Most selling advice for Instagram is written as though the platform were a shop window with a checkout attached. For a lot of businesses — especially outside the Shop-enabled markets — it is closer to a market stall where the actual transaction happens in conversation, and the conversation is governed by rules you did not write.
Two of those rules shape everything below.
You cannot start a conversation. Somebody has to message you, or comment, first. And once they have, you may reply to a direct message for twenty-four hours from their last message, restarted only by another message from them.

What those two rules rule out
Worth stating early, because a lot of published advice quietly assumes otherwise:
- No cold outreach. There is no compliant way to message a stranger.
- No follow-up sequence. A message on day three cannot be sent unless they wrote on day two or three.
- No abandoned-cart recovery. Somebody who left a cart on your website has not messaged you on Instagram, so there is no permission and no event.
- No broadcast to past buyers. Same reason.
If a tool offers those, it is either doing something unofficial with your login or describing an ads product bought elsewhere.
What it rewards instead
A channel where you cannot chase people rewards being genuinely useful in the first reply. That is the whole strategy, and it is unglamorous.
- Make the ask specific. "Comment PRICE and I'll send the list" produces a comment. "Link in bio" produces a scroll.
- Answer the literal question first. The price, the size, the delivery time — in the first sentence, before anything else.
- End with one closed question. Their answer is the only thing that restarts the clock, so make answering take one word.
- Capture them either way. Whoever asked is worth keeping whether or not they buy today.
Say the price
The most common piece of bad advice in this category is to withhold the price and "move them to a call". In a DM that reads as evasion, and it ends more conversations than it starts.
If you have a price, give it. If it genuinely varies, say what it varies with and give the range you are confident about. A real answer with a caveat beats a deflection every time, and it saves you the message that would have been spent asking again.
Build in this order
- One automation, on your best-performing post. The single question you answer most. Nothing else.
- A negative keyword list, before the second automation.
refund,cancel,not received,payment failed, plus the phrasings your audience actually uses. These stop the automated reply entirely and leave the message for you. - Test on yourself. It runs the whole pipeline and delivers to your own account, so you read the reply as a customer would.
- Go live on that one post, and read the blocked list a day later.
- Only then, the second automation.

Where the sale actually happens on each surface
The three surfaces behave differently enough that treating them as one channel is the commonest planning mistake.

A comment under a Reel is the widest net and the slowest conversation: it is public, so the reply is read by people who did not comment, and you have seven days to move it into a DM. That makes Reels the place to ask, and the public reply a shopfront rather than an answer.
A story reply is the opposite. It is already a DM, it is invisible to everyone else, and the twenty-four hour clock started the moment they sent it. Stories are where the conversation is quiet enough for a real question, and where a slow answer costs the most.
Direct messages that arrive unprompted are the warmest and the rarest. Somebody went looking for you. Almost nothing about those should be automated beyond a first fact, because the person has already done the hard part.
Pricing the automation itself
The cost shape matters when you are deciding how much of this to build. Templated replies are free and unlimited, so answering the same four questions ten thousand times costs nothing. A credit is spent only when the model has to compose something new, which means the meter tracks how unusual your inbox is rather than how large it is.
For most accounts selling a small catalogue, that lands well: the repeated questions are free and the odd genuine question is cheap. For a business whose enquiries are all bespoke, it is the expensive shape, and worth knowing before you commit rather than after.
The handoff nobody plans
At some point the conversation stops being answerable by anything written in advance — a negotiation, a custom order, someone who is upset. The moment you reply by hand, the automation stands down for that thread and stays down. You do not switch anything off.
That is the intended ending, not a failure. The job of everything before it is to make sure the person is still there, still inside the window, and already has the facts they needed.
What to count
There is no analytics product here, no attribution, no revenue reporting. What you can count, by hand, in ten minutes a fortnight:
- How many people asked something.
- How many got an answer inside the window.
- How many conversations you personally took over, and how many of those became orders.
That last number is the only one that has ever mattered, and no dashboard was required to get it.
If you are starting from an inbox you cannot face, triage first. If your question is what the reply should actually say, that is the craft post.