Why Zoho’s Zia Chat could redraw the enterprise AI playbook
Ask a sales manager which deals closing this month have gone cold, and the answer isn’t in one place. The pipeline is in the CRM, the open tickets are in the support desk, and the overdue invoices are in the finance system. Getting the answer means three logins, a few filters and a spreadsheet, and by then the moment to act may have passed.
At a press conference alongside its annual Zoholics user conference in Chennai, Zoho framed the problem in one line: the information exists, but the context is fragmented. Its answer is Zia Chat, an AI interface that lets people ask that question once, in plain language, and get a single response drawn from across the business. The launch came with Zoho’s entry into vertical software, with seven industry-specific applications.
The launch is worth a CIO’s attention because of how it differs from the standard enterprise AI playbook of the past two years. It changes what AI is expected to know, where it sits, what it costs and how far it is trusted to act.
The old playbook and where it stalls
Zoho breaks the everyday friction of enterprise work into four parts, and each will sound familiar:
Scattered information. Data sits across CRM, support, finance, projects, email and collaboration tools.
Knowing where to look. The business question comes first, and finding the right app comes second.
Connecting it manually. People switch tools, filter, and piece the story together themselves. Stopping before action. Even after the answer is found, the next step happens in another tool.
The standard response has been to add an AI assistant to each application. That helps inside a single app, but it leaves the cross-system questions unanswered, which are usually the ones that matter. Co-founder Sridhar Vembu calls the underlying issue an “interface crisis” and adds that fragmented context means AI only ever sees part of the business.
Rewrite one: From smart models to smart context
Zoho describes AI maturing in three stages. Generic AI understands language. Application-aware AI understands individual applications and their data. Industry-aware AI understands entities, rules, workflows and relationships across the business.
Zoho’s position is that AI can act effectively only when it understands the industry, the workflow and the business context. That is a shift in emphasis from model intelligence to context. Two people can put the same question to the same model and get very different answers, depending on what the system knows about who is asking, which customers, deals and invoices are related, and what the rules are.
Zia Chat is built natively into Zoho’s applications, where identity, permissions and business relationships already exist. Zoho says it uses semantic search, a “work graph”, resolvers and relationship traversal to work out how customers, deals, tickets and transactions connect, so answers are grounded in the business and not just the prompt. It works across the applications, modules, fields and records a user is authorised to access, and doesn’t need a separate integration configured for each Zoho app.
Rewrite two: from app-by-app assistants to one layer
Zia Chat is a single conversational interface across sales, marketing, support, HR and finance. Users ask questions, and it retrieves and connects information, summarises records and conversations, analyses data, and produces reports, charts, spreadsheets and presentations. Then it acts: drafting proposals and follow-ups, creating and assigning tasks, sending low-stock alerts and scheduling recurring reports.
Zoho’s own scenarios show what crossing systems means in practice:
– Sales: find which closing deals have gone cold, have open tickets or have overdue invoices, and get a prioritised list ready to act on.
– Support: see a customer’s full picture, including history, open issues, active projects and recent conversations, before picking up the phone.
– Marketing: compare a campaign against the previous one on leads, conversion rate, cost per lead and best segments, without pulling a single report.
– Finance: find overdue invoices, check for active deals in negotiation and draft a payment reminder, with context from CRM and action in Books.
– Operations: spot products likely to run out within 30 days by checking inventory, sales velocity and open orders, then get a reorder recommendation.
– Founders and managers: get a monthly snapshot of revenue against target, ticket trends, team workload and churn signals.
This isn’t entirely new ground for Zoho. Its Zoho One suite already offers Zia Search, a unified search across business data that respects employee permissions. A Zoho executive said in 2018 that it had been in development for more than seven years. Ask Zia later pulled emails, chats, documents, CRM data, tasks and tickets into one view. What Zia Chat adds is conversation, action and reach beyond Zoho. That progression, from finding information to understanding it to acting on it, is the substance of the change.
Rewrite three: from integrations to native context plus MCP
The traditional approach to cross-system AI is to build connectors and pipelines for every pairing of systems. Zia Chat takes two routes instead. Inside Zoho, native embedding means the plumbing is already there. Outside it, Model Context Protocol (MCP) integrations let organisations bring in information and capabilities from the wider technology environment, while Zia Chat keeps its deeper native integration with Zoho applications.
This won’t remove integration work, and for non-Zoho systems the quality of the experience will depend on what each MCP integration exposes. But it changes the starting point for each new use case.
Rewrite four: from open-ended agents to governed ones
The most common enterprise objection to AI that acts is trust. Zia Chat’s answer is to build the rules into the software: It can only access the information and actions available to the user. If you can view a record but not edit it, neither can Zia Chat.
Actions that need approval stay with the authorised approver. Users review and confirm actions before anything is created, changed or sent. Responses can link back to the underlying records, so users can verify where an answer came from. Data accessed by Zia Chat isn’t used to train Zoho’s AI models.
Vembu has described the design as AI for work, with guardrails and a trust layer in software to complement the AI.
Making Zia Chat fit the job: profiles, skills and agents
Zoho lets companies tailor Zia Chat in three ways. A useful comparison is a new employee joining the sales team. They get a role, a set of playbooks, and tasks they can hand off.
A profile decides what Zia Chat focuses on for a given job, and Zoho’s examples span Sales, Marketing, Finance, Customer Support Operations and Executive Leadership, each limited to the applications and data that function actually needs. Imagine a finance manager asking, “Who is paying us late?” Under a Finance profile, Zia Chat goes straight to invoices, payments and subscriptions in Zoho Books. If the CEO asks a similar question under an Executive profile, the answer might instead show overdue amounts alongside revenue and pipeline, so the picture covers the whole company rather than one function’s slice of it.
Skills are the playbook: a skill packages a specialised task so it runs the same way every time. Customer 360 works as a pre-call briefing, so before phoning a client a rep can ask who they’re about to speak to, and Zia Chat pulls purchase and subscription history, recent tickets, open issues, promises colleagues have made and the customer’s mood, replacing four browser tabs and ten minutes of digging.
Similarly, Ticket 360 does the equivalent for a single support ticket, giving a support lead the full thread, the service-level clock, earlier tickets from the same contact and a suggested next step in one view. Deal Analyzer reviews a single deal, covering stage history, competitors mentioned and risks, with recommended next actions, while Stale Deals Audit handles pipeline clean-up by surfacing deals with no movement, overdue close dates or missing next steps, ranked by value at risk so the biggest problems come first.
Agents take this further: where a skill answers a question, an agent finishes a job with several steps. The Lead Enrichment Agent adds company size and industry when a new lead arrives, spots that the company is already a customer, merges duplicates and fills in missing fields before a rep ever sees it. The
Lead Routing Agent then assigns that lead by territory, product line and rep capacity, and re-routes it if nobody has responded within the first-response window. The Lost Deal Analysis Agent reads the full history of a deal that fell through — how long it sat in each stage, competitor mentions, pricing discussions — and writes a root-cause summary, while the Follow-up Email Drafter drafts a post-demo email based on what was actually discussed, attaches the right material, and holds it until the rep approves it.
Put together, a single lead’s journey could look like this: a prospect fills in a web form, the enrichment agent completes the record, and the routing agent hands it to the right rep. That rep asks for a Customer 360 briefing before the first call, then approves the drafted follow-up afterward. Nobody has copied data between systems or chased the lead.
A note on these examples: the scenarios above are illustrations built from Zoho’s own feature descriptions. They are not customer case studies. The skills and agents named are examples Zoho showed at its conference, and the company has not said which will be available at launch or how much configuration each will need. Treat them as a picture of the design and not a confirmed feature list.
Rewrite five: from token-hungry to token-efficient
For CIOs, the question behind every AI pilot is what happens to the cost at scale. Vembu addressed it directly. Zoho’s goal, he said, is to commoditise intelligence so customers can get work done without incurring heavy token costs. He said the company is working to cut the number of tokens AI workloads need, use smaller or local models where possible, and support customers who bring their own model providers. He also said Zoho wants to make this intelligence available at an attractive, affordable price, cheap enough to bundle into its software instead of treating AI as a standalone revenue stream.
The logic holds up. When software handles retrieval, permissions and actions, the language model only has to reason over a small, relevant slice of data instead of a large dump of context. Zia Chat is also model-agnostic, so organisations choose which models power it and can switch as capabilities, cost and performance change, while the business layer stays consistent. On Zoho’s agent platform, a team that connects its own Anthropic key pays nothing to Zoho for that usage and is billed directly by the model vendor.
Rewrite six: from horizontal breadth to vertical depth
The final change is the reason the launch came with seven industry applications. Zoho’s view is that business context is shaped by the rules and regulations of each industry, so AI gets better when it sits on top of industry-specific software.
The new vertical offerings span retail (RetailIQ, a CRM for physical retail), automotive (AutoDMS, dealer management), banking and financial services (Zoho LOS, loan origination), healthcare (MedScribe, ABDM-compliant clinical documentation), education (Zoho Campus, initially in India), real estate and restaurants. Restaurants billing up to ₹12 lakh a year on Zoho POS can use it free. Zoho says the combination of suite breadth and industry depth gives AI a richer picture of how a business really runs.
Why Zoho can try this
Vembu’s wider thesis is that software is becoming a commodity as AI lowers the cost of building it. If each piece is a commodity, the value lies in combining everything into one system with AI on top, and in keeping pace with regulations and bringing structured business data to the AI. Zoho is unusually well placed to run that experiment. It is privately held and profitable, with more than 60 applications, over 150 million users and more than 19,000 employees, and it runs its own data centres. Its India revenue grew 47% in 2025, according to the company.
The bottom line
For two years, the enterprise AI playbook has been to pick the smartest model, bolt it onto each application and hope the bills stay manageable. Zoho is arguing for something different: put a permission-aware layer of business context in front of any model, let it answer across systems, keep humans in charge of actions, and make token efficiency a design goal. Vembu sums up the idea as “UI is the new AI”, meaning intelligence present everywhere, with safeguards built in.