Cloud-based app vendor Zoho recently concluded their fourth monthly AI forum, with the idea of making AI conversations with partners and customers more continuous, and providing updates on the AI front. Raju Vegesna, Zoho’s chief evangelist, emphasized the centrality of CRM and AI to Zoho’s strategy and road map.
“The usage of AI has been ramping up today,” Vegesna said. “We get about 16 billion AI calls across the board in Zoho Corp., and this is if you look at the data from the first half and that’s about a 50% yearly increase across the board. There’s a good, strong adoption happening among customers and we have been investing in AI for a long time, and we have been building up the entire AI stack, and I want to talk about today’s announcements of the various layers of this. If you look at our first layer that is the AI capability that we are exposing to our customers, we kind of break it into three. One is contextual AI. We call it Contextual Zia, and then there is Zia Agents as well as Ask Zia. A lot of the previous announcements we had were related to contextual AI integration, where the user doesn’t understand that it’s AI doing the work.
In this release, we are making big improvements to Zia Agents. If you look at the Zia Agents side, what we are doing is building several agents by default and including them as part of our products. We contextually integrated them into the products and the users can enable them and include these agents as part of workflows and buttons as well, and they can deploy these agents as in the background, so we are including several of these agents by default and all of these agents are included in the product available for paid editions, not like where some customers feel that vendors are shoving AI down their throats and trying to charge more. We are trying to deliver as much value as possible, and in this case a large percentage of the agents, if not all, are going to be included as part of the products. We will see hundreds of agents developed from Zoho, all contextually available, and these are going to be integrated into the products as well. There might be cases where the cost of the resource might be high and we may have to price agents.”
Zoho drives a lot of value through their software so there’s a lot of work going on with a lot of prebuilt agents. This includes Ask Zia, a Zoho Wide agent that is integrated contextually across Zoho applications, and which Ask Zia is gaining various skills in.
“A couple of months back, we showcased some new skills for Ask Zia in the context of CX where Ask Zia can help create workflows and customize and design some templates,” Vegesna noted. Zia is deepening skills in two areas: one is on the data analytics side, and the second one is the data engineering side, where you can go to some of our tools and ask here to import data. In this case I’m trying to import data, sales order data from Zoho, CRM and Ask Zia will connect and import the data into this into the system. Similarly, it doesn’t have to be Zoho data. You can pick that data from a third party.
“In this example, I have Amazon RDS and you can import that data into the system as well and it does the import,” Vegesna said. “But then once you have the data imported from two different sources, you may want to combine the data to join the data sets and maybe group them by order ID. In that case it basically shows you and previews the inference information that you can apply that transform and it does the work there. So once you have combined the data set, you may want to push that data set into a workspace in Zoho Analytics as a new table.
“Now once you define the destination, it will combine it and push it to Zoho Analytics and you can ask to schedule it, schedule the pipeline so that it runs, say, every day at 10:00 a.m. and it automatically does that.”
Once you are into Analytics, then you can analyze the data analytics.
“All of these are newer skills we are adding to Zia, helping data engineers, But once the data is put together, that part can also be done here with a new skill that we have added to Ask Zia as well and in this case you can go to Ask Zia and say, “hey, what’s a monthly sales contribution by channel in the last twelve months,” and it will give you a chart of this. We have added to the reasoning capability. ‘So why did my sales drop in June 2025?’ It gives you the reasons on based on the analysis why it dropped and maybe ask suggestions on how you improve sales based on historical data. It may give you some suggestions about data and of course, you can choose to share that as well if you choose to share that data. It will give you a shared confirmation and once you have that shared confirmation, you just confirm it and you can share it all during the interactions, so that’s an additional skill that we are adding to Zia as well.
“Now that’s on the AI capability, so we have Ask Zia, gaining new skills, as well as agents where we are adding a lot of new agents to Zoho contextually embedded,” Vegesna said. “Moving one level below the stack, we have been working on AI tooling for our developers and the ecosystem with several tools, which were launched in the past, in the past year or so. This includes Quick ML Auto ML, DSML and several other tools, which were available through directly low code, no code and pro code tools.
“We are adding a new tool to the tool kit, which is Agent Studio,” Vegesna continued. “While we are building tens of agents, and we expect there will be hundreds of agents to build, there are going to be cases where customers will want to build their own agent and we are introducing an Agent Studio, yet another low code-no code tool that users can build agents with as well. It’s a tool that makes it easy to create agents. and you can make it really complex or keep it simple.” In this case, users can create an MCP fairly quickly with Zoho and expose the data.
“In fact, if by default I can say, ‘hey, I want to create a sales agent’ — I’m picking a very, very simple one here — so you can say what are the functions that you want this person to do,” Vegesna asked. So even that has to be automated. In this case I want him to qualify leads, maybe take notes, assign tasks and assign leads and maybe possibly send emails. So those are the suggestions that you are giving based on that prompt that Zia tries to create an agent, and takes some instructions now you can include some knowledge as part of it. You can include knowledge from any of the systems that you have within the company or outside. You can define some tools, and we have a lot of tools out there. In this case it picked these tools and it defines some guardrails and creates an agent which you can deploy various ways.
“Once you deploy it, in this case if I want to deploy an agent as simply as a digital employee, it gives you some details on some parameter mapping and you can ask where do you want to deploy it,” Vegesna said. “In this case I want to deploy this particular agent in Zoho CRM. You can select the business and it gives you the permission structure that you have in the system. Let’s say I’m picking a administrator profile in Zoho CRM. In that case, those profiles and those details will be taken from that and you can deploy it as an individual digital employee where you give it a name and I can give it an email address and that creates an individual user with the roles and profiles. So the entire structure is integrated into the permission structure from CRM into this as part of the deployment model. Once it is deployed, go to CRM and treat that as a digital employee and go through it. You can get more comprehensive. In this case I just want to show a Quick Way of creating an agent but you can create several complex and comprehensive agents. And of course you can add sub-agents to the system and keep building that. So Zia Agent Studio is a new tool, which we are adding to a toolkit as well.”
Vegesna emphasized that as long as the tools are exposed you can make that available and perform those too.
“You can generate a chart because those tools were exposed through MCP,” he said. “You can also export this as a PDF file, and it uses the appropriate tools from analytics exposed through MCP to generate that PDF as well, and that that PDF of course, once it is available you can share it.
“That’s on the MCP side,” Vegesna noted. “The other portion of this is the LLM, so we have experience with a lot of AI algorithms. That’s a new part now moving one level below on the AI infrastructure side, such as the AI Bridge. We have built several models as well and other things like governance, where directory plays a key role. We have added a lot of those in the past. We have connected with over 200 third parties, third Party applications, Zoho plus third Party apps and you can connect to any agent or on an MCP client today. You can take the MCP URL from our MCP server, connect it to a client, go to the integrations, add a new integration and in this case we are doing this integration with CRM, add that and enable that integration so any MCP client can connect and now extract the data.
“Let’s say you pick Zoho Analytics as one of the applications that you have exposed,” Vegesna stated. You can go to any, any client –in this case we’re going with a Github copilot and connected analytics — and you’re asking copilot to analyze the data through MCP with data that you have in Zoho Analytics that has pulled the data from Zoho Desk. If you ask “why is my response time low,” the Github copilot doing the analysis connects through the MCP server in analytics.”
Vegesna emphasized that as long as the tools are exposed, you can make that available.
“You can generate a chart because those tools were exposed through MCP as well,” he said. “You can also export this as a PDF file, and it uses the appropriate tools from analytics exposed through MCP to generate that PDF as well, and that PDF of course, once it is available, you can share it.
“That’s on the MCP side,” Vegesna noted. “The other portion of this is the LLM, where we have experience with a lot of AI algorithms. We are now launching our own LLM and this is not an open source LLM where we have taken it and made some changes and are branding it as Zoho LLM. This is a completely built from scratch homegrown LLM. We are launching three models, with 3,000,000,002.6 and 7 billion parameter models out there and each one was was trained separately, so the 2.67.3 are not distillations of a 7 billion model. These were all trained separately and there are going to be more coming as well.”
Vegesna said that if you look at the how Zia is performing, Zoho likes where they are currently. But those familiar with Zoho will see it’s like seeding something and seeing it grow.
“This is the first step and we’re going to see it all over a period of time,” he stated. “It certainly performs better than Lambda 2. We’re not there yet on the Lambda 3 and Lambda 4 is a bigger context when you’re going to be having similar to Lambda 3 but we are confident on some of the work we are doing where we can optimize it. We are really not optimizing it for B2C scenarios. We are doing this LLM focusing on B2B, so it will be interesting to see how that turns out, but we like the progress we are making so far now that there are larger models.”
Vegesna said that apart from their LLM, they are also launching some ASR models and a speech recognition model, and some of these are again completely homegrown models and they are launching two models, one for English and the other for Hindi as well, so these models are integrated with Zoho Meeting, Cliq Trainer Central and all of those. You’ll see more of these integrations coming too.
“In a lot of cases larger model performance tends to be better, but it is not the only dimension out there,” Vegesna said. ‘We want to look at the price to performance ratio and various others, so it makes sense for us to develop a model and optimize it for various scenarios. “If we own the model and get creative, it will enable us to offer the right model for the right scenario and we are making the improvements in multiple dimensions as well. Certainly, right-sizing the models that we have talked about in the past IS the key approach. For a critical tech like AI, this is essential. It would be like cloud companies not owning the clouds or AI companies not owning their AI engines, It doesn’t make sense for us, so we like to be in control of the core technology that also gives us the freedom to innovate and come up with our own ideas based on our own convictions to deliver value that is going to be a long-term. We are planting a seed here and we’ll see it grow into a tree in years to come.”
Vegesna said that all of this innovation falls in line with some of their investments made in the past.
One of the most interesting questions from Vegesna’s perspective is around AI governance by design and how Zoho is managing the quality and testing and security of the agents.
‘It’s vital and fundamental, especially as you move from individual agents to fleets of agents and combined agents, and the unpredictability of those is problematic,” he said. “The permission structure or some of it if it is defined as some kind of a digital employee. Where you’re inheriting some of the things that were predefined for an employee, defining an agent without existing boundaries on them, those boundaries have to be predefined from scratch and that is where systems like Directory play a critical role. We are changing the directory system to enhance that part of it. “We cannot simply rely on the black box to expect to deliver what we promise to the customer. Can a third Party LLM guarantee that it can forget things you don’t want the LLM to remember? There’s no guarantee that LLMs can forget that so we want to make sure that we design the system that way and that’s part of the reason we are doing our own systems because we can control some of these aspects and make sure that we can guarantee the privacy and certain aspects that are critical to the governance as well. We are actively thinking about the governance part, and that’s a key part and it’s not like the industry has figured it out.”
Vegesna commented that what we are seeing in the AI market is very initial phases.
‘In the next say five to ten years is when we’ll see the real AI native applications, and we want to be one of those companies that will create some of those AI native applications. But you need an engine where you can add your own ideas on top of it, to take it in a certain direction in the B2B context and that’s where we like creating that. We want to control the core piece of technology and control our destiny. So as a company, we will continue to remain independent, but only if we have the parts of the core piece of the technology.”
CRM is key to Zoho’s long term strategy.
“When it comes to structured applications, CRM is our most advanced platform, whether it is capability, functionality, the depth, the scale and all of that,” Vegesna said. “What we are doing is taking some of the core aspects of it and and generalizing the platform and making sure that capability is available throughout Zoho. The CRM foundation is a very key part of Zoho’s strategy. If you have built something strong in the CRM side, some of those will flow into the back office side of things and vice versa as well. There’s a lot of sharing of technology, and embedding a piece of technology throughout CRM certainly is in advanced stage in several aspects, so we have taken that and we are pushing it through other applications.
“We developed and deployed our own data centres, and that gives us the freedom to expand a certain way and we have,” Vegesna stated. “We have expanded our hardware outreach, Our partnership with NVIDIA here helps, so the partnership is not just on the hardware side, but also on the software side. NVIDIA doesn’t have too many partners that are doing LLM’s for B2B, so we are working closely on that front and it’s a bi-directional collaboration where we learn from them and they’re learning from us in terms of the business use cases out there. We expect that to continue, where we can tightly integrate the hardware and the software. We will continue to expand our hardware deployments, data centres and all of those and that’s part of our investment.
Vegesna said that with respect to the API, at the moment Zoho is not looking to provide an API that would be available for consumers to directly consume.
“In the long-term, we may expose an API through a a platform so that consumers are able to access it but at the moment, we don’t have plans to do it,” he said. “Maybe we should allow somebody else to own part of the stack only because you’re fighting against just crazy numbers. We’re going to compete against Microsoft and those with public consumer alms. We are very focused on the BTB, and that is where we think there has to be more investment going in. People said you should not be investing in data centres, but we like where we are because we have invested in those data centres. We will only be able to connect these dots looking back few years from now because our conviction says that we have to invest in it and and we’ll see how things go. We can deliver some alternative views and we’ll see how things change as technology changes.”