How to Build an AI Roadmap for a Law Firm in One Week

A law firm can build an AI roadmap in one week. Map your biggest jobs step by step in a four‑hour working session, score every step for safety, capability and return, then put the work in order with what stays human written down.

That is how we run the AI Blueprint, and it ends with a 12‑month plan your partners own. You can use the same method yourself, so here is how each part works, using Marlow & Finch Lawyers, the fictional 18‑person firm in our sample Blueprint. Every figure from that firm is made up for the example.

What should an AI roadmap for a law firm answer?

A useful roadmap settles four questions your partners would otherwise keep debating.

1. Which job do we start with, and why that one?
2. What are we choosing not to build?
3. What stays with a person, whatever AI can do?
4. In what order does the rest happen over 12 months?

Day 1: how do you map the work?

Day 1 is a four‑hour working session at your office. Before it, the firm sends rough numbers: roles, hourly costs, volumes and the systems you use.

In the room you want four to six people: the ones who do the work, and the one who decides. The people doing the work know the steps that never made it into a procedure. The person who decides means the priorities are settled before anyone leaves.

Each big job is then written out step by step: who does it, how, and in which system. At Marlow & Finch, a property file from signed contract to archive came to 12 steps and 6.4 hours of file work on each matter.

How do you score each step?

Every step gets three scores.

• Safety. Is it safe to hand to AI?
• Capability. Can AI do it well today?
• Return. What is it worth, at the firm's own rates?

Those scores sort each step into one of three groups: AI does it and a person checks, AI prepares and a person decides, or it stays human. Here is how three steps on that property file came out.

• Key dates diarised. Safety 8, capability 9. AI does it, and a person checks.
• Settlement figures calculated. Safety 7, capability 8. AI prepares, and a person decides.
• Settlement attended and authorised. Safety 2, capability 5. It stays human.

A low safety score keeps a step with a person, whatever AI could do. Moving trust money scores 2.

The return then follows a plain rule. When AI does a step, all of its time comes back. When AI and a person share it, 75 per cent comes back. When it stays with a person, none does. On that file, 4 of the 6.4 hours came back on each matter.

Days 2 to 5: how does it become a plan?

Over the rest of the week, every step is costed and each job is drawn as a map. The plan is put in order, with what stays with your people written down. The firm answers a few follow‑up questions on volumes and timings.

The week ends with the finished Blueprint walked through on a call, where the partners pick the first system to build.

What goes on the "what not to build" list?

This list settles the arguments before they get expensive. Marlow & Finch's list has two kinds of entry.

• Never. Anything that moves trust money or lodges with a court, and advice from AI. The partner gives the advice, and AI prepares the file.
• Not yet. Intake automation, until enquiry volumes and conversion are measured. A coordinator joining the systems up, until each one has passed its checks.

Each thing you do build gets its own stop rule, written before the build starts. For that firm's first system it reads: stop if a draft loses its sources, or checking it takes as long as writing it.

Should you start with the biggest number?

Usually, and sometimes the proof matters more. Marlow & Finch's biggest figure was advice and document production, at about $262,000 a year. They took it second. They started on matter delivery, worth about $170,000, for three reasons: the firm's strongest precedents, the clearest bottleneck, and a draft the senior associate could judge in an afternoon.

A quick, visible win with work your team can check earns the trust for the bigger one. Their year runs like this.

1. Months 1 to 3. Matter delivery: property file documents drafted from the file, with every statement traced to its source.
2. Months 4 to 6. Advice and document production, using the same drafting pattern.
3. Months 7 to 9. Client intake and engagement.
4. Months 10 to 12. Join it up, once each system has passed its checks.

Each phase starts once the one before has passed its checks, and nothing is built until the firm accepts a separate proposal for it.

How can you start this week?

Pick one matter type your firm does often. Ask the person who does the work to write out every step: who does it, how long it takes and in which system. Then put a score out of ten for safety beside each step, and see which ones your partners would never hand over.

That one page gives your partners something real to talk about. To choose the first task from it, use the three questions in what a law firm should automate first. When you want the whole firm mapped, AI strategy for law firms sets out how the steps fit together.

Bart Puszko

Founder and Principal of Blue Seas AI. Queensland Government AI mentor. Top 10 finalist, Australian AI Awards 2025 & 2026. Before AI, 16 years in financial crime, risk and continuous improvement for major banks across Australia and Europe.