What legal AI actually costs once the research layer is in

By Michael Nadalin, Founder, Market Lead · 28 July 2026 · 7 min read

Say the quote lands at $99 per user per month. Twelve fee earners, so a shade over $14,000 a year, which reads as roughly one month of a junior solicitor. The partner running the numbers signs it off in about four minutes.

Then the practical questions start. The tool drafts a submission and cites three authorities. Someone has to confirm those authorities exist, that the passage says what the summary claims, and that the case has not been distinguished into irrelevance since. That checking happens somewhere, and in most Australian firms it happens inside a research subscription that was already costing real money before the AI arrived.

The honest version of the budget is not the licence fee. It is the licence fee, plus whatever you must keep running underneath it to trust the output, plus the time your people spend closing the gap between a confident answer and a usable one. This is how to build that number before you sign, not after.

The quoted price is the smallest number in the deal

Software pricing in legal tends to be quoted the way a flight is quoted: the fare, then everything that makes the fare functional. Per-seat AI pricing follows the same shape. The number on the proposal covers access to the model and the interface. It rarely covers the primary law, the citator, the commentary, or the integration into your practice management system.

The gap matters more in legal research than in most software categories, because the output is worthless until it is verified against a source you would be prepared to cite in the Federal Court. An AI answer you cannot check is not a time saving. It is a task you have moved earlier in the day.

Everything sitting underneath the AI licence

Australian firms rarely run legal AI on its own. It sits on a stack, and most of the stack predates the AI decision. Working out what you actually pay means listing the layers and marking which ones the new tool lets you switch off. If the answer is none of them, the AI is additive spend, not a substitution, and it needs to earn its keep on top of everything you already carry.

AustLII and Jade cover a great deal of primary material at no cost, and plenty of good practice runs on them. What they do not replace for most firms is annotated commentary, currency assurance across every jurisdiction you touch, and the citator work that tells you whether an authority still stands. That is the layer the AI vendor is usually assuming you keep.

A worked example you can rebuild with your own quotes

The proportions below are illustrative, not survey data. The point is the shape: in a typical Australian firm running AI on top of an existing research subscription, the AI licence is not the dominant line, and the two largest lines are the ones nobody put on the proposal.

Rebuild it with your own numbers. Take your current annual research spend, your quoted AI licence including the seat minimum, an honest estimate of verification hours at your average charge out rate, and any one off integration fee amortised across the term. The exercise takes about twenty minutes and it usually changes the decision, sometimes in favour of the tool.

The costs that arrive later, not at signing

Most of the unpleasant surprises in legal software are timing surprises. The pilot price is genuine, and so is the renewal price. They are just different prices, and the gap between them is disclosed in a clause rather than on the proposal.

Read the term, the notice period and the uplift mechanism before the pricing page. A CPI linked uplift is reasonable. A clause allowing an uplift at the vendor's discretion on thirty days notice, in a contract that auto renews unless you give sixty days notice, is a structural problem regardless of how good the tool is.

Questions that change the number before you sign

Vendors answer these questions readily when asked directly. The trouble is that a demo is built to show capability, not cost structure, so nobody asks. Send the list in writing and ask for written answers, because the answers become the thing you rely on if the invoice does not match the conversation.

One question does more work than the rest: what am I able to cancel because of this. If the answer is nothing, the tool has to justify itself as pure additional spend against saved hours, which is a much higher bar than the demo implies.

What actually makes the maths work

There are two honest ways a legal AI purchase pays for itself. The first is substitution: the tool retires a subscription, a service, or a category of outsourced work, and the ledger visibly improves. The second is throughput on work you are already charging for, where drafting and first pass research move fast enough that the same people carry more matters without the file quality dropping. Anything else is a preference, which is fine, but it should be budgeted as a preference rather than a saving.

The layer that decides which applies is the one underneath. A tool that reads from a subscription you already hold makes the AI cheap and the stack expensive. A tool that licences and hosts Australian primary law itself, as Legal Brain does, changes what sits on the ledger, and that is worth testing against the specific subscription you are hoping to reduce rather than against the category in general.

Practically: ask the vendor what you can switch off, put the answer in writing, price verification time at your real charge out rate, and diary the renewal notice date before the trial starts. Several Australian courts have now issued practice notes and guidelines on generative AI use, and complying with them takes supervision time that belongs in the same budget. A firm that does all of that will usually still buy the tool. It will just buy it at a number it recognises when the second year invoice arrives.

Frequently asked questions

Do I still need a LexisNexis or Westlaw subscription if I use a legal AI tool?

It depends entirely on where the tool gets its primary law. Some AI products read from a subscription you hold separately, in which case the subscription is mandatory and the AI is additive spend. Others licence and host Australian legislation and case law themselves, which is the only situation where the research line can genuinely come down. Ask the vendor directly and get the answer in writing before you assume a saving.

Why is legal AI priced per user when only some staff will use it?

Per seat pricing is standard, but the practical constraint is the minimum seat tier rather than the per user rate. Firms commonly buy a tier well above their initial user count because the next tier down does not exist, then add graduates and paralegals who push them to the tier above. Ask for the tier boundaries, not just the headline per user figure.

How do I budget for the verification time an AI tool creates?

Estimate the number of outputs going to clients or courts each week, multiply by the minutes a senior solicitor needs to check every citation and passage against the source, and price it at your average charge out rate. That figure is often larger than the licence fee in year one and falls as people learn the tool's failure patterns. A tool that links each assertion to a source passage reduces this line materially.

Is free primary law from AustLII enough to verify AI output?

For a lot of everyday work, yes, and many capable Australian practitioners rely on it. Where it usually falls short is annotated commentary, currency assurance across multiple jurisdictions, and citator depth for checking whether an authority has been overruled or distinguished. If your practice carries appellate or complex regulatory work, budget for a paid citator regardless of what AI tool you choose.