IndiaMART moves B2B buying from search to decision with agentic AI

Nikhil Prabhakar, CIO of IndiaMART InterMESH, told CIO's Aryan Sharma on Nov. 2 that B2B buying is moving from a search journey to a decision journey. A buyer says what they need. The platform weighs the request, cuts the field to a handful of qualified, responsive sellers and checks their trust signals before anyone picks up the phone. In his telling, agentic AI turns software that answers into software that acts.

Prabhakar says IndiaMART is no longer a directory that returns results. Deciding on a buyer's behalf is a much higher bar, and the platform now carries trust work that used to fall on the buyer.

That shifts the pressure to the supply side. Many of IndiaMART's sellers are small businesses with no design, IT or analytics staff, and owners juggling sales, production and collections at once.

How the shortlist gets built

Prabhakar says the matching step weighs four things: the buyer's intent, how specific the category is, the order size and the supplier's capacity. What comes out is a short list of sellers. Before, buyers got a long results page they had to scroll, call through and vet on their own.

Trust checks have moved earlier, too. Due diligence used to start after first contact. IndiaMART now shows verification status, response rates and ratings up front, and the FE FuTech account adds payment protection to that list. Buyers have also moved off formal email to WhatsApp, in-app messaging and voice. That's where VANI, IndiaMART's voice AI agent, comes in: it handles more than 100,000 (1 lakh) buyer calls across languages and accents.

Letting an agent answer the enquiry

Prabhakar sees the biggest room to improve supplier productivity and conversion on the seller side. All those enquiries land on suppliers who are also running production and chasing payments, and following up on time is the hard part.

The current tools sit inside Lead Manager, where sellers handle their buyer conversations. They rank leads by buyer intent, suggest replies and offer one-tap actions to share a catalogue, send a payment reminder or answer from a template. The responsiveness gain in the figures above comes from the suggested replies and templates.

Next comes an agent that replies for the seller. It would study the seller's past Lead Manager conversations to learn how that business quotes, clarifies and follows up. Each seller would decide how much the agent can send without review.

Prabhakar was careful about how to score it. More automated messages don't make a seller more productive. IndiaMART's stated yardstick for the seller agent is fewer missed enquiries and more qualified conversations.

The seller-controlled dial is the detail sales-automation teams should study. A cautious supplier can start with drafts it approves and give the agent more room only as trust builds. That follows the order Prabhakar sets for adoption in general: assist first, then automate. The material doesn't cover the plumbing yet, such as how conversation histories are permissioned or which messages stay off-limits, and those choices will shape how much sellers actually hand over.

Catalogue data sets the ceiling

Prabhakar points first to data readiness: incomplete catalogues limit what agents can do. Confidence also matters. A business has to trust that the agent understands its products and stays within limits.

IndiaMART's response to the first barrier is to do the structuring for sellers, since most won't learn models or prompts. AI removes image backgrounds, crops photos and pulls out product attributes when a seller builds a catalogue. Prabhakar says the company is also working on standard products, where a seller picks the item and its specifications fill in automatically. On the buyer side, bots turn needs spoken in everyday language, including local languages, into structured leads. Sellers also get analytics on competition and pricing, the kind of intelligence Prabhakar says only large firms used to have.

Coveo's Jacqueline Dooley makes the general version of this argument in a November 2025 guide for B2B commerce leaders: agents fail when a site's data and design aren't structured for them. Her list of what needs structure starts with category taxonomy, specs, variants and pricing. It moves on to inventory, lead times and quotes, and ends with what she calls action model data: definitions of what an agent may do on the platform, including required inputs and permissions.

Set against that list, IndiaMART's catalogue tools look like groundwork for the matching engine. For distributors and manufacturers running their own B2B storefronts, the same logic suggests attribute completeness belongs on the commerce roadmap ahead of any agent pilot.

Repeat orders go first

Prabhakar expects standardized, repeat purchases to move to agents first, with human judgment still central in complex procurement. As agents take on more of the work, he argues, value shifts to the one thing they can't work without: verified business data.

Web Solutions NYC's Yitz Lieblich came to a similar conclusion in a May 2026 analysis. He argues B2B is where agentic commerce hits first because its purchases come with structured requirements, negotiated terms and repetitive workflows. In his replenishment example, an inventory agent picks a supplier on total landed cost, submits the order through the supplier's API and updates the ERP. A person steps in only when the order goes over a dollar threshold.

The strongest limit on the decision-journey idea comes from Prabhakar himself. Complex buying stays human, and small businesses still have to decide whether they trust an agent to quote for them. A shortlist engine is also only as good as the catalogues underneath it. Where listings are thin, the agent has little to work with, which is exactly the data-readiness barrier he names.

In consumer retail, Salesforce describes the move toward agentic search as a structural shift in behavior.

The next thing to watch is the Lead Manager agent. It will show how far IndiaMART's decision journey reaches beyond routine reorders.