AI for Migration Agents Fixes the Afternoon. Files Stall for Months

Count the active files in your practice that nobody has touched in thirty days. Not the hard ones. The ones waiting on a client to send a declaration, a translation or a bank statement.

That number is where your practice is stuck, and almost none of the AI sold to migration agents looks at it.

Take Ravi and Ana. They signed the agreement for services and fees for their partner visa in February. Four years together, a strong relationship on any reading. Their agent is good at the law, and the submission would have taken her an afternoon. It was lodged in August.

Nothing went wrong on the legal side. Two statutory declarations from friends, a certified translation of a birth certificate, a year of joint bank statements: every request went out in one long email in February. The couple did the easy items first, then life got busy. The assistant meant to follow up, then a bridging visa problem took the week. Nobody owned the gap, so the gap grew.

Ravi and Ana are a composite of the files I hear about from principals: one or two registered agents, an assistant, a few hundred active clients and an inbox that never empties.

The law is the part of your practice that already works. The files stall in the waiting.

Where the AI pitch is pointed

Submission drafting, research assistants and policy summaries are the pitch right now, and I will be fair to them. Turning a client's history into a clean relationship timeline is real work, and a tool that gives you a first draft to check can save an evening.

But notice what it touches: the hours you spend once the evidence is in. In most practices I see, that is not the bottleneck. The bottleneck is the weeks before it, when the file sits at "waiting on client" and nobody looks at it until the client rings to ask why nothing has happened.

Why visa files stall before lodgement

A visa file is a relay, and the law is only one leg of it:

enquiry → consult → fee agreement → evidence chase → drafting → lodgement → requests from the department → decision

The leg in bold is the one with no owner. The agent is drafting or on the phone. The assistant is handling today's urgent file. The client has a forty-item checklist in an email from three months ago, and no idea which item is holding everything up.

Meanwhile, evidence goes stale while it waits. Bank statements that were recent in February are old by August, and a relationship that has moved on needs fresh evidence, which means asking the client twice. That is the conversation that makes a client feel their agent is disorganised, even when the advice was flawless.

A file waiting on the client is still your file. The client does not know it is waiting on them.

Run the numbers on your own stalled files

Treat this as arithmetic you fill in yourself.

Say you have two hundred active files, and forty of them have been sitting at "waiting on documents" for more than a month. Say your agreement holds back fifteen hundred dollars of each fee until lodgement. Forty times fifteen hundred is sixty thousand dollars of invoices you cannot send yet. If you take the whole fee upfront, the cost shows up differently: as complaints, refund requests and clients who stop referring their friends.

Every number there is invented and yours to replace. Open your practice software, filter active files by last client contact, and count the ones nobody has touched in thirty days. If the number is small, good: you have just saved yourself a project.

What AI for migration agents should do first

Fair question: why AI, and not a task reminder? A reminder says "follow up Ravi". It does not know which three of forty items are still missing, which one is slowest to get, or how to ask for it in plain words a stressed client will act on. A model can read the checklist, the file notes and the client's last emails in seconds and draft the next request. A reminder rule cannot.

So the system watches every file between engagement and lodgement, and keeps one status per client:

• It reads what has arrived and marks what is still missing.
• It flags the slowest items first: anything that depends on a third party, a translator or a government office overseas.
• It drafts a short, specific chase for the assistant to send, one item at a time, not the full checklist again.
• It warns you when evidence already received is getting old enough to need refreshing.

Every morning the assistant gets one list, most urgent first:

• Ravi and Ana: second statutory declaration outstanding 41 days, last chased 19 days ago
• Silva, M: employment reference received, unsigned and undated
• Okafor, J: birth certificate received, certified translation not yet ordered

Nothing goes to the client on its own. The assistant reads the draft, fixes what needs fixing, and sends it.

Measure two things each month: median days from signed agreement to lodgement, and the number of files waiting on documents for more than thirty days.

Automate the chasing, never the advice

The line here is not negotiable. Visa advice is the registered work, and OMARA's guidance on the use of artificial intelligence is clear that you stay responsible for the advice even when a tool helped produce it. Which visa, whether the evidence is strong enough, how to answer a request from the department: that is your judgement, and your registration sits on every word.

A system that asks a client for a missing declaration is doing admin. A system that tells a client whether they qualify has stopped being admin. I build the AI. You keep the business, and the advice stays with the person who is registered to give it.

What AI for migration agents costs

Work down this list and stop at the first rung that works.

• The features already in your practice software. Checklists, client portals and automated reminders are often included, half set up. Turn them on properly first.
• A simple routine. One person, thirty minutes a day, working a "waiting on client" list by hand, one item per message. If lodgements speed up, you have proved the value before spending a cent.
• A custom build that reads your files and client emails, ranks the chases and drafts the messages. A few thousand dollars if it only reads one system, up to the low teens of thousands if it must read your practice software, your inbox and a document portal and write back to all three, plus a modest monthly running cost.

I broke down where the running costs hide in what AI actually costs a small business. If you also run a legal practice, the same pattern of matters stalled outside the legal work shows up in AI for law firms.

The Australian layer

Three cautions before any client data touches a tool.

First, confidentiality. The same OMARA guidance warns that putting client personal details into an AI tool can breach the Code of Conduct. Get the client's consent in writing first, and ask any vendor where data is stored, whether it trains someone else's model, and whether you can delete it on request.

Second, privacy. Passports, birth certificates and relationship evidence are about as sensitive as personal information gets under the Privacy Act. Keep them out of general purpose chat tools entirely.

Third, tone. Many of your clients are anxious, and some are reading English as a second or third language. A short, specific, kind message gets the document back. A wall of legal text gets ignored.

Common questions about AI for migration agents

What should a migration practice automate first?

The evidence chase. It runs on files you have already signed, and one number tells you inside a quarter whether it worked: days from signed agreement to lodgement.

Can AI give visa advice to my clients?

No. A message that says "we still need your second declaration" is admin. A message that says "you should be fine" is advice, and it needs your eyes and your name on it.

The file that waited

Go back to Ravi and Ana. Their case was strong in February. It sat in an inbox waiting on a declaration nobody asked about twice.

If you want to know how many of your files are stuck in the waiting right now, that is what a first call is for. Book a strategy call and bring a list of your active files with their last contact date. We will find the stalled ones before we talk about building anything.