WhatsApp AI Agent for Business in India: What It Can and Can't Do

The strongest use cases are high-volume questions whose answers already exist in a system. A dealer in Nagpur asks whether 200 units of a SKU are in stock at the Bhiwandi godown. A retailer asks when last week's order will be dispatched. A distributor wants the current scheme on a product line. Today each of these lands on a sales coordinator's phone, and the coordinator checks Tally or the ERP and types back a reply.

An agent can do that lookup and reply in seconds, at any hour, in the language the dealer wrote in. Language is less of a hurdle than it used to be. Meta says its own Business AI on WhatsApp works in native Indian languages, and custom agents can be set up to reply in Hindi, Marathi, Gujarati or the Hinglish most dealers actually type.

Quotations are the next step up. An agent can read an enquiry, match the items to your price list and dealer-specific terms, and prepare a draft quotation for a salesperson to approve and send. The salesperson's job shifts from typing to checking.

Some decisions should stay with people. Approving credit beyond a dealer's limit, giving discounts outside policy, handling a quality complaint or a payment dispute, and committing to a delivery date the system cannot confirm all belong with a human.

An agent also cannot answer from data it cannot reach. If dispatch status lives in a supervisor's notebook, or stock is updated in Tally only at the end of the day, the agent will give stale answers with full confidence. Fix the data source before you switch on automated replies for that question.

Write these limits down before you build. A clear list of what the agent answers, what it drafts for approval and what it always hands over is the most useful design document in the whole project.

Meta now offers its own option. On 14 May 2026 it introduced Business AI on WhatsApp for small businesses in India, working in native Indian languages, with UPI payments inside the chat on the way. TechCrunch reported on 3 June 2026 that Meta Business Agent had launched globally after about two years of testing in India and Mexico, and afaqs reported that it was announced at WhatsApp's India Business Summit in Mumbai.

The early results from Meta's partners are encouraging. According to afaqs, Madhulika Enterprises reported 25–30% more conversions, with 50–60% of its repetitive workload handled by the agent. If most of your questions are about the catalogue, store timings and general product information, Meta's own tool is the sensible place to start.

A custom agent earns its place when the answer depends on your internal systems. Dealer-specific pricing, stock by godown, outstanding balances, dispatch status from the ERP and quotations built from your own rate cards all need a connection to Tally or the ERP, plus rules about who is allowed to see what. That integration work, along with the approval and handoff logic around it, is what a custom build adds.

Handoff is where WhatsApp automation either earns trust or loses it. The agent should hand over when a message falls outside its list, when the dealer asks for a person, or when it is unsure of the answer. The person taking over should see the whole conversation and the data the agent looked up, so the dealer never has to repeat themselves.

Route by type. Payment questions go to accounts, complaints go to the service team, and large orders go to the regional sales manager. Outside office hours, the agent should say clearly when a person will reply, and the queue should be waiting for the team when they log in the next morning.

Review the handed-over conversations every week. They show you which questions the agent could take on next, and which ones should stay with people for good.

Start with order status. It is high volume, the answer sits in one system, a wrong answer is easy to correct and the time saved is easy to count. Pick 20 or 30 friendly dealers for the pilot, run it for four weeks, and track how many queries the agent closes without help.

Before you build anything, export a month of WhatsApp chats from your sales coordinators and sort the messages by type. Most companies find that a handful of question types make up the bulk of the traffic. That list becomes your scope.

Once order status runs well, add stock checks, then quotation drafts with human approval. Each step reuses the same connection to your ERP, so later workflows are quicker to add than the first. Joistic builds custom agents connected to the systems a business already runs, and if you want to map your WhatsApp traffic to a first workflow, a 30-minute call through /contact is a good place to begin.