Most small law firms do not fail at AI because the technology is hard. They fail because adoption starts as a curiosity, one partner testing prompts on a quiet Friday, and never becomes a system the whole firm can trust. Sixty days is enough time to move from scattered experimentation to a defensible, firm-wide way of working with Claude. This plan is written for the managing partner who has to answer to other partners, protect client confidentiality, and show a return, all at the same time.
It assumes a firm of roughly three to fifteen lawyers, the size where there is no dedicated IT team and the managing partner is also the de facto technology lead. It leads with Claude because Claude handles long legal documents well, keeps a careful and precise tone, and can be configured so your client data stays out of model training. Anthropic, the company behind Claude, is the credibility tag here, not the headline. What matters to your firm is the workflow, not the vendor.
Why a plan beats a pilot that fizzles
A pilot with no end date drifts. People try a few prompts, get a mixed result, and quietly go back to how they worked before. A 60-day plan fixes three things a loose pilot never does: it sets guardrails before anyone puts client data near a chatbot, it picks the workflows most likely to pay back first, and it ends with a real decision about whether to scale. You are not asking the firm to believe in AI. You are running a short, measured trial with a clear scorecard.
The stakes are specific to legal practice. Get confidentiality wrong and you have a conduct problem. Rely on an unchecked citation and you risk misleading the court. The point of the plan is to capture the upside while keeping both of those risks in a box you control.
Days 1 to 15: Set the guardrails before anyone types a prompt
Spend the first fortnight making the firm safe to experiment in. This is the part firms skip and later regret.
Confidentiality and privilege: confirm that your Claude plan does not train on your inputs, and write a one-line rule that no client-identifying material goes into any tool that has not been approved. Your duties under the Australian Solicitors Conduct Rules do not change because the drafting is faster.
Court expectations: read the generative AI guidance issued by the courts you appear in. Judges in New South Wales and Victoria have set out how AI may and may not be used in material filed with a court, and those directions are not optional.
Data and privacy: map what client information could plausibly end up in a prompt, then check it against your Privacy Act 1988 obligations and the Notifiable Data Breaches scheme. Decide now what is off limits.
A named owner: appoint one lawyer as the AI lead for the trial. Adoption without an owner becomes nobody's job by week three.
By day 15 you should have a single page that says what Claude may be used for, what it must never touch, and who to ask when someone is unsure. It does not need to be a polished policy yet. It needs to exist.
Days 16 to 30: Prove value on three real workflows
Now put Claude to work on tasks that are real, repeatable, and low-risk if a first draft is imperfect. Resist the urge to test everything at once. Pick three.
First-pass document review: summarising long contracts, leases, and briefs, and flagging unusual clauses for a lawyer to check. Claude reads a 60-page agreement in seconds and gives your fee-earner a running start.
Correspondence and file notes: turning rough notes into clear client letters, attendance notes, and internal memos written in the firm's voice.
Research and discovery support: organising and questioning large document sets, always with a lawyer verifying anything that will be relied on.
Track the time each task took before and after. Ask the assigned lawyer to keep a simple log. You are collecting evidence, not impressions, because in 30 days you will use it to make a decision the other partners can accept.
Set one rule during this phase that no one may break: every citation, figure, and legal proposition that leaves the firm is checked by a person. Claude drafts and accelerates. It does not get the final word. That single rule is what separates firms adopting AI safely from the ones that end up in an embarrassing judgment because a citation was invented.
Days 31 to 45: Write the policy and train the team
With two weeks of real use behind you, turn the one-page guardrails into a proper AI use policy. It should cover approved tools, confidentiality, the human-verification rule, how to handle client consent where it is relevant, and what to do when something goes wrong. Keep it to two pages a busy solicitor will actually read.
Then train everyone, not just the early adopters. A 90-minute session where each lawyer runs their own real task under guidance does more than any slide deck. The goal is that a cautious senior associate and a keen graduate both leave able to use Claude on Monday morning. Adoption is a people problem far more than a software one.
Show the three approved workflows using the firm's own matters, with client details removed.
Demonstrate what a good prompt looks like, and what an over-trusting one looks like.
Make the verification rule concrete with a real example of a wrong answer stated confidently.
Days 46 to 60: Measure, decide, and scale
The final fortnight is about the numbers and the decision. Pull together the time logs, the quality feedback from the lawyers involved, and any partner concerns. Answer one question plainly: is the firm getting back more than it is putting in?
The economics for a small firm are usually clear once you look. One lawyer billing at $400 an hour who recovers just three hours a week returns roughly $55,000 a year in capacity. Across a five-lawyer firm, even conservative time savings comfortably clear $180,000 a year in recovered fee-earner time. Against that, a Claude plan at around $30 per user per month, plus a one-off setup of about $3,500 if you bring in help, is a small line item. Even after allowing for the extra verification time, a careful trial should pay for itself many times over.
If the answer is yes, scale deliberately. Roll the approved workflows out to every fee-earner, add the policy to your induction pack, and set a quarterly review so the rules keep pace with both the technology and the courts. If the answer is no for a particular task, drop it without ceremony and keep the ones that worked.
What this costs and what it returns
A realistic 60-day budget for a small Australian firm is modest. Expect software of a few hundred dollars across the trial, some partner and lawyer time for setup and training, and optionally a fixed-fee setup engagement around $3,500 if you would rather not build the guardrails and training from scratch. The larger cost is attention: the managing partner has to care enough to keep the plan on track for two months.
The return is not only recovered hours. Firms that adopt carefully report faster turnaround on client work, less late-night drafting, and junior lawyers spending more of their time on judgement and less on mechanical first drafts. Those are the things that keep good people and win repeat instructions.
Where to start
If you are a managing partner weighing this up, the cheapest first move is to block out the opening fifteen days for guardrails and name your AI lead. Everything else follows from there. If you would like a second set of eyes on the plan, or help tailoring it to your practice areas and the courts you appear in, book a brainstorm session with Automata AI and we will map the first 60 days with you.
