Most AI for Law Firms Drafts the Contract. Your Firm Leaks at Intake

It is six o'clock and the principal of a three-partner firm is back at her desk after court, doing the part of the day nobody bills for. Opening the matter from this morning's enquiry. Running the conflict check. Typing the file note before she forgets it. Chasing the client who still has not signed the costs agreement. She is a good lawyer. None of that is law. It is admin wearing a lanyard.

That is the picture behind most searches for AI for law firms. Not a robot barrister. An owner trying to claw back the hours that leak between the billable ones. Almost everything written on the subject answers a different question. It points you at legal research and document drafting, the work you already charge for, and calls that a strategy. That work is real and worth a look. It is also the crowded, riskiest end of the problem, and it is not where a small firm actually bleeds.

What AI for law firms actually changes in your firm

Split it in two, because pretending it is one thing gets firms to buy the wrong thing.

The first half is the billable craft: research assistants, contract review, a first draft of an advice or a clause. Useful in places. It is also the work you already invoice, and it carries the most risk. You have seen the headlines: lawyers hauled up for filing submissions that cited cases their chatbot invented. Anything that touches your advice or your pleadings needs a human reading every line, which caps how much time it actually hands back.

The second half is front of house and file admin: intake, conflict checks, engagement letters, chasing an unsigned costs agreement, tracking disbursements, the file note that never got typed. This work was always mechanical. It just never had anyone spare, so it slipped. That is the half no drafting demo touches, because it is specific to how your firm runs a matter.

Start with intake, not the courtroom

Pick one workflow, not the whole practice. For most firms the highest-value start is the one that runs at every enquiry and gets dropped most: intake.

An enquiry lands at half past seven at night. Instead of sitting in an inbox until someone is free tomorrow, it gets an immediate, professional reply, collects the details, runs a first-pass conflict check against your existing matters, and sets the file up in your practice management system ready for a fee earner to open in the morning. The client feels looked after on day one. No partner loses an evening to data entry. I pulled apart how that front-desk automation works, and where it must hand back to a person, in what an AI receptionist actually does for a small business.

The firm that answers the enquiry first usually keeps the client. The rest are still drafting a reply.

One workflow also gives you a number to defend. If two enquiries a week go cold because nobody replied fast enough, and each is a matter worth real money, recovering even half of them is a result, not a vibe.

The unbilled hours are the real leak

Every hour a fee earner spends chasing an unsigned agreement or retyping a file note is an hour that is neither billed nor billable. Firms treat that time as the cost of running a practice. Most of it is a communication gap you can close.

A rule that follows up the unsigned costs agreement on schedule, in your firm's tone, and nudges the client for the missing document recovers days a month that were leaking quietly. The same logic runs the file: turning a rambling client email into a dated file note against the right matter, flagging the limitation date nobody diarised, prompting for the ID a new client still has not sent. None of that is the clever part. All of it is the part that keeps people back late.

The drafting tools automate the work you already bill for. Your losses live in the work you can never put on an invoice.

What AI for law firms costs to build and run

The bands are real. A per-seat legal research or drafting subscription is cheap enough to trial on one lawyer. An intake-and-admin workflow built to fit your firm and connect to your practice management system usually lands between a few thousand and the mid teens of thousands of dollars to build and go live. A connected setup across several systems starts higher again, and is rarely where a firm should begin.

The figure that catches owners out is not the build, it is the running cost: the subscriptions underneath, the model usage itself, and the staff time still spent reviewing output and handling what the system escalates. I broke those layers down in what AI actually costs a small business. Match the spend to the task. Never pay for a custom build where a setting in software you already own would do the job. A vendor whose every answer is their priciest option is solving for their invoice.

The Australian layer: your PMS, trust accounting, and client duty

Two things separate AI that works in an Australian firm from a generic overseas template.

The first is integration. You run on LEAP, Smokeball, or Actionstep, and your trust accounting is not a detail you let a chatbot near. A tool that cannot read and write your real matter list has not removed the double handling, it has moved it to a paralegal. Check the integration, and keep anything automated well clear of the trust account, before you fall for the demo.

The second is duty. The moment a system touches a client file it is handling privileged and confidential information, under the Privacy Act, your confidentiality obligations, and legal professional privilege. Before anything connects to real client data, get plain answers: is it stored in Australia, is it used to train someone else's model, and can you delete it on request. A vendor who cannot answer that has just told you how carefully they build.

How to not get sold the wrong thing

Run any pitch past three questions before you sign. Which task does this remove, and how many hours or lost matters is it worth. Does it connect to your practice management system cleanly, and stay clear of the trust account. When the work is done, do I own the system and the client data, or does it stop the day I stop paying. A lawyer knows a straight answer from a hedge. Point that instinct at the vendor.

Common questions about AI for law firms

What should a law firm automate first?

Intake and client chasing, not legal research. That is where the unbilled hours and the cold enquiries live. Prove the hours and matters recovered on that one workflow, then widen. Automating five things at once means you cannot tell which one paid off. I wrote about starting narrow and proving it before you widen in most AI automation for small business automates the wrong task.

Will AI replace lawyers or paralegals?

No, and firms treating it that way aim it at the wrong target. It clears the intake, the chasing, and the file admin so your people can do the advising and appearing that clients actually pay for. It does not do the judgement, and on the billable side it still needs a human checking every line, because the court holds you responsible for what your AI wrote, not the vendor.

Is a custom build worth it over an off-the-shelf tool?

Only once the off-the-shelf tools are on and you have hit their limit. The custom value sits in the intake-and-matter workflow that fits your firm and connects to your software, and only if you own the system and the data at the end.

If you want a straight read on which task to fix first, and whether it needs a custom build at all, that is what a first call is for. Book a strategy call and bring the part of your week that keeps a fee earner back after the billable work is done.