Instagram Automation for an Online Shop, With the Flows Written Out

An online shop’s Instagram inbox is unusually predictable. Four questions account for most of it — price, size or variant, shipping, and whether a thing shown three weeks ago is still available — and all four have fixed answers that somebody currently types by hand.

This is a playbook rather than a feature list: what to build, in what order, and where automation makes the experience worse rather than better.

Automation one: the product post to Direct route

This is the highest-value automation for a shop and the reason most shops arrive at this category.

You post a product. People comment — “price?”, “link”, an emoji, a friend’s handle. A rule answers each of them in Direct with the price, the link and one question that invites a reply.

Three decisions make the difference between this working and being noise:

• Name one keyword in the caption and only one. “Comment SIZE for the size chart”. Multiple keywords per post split the instruction and halve compliance.
• Reply publicly as well. Every failed private reply is invisible unless you post a public comment saying the DM was sent. This is worth more than any wording improvement inside the DM.
• Send the product, not the catalogue. A person asking about one jacket does not want the shop homepage. A deep link to that product converts several times better, and it is the same amount of work if you attach the rule to the post rather than using one generic rule everywhere.

The mechanics are in the comment-to-DM guide.

Automation two: the four standing questions

Read your last fifty Direct conversations and count. For most shops the distribution is roughly: price, shipping cost and time, size or variant availability, returns.

Each of these is a lookup, not a conversation. Automate all four, and write the answers to a strict format:

Shipping is €4.90 anywhere in Spain, free over €60, and usually arrives in 2–3 working days.
Want me to check if something is in stock?

Answer first, next step second, nothing else. The method for choosing these is in building an FAQ bot from the questions you already get.

Automation three: the story mention reply

Shops get tagged. Someone unboxes the thing, photographs it, tags you, and the story lives 24 hours. Most shops answer with a heart the next morning, by which time the story is gone.

An automatic reply that arrives while the story is live can do two things worth real money: ask permission to reshare (free content, and the asking reads well), and offer something small and specific for the next order. The detail is in story mention auto replies.

Automation four: the out-of-hours answer

Shopping happens at night. A message at 22:40 with no reply until 10:00 the next day is a conversation that has usually moved on, and for a weekend closure the 24-hour window can expire before anyone opens the inbox — at which point you cannot answer at all.

An after-hours rule that answers the four standing questions and states a specific reopening time converts most of that loss into a delayed sale rather than a lost one.

What should a shop never automate?

Three things, and the third is the expensive one.

• “Where is my order?” This requires a lookup in your actual system, and a generic “orders usually ship in 2 days” to someone whose parcel is nine days late escalates the complaint. Route these to a person with the order number captured.
• Refunds and damage. An automated message about a broken item is the thing customers screenshot.
• Stock confirmations you cannot actually verify. Telling someone an item is available when it is not produces a cancelled order and a bad review. If the bot cannot see the stock level, it should say “let me check” and mean it.

How does this handle a post that goes viral?

Badly, unless you planned for it, and this is the scenario worth rehearsing because it is also the one you want.

A Reel with 600 comments produces roughly 600 private replies in the hour it peaks. Three things break in that hour:

1. Your cost, if your tool bills per contact. Six hundred contacts created in an afternoon can move you two pricing tiers for the month, before you know whether any of them bought.
2. Your inbox, if it is a plain list. Six hundred new threads with no filters, no assignment and no search is not workable, and the team goes back to answering on a phone.
3. Your stock answer. If the item sells out at comment 200, comments 201 to 600 are getting a promise you cannot keep. Have a plan for switching that rule off quickly.

The first is a vendor choice, the second needs a real shared inbox, and the third is a person watching.

Where does an AI agent fit in a shop?

Between the fixed answers and the human. A rule answers “shipping”; an AI agent trained on your size guide, your shipping page and your returns policy answers “I am usually a 38 in Zara, what should I take in this and would it get to Valencia before Friday” — which is the question that actually precedes a purchase.

Two constraints that make this safe in commerce specifically: the agent must answer only from your material, and it must not confirm stock or delivery dates it cannot verify. “I will check that with the team” is a correct answer and should be an explicit instruction, not a hope.

What to measure

• Share of conversations where a human typed anything. Should fall. If it does not, the automation is decorating the inbox.
• Time to first human reply on the conversations the bot could not handle. Should also fall — if it rises, attention has been absorbed rather than saved.
• Comments per Reel that trigger the rule, divided by total comments. Low ratios mean the caption instruction is unclear, not that the audience is bad.

Chatonaut connects Instagram through Meta’s official login, keeps comments, Direct and story mentions in one inbox, and includes an AI agent trained on your own material. Plans are on the pricing page.