Prashanth “PVK” Krishnaswami of Zoho - We Don't Acquire Companies for IP, We Build It from Scratch

Today Zoho announced a seies of artificial intelligence investments and offerings, including its own large language model, Zia LLM, and 40 ready-to-deploy Zia Agents. These developments aim to help organizations maximize the value of AI technology and improve daily workflows.

I had an opportunity to spend "A Few Good Minutes" with Zoho's head of CX strategy, Prashanth “PVK” Krishnaswami, to get the inside scoop on how this move fits into Zoho's overall stragey of designing and building its own foundational technology and how these AI tools and services fit in with its Ulaa browser.

After checking out the release and attending an analyst call, a few things stood out:

• Zoho continues itscommitment to owning the entire AI stack, from AI capability to infrastructure and data centers.
• This approach should allow Zoho to have more control over cost related to AI processing power consumption, avoid being a costly input to customers as they gravitate to higher levels of AI usage.
• Zoho highlighted the increasing usage of AI with more than 16 billion AI API calls in the first half of the year, representing a 50 percent year-over-year increase.
• Zoho is exposing data through the Model/Content Protocol (MCP) server, and support for the Agent2Agent (A2A) protocol will be implemented.
• Zoho's partnership with Nvidia extends beyond hardware to include software collaboration, with a focus on LLMs for B2Bs.
• The Ulaa browser was developed with similar motivations—privacy and an integrated platform for AI and apps. The browser is seen as an entry point for customers and a platform where AI (such as phishing protection and future in-browser agents leveraging computer vision) can be embedded invisibly.

Privacy and security are primary reasons Zoho decided to build itsown LLM from scratch rather than acquiring IP, maintaining high standards for cloud privacy and security.  The increasing amount of data from Zoho’s diverse apps (business applications, productivity apps, collaboration tools like Cliq, Creator development platform) presents opportunities to leverage AI across many data types for productivity gains and improved outcomes.

One final thing to point out: By developing its own LLM(s), AI agents, and core infrastructure, Zoho reduces reliance on third-party providers and expensive licensing. And because of that, it doesn't plan to charge high premiums for AI agent features. There might be a small premium in certain high computational processing needs, but there should be no premiums at all for in-house AI-powered functions.

Below is a clip from our conversation, along with an edited transcript.

Edited Transcript

Brent Leary:  So let's talk about LLMs.

PVK: The first reason we decided to build our own LLM from scratch was to maintain privacy and security standards. We've set a really high bar for ourselves with respect to this SaaS layer, and we want to maintain that bar with all of the upcoming AI value, so that's the first reason.

The second is, as a company for almost 30 years now, we've had a drive to build foundational technology, and we're more interested in the know-how of being able to build these foundational technologies because we typically don't acquire companies for IP.

We build IP from scratch. So the process of building the IP teaches a bunch of things that can be laterally applied to various other areas of the business. There is a certain amount of value for us in doing that, so that's certainly in line with being able to build our own technology from scratch, our own browser, we thought our own LLM was next step. So we certainly did that.

Leary: I think it was last week where Perplexity came out with its own browser, and I remember being at one of these Zoho Day events. I think it was Raju who introduced the Ulaa browser. I remember seeing some people that were like, "What? What? Why are they doing a browser?" And that was before the new AI revolution. Now it feels like people are starting to understand.

Things that were commonplace a few years ago are being completely disrupted, and now you have this new infrastructure, new foundation. Well, I shouldn't say new because it's been there for a while. It's just you guys are ramping it up more and more and more.

Leary: What role does the browser and the LLMs and all this stuff play in the overall strategy of where you guys are taking this?

PVK: I cannot, in all honesty, say that we had all this figured out 10 years ago when we were building this. Somewhere along the way we learned how to connect a lot of these elements as they're being built.

So we realized, "Hey, that looks interesting, and that looks interesting. We can just put those together, and I'm sure there will be something that's valuable to the customers for that." We've already had a little bit of AI in the browser itself,  predominantly to do with phishing and malware and things like that, just to protection of all of that stuff. Now we're able to integrate a lot of our apps into the browser, and in the future, we expect that there will be a demand for in-browser agents, and basically take advantage of computer vision and things like that. And be able to build more and more value on the browser.

But we see the browser as an entry point. We always have. It took us a little bit of time to build the browser to our standards of privacy and security, but we did it, and we're starting to see a lot of interest in it already because there's not that many browsers that have the luxury of prioritizing on privacy.

Leary: When you see what's going on now with what Perplexity just announced, and Google is already trying to figure out how to integrate more AI into its traditional stuff, it feels like it could be a little luck, but it also looks like a lot of foresight too.

You know, all the focus on agents and agentic AI. You just mentioned you provide some kind of out of the box. What are some of the ones that you think are gonna impact your customer base up front?

PVK: One of the primary bottlenecks is being able to nurture customers into your funnel. The other one is in being able to support them after a sale or or during a sale.

And these are the two primary bottlenecks, and they're also, in a certain sense, not linearly scalable because you may not find the right talent to be able to scale up when you need it, and this can constrain your business.

Brent Leary: When you think about all the different kinds of data, I mean, there's the stuff that goes on in the applications, and there's stuff that goes on with interactions via a browser. Being able to collaborate on, is it Cliq? There's a lot of interactions and discussions, apps that are more productivity, so there's a lot of stuff going on in there. Business applications, now you got a lot of stuff going on.

People using Creator and, and development stuff. There's a lot of data now that there wasn't, like, 10 years ago. All of that data can now be applied. You guys have built the infrastructure that allows for more and more interaction and data types to be flowing through. Give us a little look a year or two years down the line.

PVK: We will certainly invest in smaller language models, visual models, and things like that. And the idea is on one side we want to build a lot of AI that's invisible to the user, that nobody knows is AI. And honestly, they shouldn't care about whether it's AI or not as long as it's  computationally efficient for us and we're able to run it on existing infrastructure; and as long as it gives us better quality outcomes for our customers. It should be invisible and should run in the background. So that's one layer we've built.

In the past year, we've consciously branded a lot of it as Zia, which we didn't do before, but now we're essentially pulling it up just to talk a little bit more about it and say, "Hey, all of this that you've been using so far is also Zia." And we'll continue to build on that invisible AI layer for sure.

Another one that we're seeing is agents for configurations of software.  In the future, and I'm coming at this from a very dreamy angle, I'm not saying this is going to be ready in the next six months or anything like that, we hope that there will be a situation where you're able to draft a document, maybe about two or three pages, about what exactly you need, what you're expecting from a software, and the agent will build that on Zoho for you.