When you choose a software provider, how much ownership do you really have over the technology that powers your business?

It's a question that many companies are asking as they integrate artificial intelligence (AI) into their operations. While most simply plug into a third-party AI service or a massive open-source model, cloud software giant Zoho has taken a different path. The company decided to build its own proprietary AI models from the ground up, a strategy its global chief evangelist, Raju Vegesna, likens to car companies that build their own engines.

This unconventional approach, while requiring significant investment, gives Zoho unparalleled control over the privacy, performance, and long-term vision of its AI deployments.

One of the most striking aspects of Zoho’s AI, named Zia, is that it isn’t a single, monolithic model designed for every task. Instead, it’s a collection of smaller, specialised language models. The system automatically routes a user’s query to the most suitable model based on its complexity.

For a simple query, Zia might use a smaller model with just a few billion parameters. For a more complex or demanding task, it can be routed to a larger, more powerful model. This intelligent routing ensures that Zoho uses only the amount of computational power necessary for any given task. The result? Significant cost savings and greater energy efficiency - a crucial benefit in a world where AI consumption is rapidly increasing.

As Vegesna noted, "You don’t need an LLM for anything; that will require a bunch of resources." This is a philosophy that prioritises solving a problem efficiently over using a sledgehammer for every task.

In a business landscape increasingly concerned with data security, one of the biggest benefits of Zoho's approach is privacy. By creating its own models, Zoho has complete control over the input and, therefore, can guarantee the output in terms of data privacy.

The company has a strong stance on not training its AI models on customer data. This is a critical distinction from many other AI providers who use user-generated data to refine their models. By controlling the entire process, Zoho can stand by its promise to respect customer data privacy, a promise that would be nearly impossible to make if they were relying on a model built and trained by someone else.

DELUGE is used throughout the Zoho ecosystem to provide a high level of customisation and automation. Examples include:

• Zoho CRM: To automate lead scoring, send a personalised email based on a customer's action, or create a custom button that performs a specific task.

• Zoho Creator: To build custom, business-specific applications from scratch, giving companies complete control over their internal software.

• Zoho Flow: To create complex multi-step automations that connect Zoho apps with hundreds of third-party services.

The existence of DELUGE proves that Zoho has long understood the value of owning its core technology. It empowers businesses to take full ownership of their tech stack, just as building Zia allows Zoho to own the core of its AI.