Zoho Corporation [www.zoho.com], a privately-owned company I like a lot and have been following for some time, has just announced a major evolution of its Zoho cloud-based software suite – essentially this is an operating system for the business, which offers it some 50 integrated business, productivity, and collaboration applications (its Customer Relationship Management, CRM, tool is the most popular part of the suite but there is a lot more, if you want it). Part of this suite is Zoho Analytics [https://www.zoho.com/analytics/], which should now be considered as an Agentic Analytics platform, “a trusted intelligence layer for business insights and AI-driven action”.
I’ve talked to the AI people in Zoho in the past and I have high expectations of this new venture, partly because I like Zoho’s culture and focus on the business, and partly because of the lack of stock market-driven distractions from Zoho’s company ethos. I do think Zoho has experience in AI, since before it became an uber-fashion, and understands the importance of making AI solutions verifiable.
Now, I particularly like Zoho’s focus on Ontology and Semantics for its new platform, although the large number of data connectors available (well over 500, with 30 new ones just added) is part of the story too. In order to use AI effectively in the business, you need to understand what it is actually telling you – for a start, there are two broad categories of AI analytics, one based on advanced computational maths (which delivers trusted, unequivocal answers) and one based on large language models and similar, which can often prompt unanticipated innovation, but which can sometimes hallucinate dangerously unrealistic scenarios.
This second sort of AI can revolutionise the quality of the User Experience (UX) associated with existing tools and facilitate the generation of new tools, but there is a lot more to the use of AI in the business than simply trying different prompts until the AI supports the decisions you wanted to make in the first place – “The Computer Says YES” can be as big a problem as the “Computer Says NO” meme we used to laugh at in Little Britain [https://youtu.be/x0YGZPycMEU?si=MbNPdzyymQl8ff9r]. As Tony Thomas, CEO of Zoho USA, says: “Agents, however, cannot function only with data, but require a robust intelligence layer. Today’s updates to Zoho Analytics provide that and more: one open and extensible trusted intelligence layer that gives every AI tool in your business the context, metrics, and governance to deliver answers users can rely on”.
So, Zoho provides a framework – OK, a platform – in which AI can operate effectively. We have an Ontology Layer, which captures the business units in an organisation and, more importantly, how they relate to each other. According to Zoho (and it makes sense to me), this helps the AI to “generate insights and recommendations that are relevant and reliable, addressing real business
needs and ensuring results are tailored precisely to achievable outcomes and tied directly to process”. And then we have a semantic layer, which user can use to “produce a business glossary for AI that standardizes descriptions, synonyms, instructions, and metrics” (and Zoho supports different semantic models for different contexts). The need for semantics should become obvious when you think that large language models simply process the statistics of text without its meaning – they can’t know (unless told) that a fully accredited customer, with a credit rating and verified address, is not the same entity as someone called Micky Mouse who just visited your website once but still resides in a database called “customers”. Unless, perhaps, you ask the AI to tell you how hogenous the business’s idea of “customer” is, which is a human prompt.
Another Zoho enabler is compatibility with Apache Ossie [https://ossie.apache.org/], which is a vendor-neutral, open-source YAML and JSON specification that standardises how semantic data models, metrics, and business logic are defined and exchanged across analytics, business intelligence and AI systems. This ensures that an organisations metadata definitions are not locked inside of Zoho – the widest possible usage will help to ensure that they are kept up to date and complete. The Zoho Agentic Analytics platform also introduces the Zoho Analytics MCP server (available in remote and local options), which supports data analysis, data preparation, and visualization. A new command-line interface (CLI) is included, so users can work with Zoho Analytics from inside Claude, say, through a Claude plug-in. This all means that there is no danger of lock-in to the World of Zoho (unless, I suppose, a customer wants to build it that way; not really a good idea, I’d think).
Key to my heart is first-class support for AI governance, built in. A unified data catalog of data, metrics, analytics assets, and models gives organisations the opportunity for lineage
and impact analysis, with broad search options, and management of data and access policies. As long as, that is, organisations use them, this provides the basis for them to be able to demonstrate governance to third parties. Administrators can see which agents ran, who used them, what they accessed, and how they answered queries. AI usage and cost tracking across teams, along with administrative controls for Ask Zia (Zoho’s query engine), should help organisations monitor and manage AI adoption. At the operational level, teams can develop and test analytics artifacts and agents in a sandbox, publish to production when ready, and roll back to previous versions if necessary.
This is impressive, to me, but actually implementing effective governance isn’t trivial. It needs employees with governance skills, budgets for putting the effort into building governance, and (role-playing?) testing, with the sort of questions that internal and external stakeholders in governance might ask – and governance must be holistic, organisation-wide, you can’t achieve well-governed AI in a wilderness of ungoverned chaos. Still, without the sort of facilities provided by Zoho for its platform, you couldn’t really start building governance.
My initial impressions of Zoho’s Agentic Analytics platform are very positive. Do I have any concerns? Well the obvious concern, which is that everyone is mindlessly jumping on the AI bandwagon just now and many of these initiatives will fail to deliver on their promises, is mostly obviated by Zoho being a private company, unlikely to be over-influenced by the stock-market and its AI hype. There is, of course, a need for reasonably mature customers, themselves resistant to all the AI hype. One of Zoho’s competitors (well, a sort-of competitor) told me about one of their IT Director customers who’d had great success with using AI for coding – and now had a strategic objective for eliminating programmers altogether and having 100% of the organisation’s code written (and, presumably, tested) by AI. We both agreed that this was a pretty risky objective, not least because an organisation’s strategic objectives should really be focused in business outcomes, not tech wins.
In other words, given that I like what I’ve heard about Zoho’s new platform, I think that the biggest issue that Zoho might face is overenthusiasm in its customers for the AI Kool-Aid, coupled with a lack of customer interest in boring things like AI governance. Perhaps I should be taking another look at Zoho education for the new world of AI[https://help.zoho.com/portal/en/kb/zoho-learn/ai], but I have been pretty impressed by initiatives such as the Zoho University now Zoho Schools of Learning[https://www.zohoschools.com/] in the past and the Zoho customers I’ve met have always seemed pretty clued-up and well supported by Zoho, so I expect that Zoho is going to manage this risk OK.
I do think this new platform will be a success and I look forward to following its development.
Print Article for this blog Download here
