job. So, most companies would tend to buy. For engineering-rich companies like OpenAI,  they are going to build it because they have the engineering talent to do that. But the next shift that is probably going to show up, and this conversation has already begun in VC circles and private equity circles, is that platforms that build entire platforms.

So, you can say a souped-up version of ChatGPT where I can put in a prompt saying build me an ERP system. I have 5,000 employees. I have operations in these 20 countries, and this is my revenue. This is my cost base. It’ll build you an ERP system. So, that shift is going to happen at some point where all the legacy ERP providers probably won’t survive if that comes to fruition. If you can build your own software, now you’re talking about moving towards an AI operating system for the company where everything is under one roof, everything is yours, and everything is provided by one vendor.

So, that may happen, and that comes with a big risk of vendor lock-in. It’s already happening because there are only a couple of good, reliable vendors. So, the build versus buy decision may shift over the next five years, but right now, if you want to buy, you’ll have to gamble because the AI native companies are small. They may go bust. You’ll have to put your faith in them. One way we’ve gotten around the issue is by making investments in the products that we buy through our Zendesk Ventures arm, which everybody cannot do. You’ve got to have money to do that.

Two, having a fallback plan: if the vendor goes out of business, do I get the source code to keep running? We’ve had those conversations with a couple of vendors, and I think that’s going to mature out as these companies grow and become bigger over the next 5-10 years. But companies cannot wait; they have to make decisions now, even if it means just learning how to manage and operate AI systems. Just learn from that experience, and you can always fall back to legacy processes if it doesn’t work out. So, that’s where proofs of concepts help and vet the companies.

So, I know it’s a nebulous answer, but that’s where we are today. It’s going to get better as we move forward.

Kashyap: Fantastic, Akshaya. This was a lot of fun, and I’m very sure that it’s going to add a lot of value to our audience, especially senior leaders who are working in this space. I don’t think anybody is taking up data and IT roles separately anymore. They are going to be working together very closely. So, it’s going to be interesting to see how things unfold. Thank you so much for your time.