In this post, we share highlights from a recent conversation with Prashanth “PVK” (PVK) Krishnaswami, Head of Marketing for Zoho CRM, about Zoho’s latest AI announcements. From building its own large language model to launching pre-built AI agents, a custom agent builder, and an agent marketplace, Zoho is taking a distinctive approach to agentic AI. We discuss the strategy behind these moves, the company’s focus on privacy, sustainability, and practicality—and what it all means for Zoho customers. Read on for a summary of the key insights from our discussion.
LAURIE: Today we’re here to talk about—what else!—AI, with PVK, head of marketing at Zoho for CRM. AI developments are moving fast, and business application vendors are hustling to outdo each other. Zoho has recently made some big announcements, which I’m looking forward to learning more about. PVK, can you start with the big picture view on Zoho’s overall AI strategy?
PVK: Of course. First, Zoho has been working on AI for over a decade. We look at AI as more than a multiplier—it can provide exponentially more value, enabling us to tackle time, effort, quality, and cost issues in business automation simultaneously.
Second, Zoho has always focused on building the foundational technology that underlies business solutions. We don’t usually grow by acquisitions, we prefer organic growth. Building foundational IP (intellectual property), including for AI, enables us to reuse a lot of our knowledge across our solutions.
Third, we build for value, not vanity. We’re usually not the first vendor to ship something because we choose to observe the technology landscape and take the most pragmatic approach that works best for our customers.
We also have a vast library of 50-plus SaaS applications. All of them have APIs that agentic AI will rely on to take action—making our agent layer extremely powerful right off the bat.
LAURIE: Zoho is building its own large language model (LLM) instead of solely relying on publicly available LLMs. Why did you decide to create a native LLM?
PVK: It may seem a little counterintuitive when many models are available and we can integrate them with APIs. But the top reason for building our LLM is to support our commitment to privacy and security—a major reason why customers choose Zoho. Also, we like to build foundational technology, as I just discussed.
We’ll continue to work with LLM partners, because customers have valuable use cases with these models. But with our LLM, we can right-size the model. Not all use cases are equal in terms of complexity and value. You want to choose the model that’s most appropriate. Simpler use cases can use a smaller model, which helps us manage costs over the long-term.
LAURIE: Zoho also build its own web browser, Ulaa. What have you learned from that that fed into the decision to build an LLM?
PVK: With Ulaa, we also wanted to double down on privacy and control as much of the technology stack from a privacy standpoint, from the browser to the applications. Now we have these two fundamental components. I’ll leave it to your imagination about how we could put these together in the future.
LAURIE: That could get very interesting. Another question I have is, how do you think about the environmental impact of building another LLM?
PVK: You don’t need a large model for many use cases. Narrow to smaller models will often do the job, with less environmental impact. We’re testing our models on internal use cases first, assessing them for sustainability before rolling out to customers. We also run our operations on renewable energy. We want to use resources in a manner that is both commensurate with the value of the use case, and also environmentally responsible.
LAURIE: Shifting gears, our latest SMB Group survey indicated that SMBs’ top concern about AI is the accuracy of AI outputs. What is Zoho doing to address this?
PVK: We are deploying an additional layer of context that comes from our business applications. Combined with specific context from customers’ use cases, this automatically lessens hallucinations. From an agentic standpoint, agents are provisioned to specific applications with a layer of permissions and data privacy baked in. We’re also building dashboards for our internal use cases, which we will provide to our customers as well.
LAURIE: Speaking of agents, this is the perfect segue to the other part of this announcement, which is about agentic AI. What are your goals here?
PVK: Our primary goal is to help our 900,000 paying customers modernize their businesses with operational and strategic efficiencies that save time and money, and raise the quality of customer and employee experiences. We’re looking for initial use cases that involve rote, redundant work that will give customers a win right off the bat.
We’ll have a dedicated section within our marketplace for our Zia AI agents. But you don’t have to visit the marketplace to install them. They’ll be baked into the products you’re using.
One example is a revenue specialist agent, which helps sales reps identify opportunities to upsell and cross-sell existing customers. Another is a customer service agent for simple customer service inquires. The agent can process incoming customer communication, understand the context behind it, select responses from a knowledge repository, and send a response. If necessary, the agent will triage it to a human service rep.
LAURIE: How about the custom agent builder?
PVK: The agent builder gives the average business user tools to customize agents using prompts. You can build agents for about 700 actions, and deploy them on Zoho and on any of your other applications.
LAURIE: Last, but not least, Zoho is launching an agent marketplace. How will you curate and certify third-party agents?
PVK: Zoho Marketplace for applications has been in existence for many years, and has over 2000 extensions. The curation process that we use in the app marketplace, and our partnership expertise with ISVs gives us a great foundation for the agent marketplace. Our curation process is the same, with an emphasis on privacy and security.
LAURIE: When will these different offerings be available?
PVK: Zia LLM will be generally available at the end of 2025. We’re rolling out Zia Agents, Zia Agent Studio, Agent Marketplace and Zoho MCP Server to early access customers now, with general availability at the end of the year.
LAURIE: Thank you again for sharing this update and Zoho’s AI vision.
© SMB Group