AI and automation
Design human review into AI workflows
Make the boundary between a generated suggestion and a real business action explicit in the product.
ExploreA WhatsApp assistant for a Bikaner business: what it can and cannot do
A realistic look at using an AI assistant on WhatsApp for a shop, a wholesaler, a coaching institute or a clinic.
The problem it solves
A business in Bikaner may get the same questions all day on WhatsApp: price, stock, timing, address, fees. Staff answer the same ten questions repeatedly, and a message that arrives at night waits until morning.
An assistant can give a first answer immediately, collect the details a person needs, and pass the conversation on. It does not replace the person.
What it can do well
Repeat questions
Timings, location, documents required, product availability from a sheet you maintain.
Collect details
Name, need, quantity and city, written as a short summary for your team.
Hindi and English
Modern language models handle both, and mixed Hindi and English too. We still test with real messages because spelling varies a great deal.
Out-of-hours replies
An honest "we will reply at 10 am" with the details already recorded.
Where it fails
It can misread a short message, repeat an old price, or confidently answer something it should not. That is why prices, discounts, refunds and confirmations should wait for a person. It also needs good source data: a stock sheet that is a week old will produce week-old answers.
How to start safely
Pick one queue, such as admission questions or product availability. Write what the assistant may say. Run it on past messages first. Read the log each week. Widen it only when the first queue is dependable. WhatsApp Business has usage rules and charges set by Meta, which we explain before you commit. See AI agent development and Bikaner.