Most stores set up Instagram automation on a Sunday afternoon, switch on six triggers at once, and spend the following week unable to tell which one produced the reply that embarrassed them.
This is the slower version. It takes four weeks, most of which is waiting and reading, and it ends with something you will still be running when the festive season arrives.
First, the three surfaces and the clock each one runs
A store touches customers on Instagram in three places, and they are not interchangeable. Getting them confused is the single most common reason a setup underperforms while looking correct.
A comment on a product post gives you seven days. A story reply and a direct message both put you inside the twenty-four hour rule, counted from their last message and restarted only when they send another one.
For a store the practical reading is this. Posts are your durable surface — a Reel from last week is still generating conversations now. Stories are your fast one, good for drops and restocks, dead by tomorrow. DMs are where the money is and where the clock is cruellest.
Week one: templates only, on one post
Turn the model off entirely. Write four replies by hand.
Pick the product that gets the most comments, not the one you wish got the most. Read the actual comments under it and take the word people already use. If half of them type "price" and half type "rate", your trigger matches both, because matching ignores case and punctuation and you should be matching on the word appearing anywhere in the comment rather than the whole thing.
The five fields, in the order the builder asks:
- Trigger. The post, and the words. Add negative keywords immediately — "not interested", "worst", "scam". A store that automates replies to angry comments has automated the wrong thing.
- Public reply. Six words under the post. "Sent you the details in DM." It is a receipt, not the answer.
- Private reply. The actual answer, with the price in it. Written the way you speak to a customer, not the way you write a policy page.
- Capture to Leads. On. This is the part that pays for the month.
- Test on myself. Before Go live, always. The full pipeline runs and the result arrives on your own account.
Templated replies are free and unlimited, so week one costs nothing at all and tells you what your inbox is actually made of.
Week two: read what got stopped
Open Activity, filter to blocked, and read it like a store report rather than an error log.
Ten checks run before every send, in a fixed order, and the first failure is recorded against the run. For a store the reasons cluster in predictable ways:
- A lot of closed windows means your trigger is catching comments on posts older than seven days. Normal, and mostly harmless.
- A lot of duplicates means Instagram redelivered the same comment. Also normal.
- A lot of takeovers means you are answering by hand faster than the automation. That is a signal to narrow it, not to widen it.
- Rate budget hits mean a post travelled and the hourly cap turned a spike into a queue. That is the cap doing its job on the day you most need it doing its job.
Week three: add the second surface, and let the model in
Now add stories. A drop announcement with "reply DROP for the link" feeds the same inbox from a direction most stores never set up, and the story trigger is the one almost nobody configures.
This is also the week to turn on generated replies, for the questions no template fits. Credits are spent only here, one credit per reply the model writes, and the free tier gives you a hundred of them without a card. For a store, that is roughly a fortnight of the genuinely unusual questions — the fit queries, the "can you do this in a different fabric", the customer with a problem.
Keep the templates underneath as fallback. If the model is slow or unsure, your own words go out instead of nothing.
Week four: quiet hours, caps, and then stop touching it
- Quiet hours in your customer's timezone. Around 9pm to 8am IST is what most Indian stores settle on. A price reply at 3am reads as a bot even when the wording is perfect.
- First-time senders only, if you post daily. It stops the same regular follower being greeted under every reel.
- The hourly cap. Set it before a post travels, not during. Reels do not warn you.
Then leave it alone for a month. Continuous tuning is the failure mode of week four — you change three things, something improves, and you have no idea which change did it.
A month of data on an untouched setup is worth more than four weeks of adjustments you cannot attribute.
What this plan deliberately does not include
No cart recovery sequence, because the version everyone sells does not exist on this API. No multi-day nurture drip, because the twenty-four hour rule means nobody has the window a drip assumes. No posting or scheduling, because this replies to people, it does not publish for you.
And no catalogue sync. The replies know what you wrote in them and nothing about your stock levels, which is why the availability template should point at the product page rather than promise a number.
If you want the scope of the whole thing in one page before committing four weeks to it, start here. If your volume is coming mostly from Reels rather than static posts, that funnel behaves slightly differently and is worth reading alongside week one.
The plan works because it is boring. One automation, tested on yourself, live on one post, read for a week. Everything after that is repetition with a wider net.