Ten questions, a hundred times
Delivery, sizing, stock, returns, opening times. Typing those out personally is not customer service, it is transcription — and the better a post performs, the more of it there is.
Meta approved partner
Instagram customer support, automated
The same ten questions arrive a hundred times a week, and the answers are already written on your own site. An agent handles those in under a second, shows where each answer came from, and hands the rest to a person with the thread attached.
Free to start · No credit card · Official Meta API
Hey, do you ship to Ireland?
We do! Orders to Ireland arrive in 3–5 working days, and you’ll get tracking as soon as it leaves us.
northlane.co/shippingWhat’s your return window?
You have 14 days from the day it arrives. Anything unworn with the tags still on can go back, and returns are free within the UK.
northlane.co/returnsMy order turned up damaged :(
I’m sorry about that. I’ve passed this to the team with your order details and someone will come back to you to sort a replacement.
handed to a personTwo of these never reached a human. The third was tagged and assigned to someone who can fix it.
Why automate this
Support is the work that grows with your reach, not your revenue
A post that does well does not bring you more orders so much as more questions, and they are almost always the same ones. Three things make this worth handing to an agent, and only one of them is about what time it is.
An answer you can trace
Every reply names the page or document behind it. Automation you cannot audit is a liability waiting to happen; an answer with a source attached is one you can check and correct.
Open when you are not
Evenings and weekends are when people browse and ask. A reply in a second at midnight beats a considered one at nine, because by nine they have asked a competitor the same thing.
The play
What has to happen between a question and a resolved thread
Support automation nearly always fails in the same place. It answers the easy thing confidently, and then has nowhere sensible to put the hard thing. All four stages have to exist for it to be safe to switch on.
northlane.co · 47 pages
returns-policy.pdf · 9 pages
Shopify · 212 products
Stage 01 · the sources
It is only as good as what you point it at
You are not writing answers from scratch. You are pointing the agent at the ones you have already published, and policy pages beat marketing pages every time.
- Your site, your store, or a PDF
- Refunds, shipping and sizing first
- It answers from those and nothing else
Do you ship to Ireland?
Yes — 3–5 working days.
ⓘ northlane.co/shipping
Stage 02 · the answer
Fast is table stakes. Traceable is the point.
Any tool can reply quickly. What makes an automated answer safe to send is that you can see which page it came from, so a wrong answer is a document to fix rather than a mystery to investigate.
- Under a second, any hour
- Cited to the page it came from
- One agent across Instagram and Facebook
Where is my order?
Getting a person on this now.
→ assigned · tagged order status
Stage 03 · the boundary
Knowing what not to answer
Anything about one specific customer — where their parcel is, whether they get a refund — is not in your policy pages and never will be. The agent does not improvise. It stops.
- Outside its sources, it does not guess
- Order and refund questions go to a person
- Tagged on the way, not just forwarded
RKRavi K.Open
Assigned · order status
Agent handled 24 before this
Stage 04 · the queue
A human picks up where it left off
Handover is where most support automation becomes worse than none, because the customer is made to start again. Here the thread arrives whole, with what was answered and why it escalated.
- Assignment, tags and tickets
- Labelled agent, workflow or person
- Everyone who messaged is a contact too
Most tools stop after stage two. They answer the easy question well, then answer the hard one just as confidently, because nothing was ever built to catch it.
What you get back
You can see everything it said, and step in whenever you want
Every conversation lands in one queue, labelled with what is driving it — an agent, a workflow or a person. Nothing happens in your DMs that you cannot read back, which is the difference between delegating support and losing sight of it.
Threads carry their full history, so anyone picking one up has the context already. And everyone who messaged becomes a contact with whatever the conversation collected, which means a support queue quietly builds the same list a lead form would.
The agent
What the support agent can be told to do
An agent is exactly as good as what you point it at, and exactly as safe as the boundary you draw around it. You set both.
Answer from your own sources
Your website, your Shopify store, or a PDF such as a size guide, price list or returns policy. It answers from those and nothing else, so it cannot invent a returns window you never published.
Show where every answer came from
Each reply names the page or document behind it. When something is wrong, you fix the source rather than hunting for what the bot decided — and you can check its working before a customer acts on it.
Stop at what it should not answer
Order status, refunds, anything about one named customer. None of that lives in your policy pages, so the agent does not attempt it. The boundary is the feature, not a limitation.
Escalate with the context attached
The person who takes over sees the whole conversation, what was already answered and why it stopped. Nobody asks the customer to repeat themselves, which is what usually makes handover worse than no automation at all.
Cover every channel from one place
The same agent handles Instagram DMs and comments alongside Facebook Page comments and Messenger, set up once rather than twice. WhatsApp is coming to the Business plan at no extra cost.
Answering, and knowing when not to
Custom AI agents are on the Business plan. See how agents are trained →
Proof
Brands that let the DM answer first
“Augst AI became our first-line customer support on Instagram. It replies faster than any human, and it's never off duty.”
FAQ
Questions about support automation
01What if it gives a customer the wrong answer?
It answers only from the sources you gave it and names the one it used, so a wrong answer points at a page that needs correcting rather than at a black box. Fix the source and every future answer changes with it. You can also read back every reply it has sent from the Unibox and step into any thread yourself.
02What can the agent be trained on?
Your website, your Shopify store, or a PDF such as a rate card, size guide or policy sheet. It answers from those sources only and cites the one each answer came from. Custom AI agents are on the Business plan at $99/mo.
03What happens when it does not know?
It stops rather than guessing. Anything outside its sources — an order problem, a refund decision, anything about one specific customer — becomes a thread for a person, tagged and assigned so it is not lost. On the Advanced plan the Unibox gives you that queue with assignment, tags and tickets.
04Can I see everything it has said?
Yes. Every conversation sits in one queue and every reply is labelled with whether an agent, a workflow or a person sent it, so there is nothing happening in your DMs you cannot read back.
05Can a person take over mid-conversation?
At any point, and the thread arrives with its full history so the customer is not asked to start again. That handover is usually where support automation falls down.
06Which channels does it cover?
Instagram on every plan and Facebook from Starter — Page comments and Messenger alongside Instagram DMs and comments, with one agent across both. WhatsApp is not live yet; it is coming to the Business plan at no extra cost when it ships.
07Is automating replies allowed under Meta's rules?
Yes. Augst AI is an official Meta partner and everything runs on the official Instagram API, within Meta's guidelines for chat automation. There is no unofficial automation of your logged-in account.
Related
Your team should only ever see the hard ones
Point an agent at what you have already written, and see how many questions never reach a person at all.
No card · Live in minutes