We’re not chasing the AI hype: Zoho CEO Mani Vembu
The company’s new CEO shares insights into building Zia LLM, the shift toward agentic AI, learnings from Sridhar Vembu and more
By Naini Thaker
Sep 10, 2025, 17:08 IST5 min
Advertisement
From starting out as a software developer to stepping into the CEO role at Zoho, Mani Vembu’s rise has been steady and deeply rooted in execution. In early 2025, he officially took over the reins from his brother and Zoho co-founder Sridhar Vembu, who transitioned to focus on research and innovation as chief scientist. Mani’s appointment marks a new chapter for the Chennai-headquartered SaaS firm, as it doubles down on platform-driven growth and enterprise adoption.
One of the most significant milestones under his leadership has been the launch of Zia LLM, Zoho’s proprietary large language model, trained entirely in-house and tailored for specific business use cases. This move reflects Zoho’s long-term commitment to building AI capabilities that are cost-efficient, privacy-conscious, and deeply integrated into its product ecosystem. With 32 percent year-on-year growth in India, driven largely by inbound interest from larger enterprises, Zoho is now looking to replicate its success globally.
In a conversation with Forbes India, Mani Vembu shares insights into Zoho’s AI strategy, the shift toward agentic AI, and why decentralisation remains central to the company’s leadership philosophy. Edited excerpts:
Q. Why did you build Zia LLM from scratch? And what makes it stand out?
Most public LLMs are consumer-focussed. At Zoho, we initially used and hosted open-source LLMs, which gave us insights into the infrastructure, investment, and optimisation required. We realised that smaller models (such as 3B, 7B parameters) could be optimised for specific business use cases like summarisation across emails, support tickets, etc. without needing massive models.
So, we built three different LLMs—1.3B, 2.6B, and 7B parameters—each trained for specific enterprise use cases. These models are not open to customers directly but are integrated into Zoho’s products via agents and prompts. This makes Zia LLM an enterprise-grade model, optimised for cost and power efficiency, and tailored to Zoho’s ecosystem.
Q. When did Zoho decide to build its own enterprise LLM, and how long did it take?
Zoho has been working on AI since around 2011–2012. The launch of OpenAI’s GPT accelerated interest in generative AI, prompting us to explore our own capabilities. About 18 months ago, internal experiments showed promising results, which led to the decision to build our own proprietary models.
Training the models took significant time because we chose not to use public or customer data. Instead, we hired annotators (someone who labels raw data with meaningful information to create datasets for training AI models like GPT) and trained the models from scratch.
The three initial models (1.3B, 2.6B, and 7B parameters) were trained over 12-15 months, each tailored for specific use cases. Larger models (30B to 100B parameters) are now in training and will take another 4-6 months per model to be ready.
Q. How is Zoho preparing for the shift from generative AI to agentic AI? What trends are you seeing among customers?
The rise of digital employees is driving automation through agents. Customers want agents to handle tasks like meeting notes, deal analysis, and ticket summarisation. We have launched Zia Agents—pre-built agents for native use cases—and an Agent Studio, a builder that allows users to create agents from scratch or via prompts.
Interestingly, even the Agent Studio includes an agent that helps build other agents. These can be deployed across all our products as digital employees.
However, accuracy and reliability are key. Some tasks (like summarising emails) are low-risk, while others (like responding to customers) require high precision. We ensure that agents are powered by its own search infrastructure and data ecosystem, making them more effective and reliable.
Q. With so much investment going into AI globally, is Zoho already seeing returns on its AI investments?
The return on AI investment is still evolving. Our philosophy is to bundle AI within the existing subscription rather than charging extra. This drives the need for cost optimisation—for example, using a 1.3B parameter model for summarisation instead of an 8B one.
We track AI adoption through API usage metrics, and there’s been a 100 percent increase in AI feature usage year-on-year, indicating growing customer engagement. However, the broader industry is still figuring out how to balance massive AI investments with sustainable returns. We are focused on efficiency and value delivery rather than just following the hype.
Q. Zoho has traditionally focussed on SMBs. Why the shift toward larger enterprises?
We began by targeting business teams and citizen developers with low-code tools, allowing non-technical users to customise and deploy software easily. Over time, as the platform matured, larger enterprises began showing interest—especially in India—through inbound inquiries and word of mouth.
Our origin as a telecom platform company gave it the expertise to evolve from products to platforms. This shift has enabled Zoho to win larger deals organically, without aggressive outbound sales, and expand its reach into enterprise segments.
Q. The company is now taking a vertical-first approach. How is this helping accelerate growth?
We are seeing strong traction in specific verticals. In India, BFSI (Banking, Financial Services, and Insurance) and Auto DMS (Dealer Management Systems) are growing rapidly. In the US, real estate is a key vertical, while in Europe and Asia, BFSI and Auto DMS continue to perform well.
By focusing on verticals, we can reduce implementation time and cost, offering faster time-to-value. Our low-code platform enables quick adaptation to changing regulations and market conditions—especially important in sectors like banking, where processes and compliance requirements evolve frequently.
Q. What are your thoughts on the current talent landscape—layoffs, lack of skilled labour, and the debate around hiring freshers vs experienced professionals?
Our philosophy has been to build talent from the ground up, especially in areas where opportunities are scarce. Since its inception in 1996, Zoho has focussed on hiring freshers—over 90 percent of our workforce—and training them on the job. The company’s mission includes rural development and research & development, not just profit optimisation.
Regarding industry layoffs, I’m not sure whether AI is the main cause. It could be due to slower growth, past over-hiring, or other market dynamics. At Zoho, we use AI to assist employees, not replace them—helping improve productivity by 20–30 percent across roles like legal, support, and sales.
Q. What leadership lessons have you taken from Sridhar Vembu, and what changes have you made in your new role?
One of the biggest learnings from Sridhar is the value of decentralisation. Zoho’s decision-making is highly decentralised, which brings some chaos—like overlapping efforts—but also empowers teams and fosters agility. I really want to preserve this.
Sridhar himself is constantly evolving, so change is part of Zoho’s DNA. We are experimenting with regional and rural expansion, moving upmarket, and transitioning from product to platform. The focus is on staying agile and flexible, especially in a fast-changing AI landscape.
Q. What’s the one big challenge that keeps you up at night?
The biggest concern is security and privacy. As a cloud provider, Zoho places immense importance on ensuring that customer data is safe. Every decision is made with this priority in mind, and the company has strong privacy measures in place to uphold that commitment.
- Home /
- Leadership /
- Were-not-chasing-the-ai-hype-zoho-ceo-mani-vembu
Advertisement
Advertisement
Advertisement