WhatsApp chatbot for ecommerce: what to automate and what not

Every online store hears the same five questions all day. Where’s my parcel. Can I change the address. Is this in size M. Did my order go through. Can I return it.

A WhatsApp chatbot for ecommerce takes the first few off your team’s plate and passes the rest to a person with the context already written down. That is a good deal. A bot that tries to win every argument is a bad one.

The four flows a WhatsApp chatbot for ecommerce should run first

Skip the clever stuff. These four cover most of the volume in a typical Indian D2C store.

1. Order status. Customer sends “where is my order” or taps a button. The bot looks the order up and replies with the status and tracking link.
2. COD confirm or cancel. A cash-on-delivery order triggers a message with two buttons: Confirm, Cancel. This is the cheapest protection you have against orders that bounce back from the courier.
3. Product questions. Stock, size, colour, price. A short menu or an AI agent answering from your own catalogue and policy pages.
4. Handoff. Anything the bot can’t settle goes to a human in the shared inbox, with the chat attached.

Returns and refunds are the fifth flow people ask for. I’d keep that one human-led until the first four run clean. Money questions deserve a person.

Buttons first, free text second

WhatsApp gives you structured replies for exactly this job. Meta’s Cloud API documents reply buttons (up to 3 per message, with titles of up to 20 characters) and list messages (up to 10 rows in total, row titles up to 24 characters). You’ll find the limits in Meta’s interactive reply buttons documentation.

So a menu looks like this:

Hi Aarav, welcome to the store. What do you need?
[Track my order] [Change address] [Talk to a person]

Three buttons. Tap, done. People answer buttons far more reliably than they write a clean sentence, and the bot never has to guess. For a longer menu (product categories, say), use a list message and keep it under 10 rows.

How the flows connect to your store

This is where most guides go vague, so here’s the specific picture for WaMark.

• Shopify and Ecwid connection. WaMark sends order confirmations, shipping updates, cash-on-delivery confirm-or-cancel messages and abandoned-cart reminders from your store events. You don’t build that by hand.
• Orders list. Orders sit in one list, with GST tax invoice PDFs and credit notes.
• Shipping. If you ship with your own Shiprocket or Delhivery account, tracking updates go to the customer on WhatsApp.
• No-code chatbot builder. Buttons, conditions, webhooks and Google Sheets steps. It can also call your own CRM or any REST API in the middle of a chat, which is how you look up a customer or raise a ticket without leaving WhatsApp.
• AI agent. It answers from your own documents and URLs, reads the whole chat before it replies, and hands over to a person when asked.

The practical split: automatic messages go out when something happens in your store (order placed, shipped), and the chatbot answers when the customer writes in. Both live in the same inbox, so your team sees the whole thread.

Want a sense of the broader tooling first? The WhatsApp chatbot guide for small businesses covers the basics, and the AI WhatsApp chatbot post explains how the AI side works and what Meta allows.

COD confirmation: write it so people answer

Cash on delivery is common in Indian ecommerce, and it’s also where a store loses money on dispatches that no one receives. A confirm-or-cancel message sent soon after the order catches the “I ordered by mistake” and “wrong address” cases before the parcel leaves.

A message that works:

Hi Meera, we got your order #1042 for ₹1,499 (cash on delivery). Delivery to Pune 411001 by Thursday. Please confirm so we can ship it today.
[Confirm order] [Cancel order]

Keep three things in it: what they ordered, the amount they’ll pay at the door, and where it’s going. Customers who recognise their own details tap Confirm. Customers who don’t, cancel, and you save the shipping charge.

I’m not going to promise you a drop in returns. Your mileage depends on your category, your pincodes and how fast the message goes out. Test it for 2 weeks on one product line and compare the return-to-origin count before and after.

Where the bot must stop

A decision rule I use: if the answer can be looked up, the bot answers. If the answer needs judgment or money changes hands, a human takes over.

Make the “Talk to a person” button visible in every menu. Hiding it makes people angry, and angry customers write one-star reviews and report your number.

The AI side has a rule from Meta as well. Meta’s updated Business Solution Terms bar general-purpose AI chatbots (assistants that answer anything) from the WhatsApp Business Platform from 15 January 2026. A store bot that answers questions about its own products and orders is a different thing, but I’d read the terms yourself and keep the AI tied to your own documents. The terms also bar using WhatsApp Business Solution data to train AI models.

Worked example: what a bot saves

Take round numbers. They’re assumptions, so swap in your own.

• You ship 300 orders a day.
• 1 customer in 3 asks “where is my order” at least once: 100 chats a day.
• A person takes 3 minutes per chat to look up and reply.

100 × 3 minutes = 300 minutes, or 5 hours of agent time every day. If a bot with a tracking link settles 70% of them, you win back 210 minutes a day. Over a 26-day month that’s 210 × 26 ÷ 60 = 91 hours.

At an agent cost of ₹150 an hour (assumed), that’s ₹13,650 a month. The Growth plan, which is where chatbots and the AI agent sit, is ₹6,999 a month before GST, or ₹8,258.82 with 18% added. On these assumptions the time saving covers the software with about ₹5,391 to spare, and customers get faster replies as a bonus. At 40 orders a day the sums don’t work, and the automatic shipping messages from the Shopify or Ecwid connection are enough on their own.

Mistakes that wreck an ecommerce chatbot

• A menu with 9 options. People tap the first or the last. Keep three buttons on the front screen and push the rest behind a list.
• No fallback. When the customer types something the bot doesn’t understand, it should say so once and offer a person. Looping the same menu five times is how a chatbot earns a bad reputation.
• Promotions inside service flows. A customer asking where the parcel is doesn’t want a coupon. Keep marketing in opted-in campaigns.
• Stale data. If the bot reads from a Google Sheet that someone updates on Mondays, it will tell people the wrong status on Thursday. Pull from the live store where you can.
• Night-time silence. A bot’s biggest win is the 11 pm “is this in stock” question. Write a clean answer for it, or at least a promise of when a person will reply.

Fixing these costs nothing but an hour of reading real chats. Do that once a week for the first month.

Set it up in this order

1. Week 1. Connect the store. Turn on order confirmation and shipping updates. This alone answers a large part of “where is my order”.
2. Week 2. Add the COD confirm-or-cancel flow on one product line.
3. Week 3. Build the menu with 3 buttons and a handoff. Read the first 50 real chats and fix the wording.
4. Week 4. Add the AI agent on your return policy, size guides and FAQs. Keep refunds with humans.

The time to start is before the festive rush, when message volume jumps. If you’re planning a sale, read the Black Friday WhatsApp plan next, then see how the WaMark features fit together.