You posted the reel at seven. By nine it has done exactly what you hoped: people are watching it, saving it, and sending it to each other. It has also done the other thing a good reel does, which is fill your inbox. “Price?” “Is this in M?” “Do you ship to Indore?” “Is the blue one still there?”
You are on the sofa with your phone. You answer the first eleven. Somewhere around the twelfth you put the phone down, and by the time you pick it up again the thread has gone cold. The shopper who asked about the blue one has bought something else, from someone who answered.
This is the shape of selling on Instagram for most D2C stores in India. The content works. The questions arrive. And the one place a sale is decided, the DM, is the one place nobody is staffing.
Why the Inbox Is Where the Sale Happens
Indian shoppers ask before they buy. That is not a flaw in the audience, it is how buying a ₹3,500 shirt from a brand you found on a reel works. Fit, fabric, delivery time to a city the size chart does not mention, whether an exchange is possible if the size is wrong. A website answers some of this on a policy page nobody reads. Instagram answers it in a conversation, which is why the conversation is where people go.
So the DM is not a support queue that sits beside the shop. It is the shop floor. A question about size in a DM is the same moment as a shopper holding a shirt up in a store and looking around for someone to ask. If nobody comes, they put it back.
The trouble is that this shop floor is open at nine at night, on a phone, in batches, and hiring someone to sit in it does not scale with a reel that lands. Which is how most stores end up trying a bot.
Why Keyword Bots Fail
The first thing most stores try is a keyword trigger. Someone types “price” and a canned reply goes out with a link to the product page. It feels like automation, and for about one message it is.
Then the shopper replies. “Which link is for the blue one?” “I am usually an L but your last shirt was tight.” The keyword bot has no answer, because it never understood the question, it only matched a word. It sends the menu again, or nothing. Either way the shopper knows nobody is home, and a bot that reveals itself is worse than silence, because silence at least looks like you are busy.
“A shopper can tell within one reply whether anybody is home.”
The other failure is subtler. A keyword bot cannot remember that the shopper already said they wear L, and it cannot tell that “price?” under a sold-out product needs a different answer from “price?” under one you have in every size. Selling is mostly context. Keyword matching throws the context away.
What a Real Thread Looks Like
Here is a sample thread, invented for this post, with the mechanics exactly as the channel runs them. Everything in it is plain text or a card, because that is all Instagram allows.
A shopper writes: “hey is the linen shirt still there in L.” The assistant checks the live catalog and replies that yes, the Kora Linen Shirt is in stock in L, and adds that it runs a little relaxed through the chest, so someone who normally wears L is fine staying with it. That second sentence is the part a keyword bot could never write. It comes from the product data, and it answers the question the shopper was about to ask next.
Under that reply come two product cards: the Kora Linen Shirt at ₹3,499 and the Ridge Linen Trousers at ₹5,199, each with a Buy now button and an Add to cart button. The shopper taps Add to cart on the shirt. “Added. The Kora Linen Shirt in L is in your cart.” They tap it on the trousers too. “Added the Ridge Linen Trousers too. Here is where you are,” followed by a single cart card: Your cart, ₹8,698, with a Checkout button and a Returns button.
Then the shopper asks the question that decides it: “do you ship to indore, and what if the fit is off.” The assistant answers from the store’s published policies. Yes to Indore, usually in three to five days. Exchanges are open for seven days from delivery under the store’s returns policy, so a size swap is straightforward. Under that reply, a row of short suggested replies: Track my order, Sizing help, Shop new arrivals, and Talk to a human.
Notice what did not happen. The shopper never opened the website and never searched for the trousers: the assistant offered them because they go with the shirt. The cart total is the sum of the two products they tapped, drawn from live prices, not a number typed into a template. And there was no bullet list and no wall of text, because the assistant writes the way a good salesperson texts: short sentences, one recommendation, a reason, and a question back.
What Instagram Will and Will Not Let You Do
Most “AI for Instagram” demos show you a dashboard, because the thread itself is constrained and the constraints are not flattering. You will find them in week one anyway, so here they are.
Replies are plain text. There is no bold, no bullets, no tables, and a long answer has to be written as sentences and split into chunks. Product cards are the one piece of structure you get: a carousel of up to ten products, each with two buttons, and one cart card with at most three. Suggested replies are limited to a handful of short chips, and Instagram will not let chips share a turn with a carousel, so they arrive on the next message.
The assistant also waits before it speaks. People on Instagram send three messages where they would send one email, so the channel holds for about ten seconds and answers the whole thought rather than the first fragment. That means the first reply is considered rather than immediate, and anyone promising otherwise on this channel is either ignoring how people type or ignoring the platform.
Instagram does not hand over an email or a phone number, ever. Those arrive only when the shopper types them. It also gates a shopper’s name and picture behind extra review, so an empty profile in your dashboard is normal, not a bug. And when someone taps Buy now and lands on your checkout, Instagram sends no signal that they did. Whatever numbers you read from this channel are a floor, not a ceiling.
Finally, the assistant answers text, and a story mention. A photo of a shirt from another brand, a sticker, a voice note: none of that is understood, so it is left alone rather than answered badly.
The Honest Limits
Two more things a vendor is unlikely to volunteer. First, this runs in English. There is no translation layer, so if your customers write to you in Tamil or Marathi, do not plan a regional-language rollout on this channel yet. Second, the assistant is grounded in your catalog and your published policies, so the answers are only as good as what you have published. A store with a vague returns page will get vague returns answers, and the fix is on the page, not in the assistant.
And a person still matters. “Talk to a human” is a permanent option in every thread. Asking for it pauses the assistant for two hours while your team is notified, and if one of your team simply starts typing in Instagram the assistant notices and stays quiet for thirty minutes, so nobody is answered twice by two different voices.
How to Start
You need an Instagram business account, a store connected to MerchantIQ, and someone who can approve the connection. You connect the account once, choose whether the assistant handles DMs only or DMs and comments, and then rewrite the openers a first-time visitor sees so they read like your store rather than like software. Three are seeded for you and you can add a fourth.
From then on the same assistant that runs on your storefront answers the inbox, against the same catalog, the same policies and the same shopper record, so a question asked in a DM on Tuesday and on a product page on Friday gets one consistent answer. You can read how the channel works, edges included, on the Instagram DM automation page.
Conclusion
Your reels are already sending people to the place where the sale is decided. The question is whether anybody is there when they arrive. Answering the first eleven and letting the rest go cold is not a strategy, and a keyword bot that reveals itself in one reply is a worse one.
The inbox can be staffed, as long as the thing staffing it understands the question, knows your catalog, respects what Instagram will and will not let it do, and steps back the moment a person should take over. That is a sales channel. Treat it like one.