Your browser doesn’t support HTML5 audio
Every enterprise vendor has an AI strategy - so how is Zoho's approach different? Let's face it - most enterprise vendors talk a pretty good AI game. Yes, some vendors make too many vague boasts about autonomous agents. But: they also tell you about their emphasis on data quality and data privacy. Rightly so: garbage in, garbage out is one of the toughest AI project lessons (what organization likes to look in the data mirror?)
When it comes to achieving better AI, the details matter. Businesses rarely achieve a trusted AI result with out-of-the-box LLMs; we need to understand the architecture that leads to better AI results. We need to hear from vendors not just about their data privacy efforts, but their take on ethics, governance and risk.
Zoho is known for memorable (and provocative) AI statements. Want a sample? How about Raju Vegesna, Zoho Chief Evangelist telling a room full of industry analysts:
If someone can pull the plug on intelligence, are we really sovereign?
Vegesna was speaking of the global trend of data sovereignty - and localized models that break dependence on a single LLM/intelligence provider or hyperscaler.
But the depth behind those statements is what I'm looking for. After all, isn't that what customers need right now to parse the AI substance from the marketing noise? So, during the Zoho Analyst Day opportunity, I seized that chance, via two separate video discussions. Now, we've pulled that audio together into one Executive Intelligence podcast. First, I talk with Ram Ramamoorthy is Director - AI Research at Zoho & ManageEngine.
During my talk with Ramamoorthy, we delved into:
- What is "delayering the stack" - and why does it matter?
- How do we get to contextual intelligence?
- And how should we approach data privacy and AI ethics in our development and application choices?
During our talk, Ramamoorthy explained what contextual intelligence means to Zoho:
Setting the right context that is available to the person who is using the LLM or the agent becomes super important, because what we have done is a full stack approach. Now, our AI evolves over our search indexes. So basically, an agent that I built will only be able to access data that I have access to.
Now, let us say we both work for Zoho. I will not be able to have my agent summarize your emails, your chat messages, because I do not have access to them. So because of this, privacy is baked into the system, the hierarchy. It knows who my reporting manager is. My reporting manager has to approve my travel request, my leave request, and all of that. Now that information, that meta information, that context, is carried forward across systems, because we have unified processes, because we have unified all the data in the system.
Next up in the podcast, I speak to Raju Vegesna, Zoho Chief Evangelist. We went deeper on:
- Why Zoho builds its own technology platform - including LLMs
- The difference between a frontier model approach to AI and a sovereign approach, where context and privacy are the priorities.
During our podcast, Vegesna hammered the so-called "open source" virtue signals of big AI vendors - and how this always seems to shift towards proprietary value extraction - a process that now has a famously edgy name. Vegesna explains:
At some point, these closed system will become [value] extraction systems. That is something that is making customers uncomfortable. Of course, we feel uncomfortable with that.
We think there is, a different model, where we should not move towards centralized systems, trying to extract value. Customers and their businesses should be in control. That's why we are looking at alternative approach, where we build the models for the customers, let them train, give them the tools, let them train the models with their data, and use it for their business. Increasingly, those discussions are becoming highly common.
But how should Zoho's own customers navigate these changes? And what about the market challenge to SaaS vendors to justify their value? How can we reframe AI around different values than headcount extraction? Time to hash it out... Hope you enjoy.