Software firm Zoho on Thursday launched its own enterprise Large Language Model (LLM) – Zia LLM, Automatic Speech Recognition (ASR) models, and a no-code Agents builder. Zoho CEO Mani Vembu said their AI strategy is a ‘marathon’ and they have chosen to build all of this from scratch and not go in for massive-sized and costlier LLMs in the market, thereby ‘right-sizing’ the models to real business use cases.
Speaking to businessline, he said that Zoho, currently, “does not expect to pass on the costs” of the AI capabilities to customers as they are able to keep costs in check with own data centres, owns GPUs, and not having to pay model providers. This helps us add more value to the customers and we will continue to add more functionalities to Zia LLM, said the CEO.
Excerpts:
Why did you decide to build your own enterprise-grade LLM from scratch and not on top of other models?
We started with the question of how do we build efficiency and scale? If you look at some of Zia LLM’s use cases today, we believe that those use cases don’t need a model with very large number of parameters. So, by building on our own, we have the flexibility of right-sizing the models. We train them on specific use cases and we get better results at a lower cost.
That’s why we have launched three models now with 1.3, 2.6, and 7 billion parameters, each trained separately. In doing so, we see a long-term value for our customers. It is also being built entirely on publicly available data and on some private data that we got annotators to create for us, ensuring privacy. The differentiation also comes from combining the low-code and the AI model, which is helping us co-create with the customer with the user in loop.
What are the customers looking for today when they adopt Agentic AI systems?
Users want to really look at what kind of AI functionalities we have; what kind of builder tools we give for them to build their own models including what Agents offer. I do not have specific numbers to share but we have already opened it (Zia LLM and Agentic AI builder) up for early access to quite a few customers. Some of them have used our Agents builder to build their own tools and deploy it and have noticed productivity improvements.
With Zoho One now going to have these additional features, will we see the shift to new pricing models?
Definitely, in the long run, outcome-based pricing is ideal for AI SaaS, but the question today is how do you measure it. Currently, we have kept the pricing of Zoho One (full suite of business apps) fixed but we keep adding new apps and now features like Zia LLM are also part of the package.
But all these come with some costs, so how do we absorb that cost? That’s why we are right-sizing our models rather than partnering with a LLM provider and then passing on that cost to consumer. So, we are taking small steps by identifying common business use cases, right-sizing the model and solve those first.
How do you view all the big tech layoffs, which are said to be due to AI?
I honestly don’t know whether the layoffs are because of AI only, or the structural extra (people) that they have hired over a period of time. So, you hired expecting some growth and once the growth slowed down, they’re being let go. Today, there are no real instances where a company can say I can send all my staff home and AI can do their job. AI is not at that level because there is a lot of hallucination. If you make it fully autonomous and respond to a customer and if it hallucinates, then your customer experience drops. So, AI can actually co-exist but not replace people.
How is Zoho and the SaaS sector coping with the slowdown?
The growth levels for SaaS during 2022 – those days are are over. It’s going to be gradual and it’s not going to come from the boom. We are seeing good growth in India, West Asia, LATAM (Latin America). (Zoho grew 32 per cent year-on-year in India, in 2024).
We are seeing good growth in developing countries and winning more deals there. This is also because we have set up strong local teams in these regions. For us, the large enterprise clients are growing faster than small and medium-sized businesses (SMBs) as there is still pressure in SMB, especially in the US market.
Published on July 17, 2025
