Boltise Logo
Get Started
Back to all articles
WhatsApp AI · Commerce6 min read

WhatsApp AI Ordering for a Gujarat Dairy Brand: What Worked and What Broke

AP
Anil Patel
Founder & Systems Architect · Published Oct 2, 2026
Article cover · WhatsApp AI · Commerce1200 × 500

01. The Problem: Orders Lost in Chat

A dairy brand in Gujarat sells most of its products over WhatsApp. Customers write in Gujarati, often from villages with addresses that do not fit a standard form, and many of them ask questions in the middle of ordering. The team was answering hundreds of messages by hand, and orders slipped through between "how much?" and "paid".

02. What We Built

  • An AI agent with real tools: It can search the brand's knowledge, find products, check an order's status, check a PIN code and resend a payment button, instead of guessing from text alone.
  • A Gujarati address form: Customers pick their village from the official government village list, shown with Gujarati names, so deliveries stop failing on misspelt addresses.
  • Native WhatsApp payments: Orders end in WhatsApp's own review-and-pay flow, with one gentle reminder if a payment is left unfinished.
  • A clean hand-off: When the AI is unsure, or a payment is stuck, the chat moves to the team with a summary card showing what the customer wants and what the AI already tried.

03. Form vs Free Chat

We ran the structured order form and free-text chat side by side on the same day. The difference was large enough to change the default for everyone.

FlowStartedOrders placedPaid
Guided WhatsApp form311913
Free-text chat2491
Takeaway:

AI is best at answering questions. A short guided form is still the best way to take an order. Use both, and let each do its own job.

04. What Broke, and the Fix

  • Questions inside a form: A customer asking "when will it arrive?" while the bot waited for a village name was treated as the answer. Now each reply is checked first: real answers fill the step, questions go to the AI, and the step is asked again afterwards.
  • The AI provider went down: For two hours the AI could not reply at all. Now any AI failure hands the chat to the team with the usual message and sends the team an alert.
  • Silent failures: A database rule was quietly rejecting every hand-off card. Errors are now loud, and every new message type is tested end to end before release.

05. Built Once, Used by Every Client

Everything in this project was built into our own WhatsApp platform rather than as a one-off, so the AI tools, the address form, payment reminders and the team hand-off now work for every business we onboard. The next steps are UPI payments matched automatically to bank alerts, and a nightly review that turns unanswered questions into new knowledge for the AI.

AP

Written by Anil Patel

Founder & Systems Architect at Boltise

Anil builds AI agents, WhatsApp automation and developer tools for businesses across India, from Tharad, Gujarat.

Related content

View all articles

We use cookies

We use cookies to help this site function, understand service usage, and support marketing efforts. Visit to change preferences anytime. View our Cookie Policy for more info.