Skip to main content

How Should a Law Firm Start With AI? (Hint: Not With Client Letters)

Professional reviewing documents with an AI-assisted workflow

A small law firm should start with AI on internal work an attorney can check: document and contract review, first drafts of research, search across the firm's own precedent, and back-office tasks like intake and billing. Keep AI away from anything that reaches a client without review. That order builds skill and trust before AI goes anywhere near the client relationship.

If you're a managing partner or practice lead, you're getting pitched on AI constantly. Every legal tech vendor has an AI story, the bar associations are running panels on it, and associates are already using it on their own. The pitch is usually efficiency: faster research, lower overhead, more output per hour. What doesn't come up is the risk profile, the malpractice surface area, or the fact that AI is particularly confident when it's wrong.

I spent years working inside law firms as an AI advisor. Here's what I saw work, and where firms got hurt.


Where does AI actually work in a law firm today?

Document review and contract redlining. This is probably the strongest use case in legal AI right now. Reviewing large volumes of documents for specific terms, flagging deviations from standard language, identifying missing clauses: AI handles this well, because it's pattern-matching over structured text. The key word is review. An attorney still confirms what the AI flagged, but the time savings on due diligence and contract analysis are real.

Research drafts as a starting point. AI can surface relevant cases, summarize holdings and draft an initial research memo faster than any associate. What it can't do is verify its own citations with certainty, or catch the nuance that decides whether a case applies in your jurisdiction. Used as a starting point, it's genuinely useful. Used as a final product, it's a liability.

Knowledge management and precedent search. Most firms have years of work product sitting in folders nobody navigates well. AI search across internal documents, past briefs and standard templates is one of the most underrated applications in legal. It doesn't require trusting a model with anything client-facing. It just makes the firm's own knowledge easier to find.

Internal process automation. Billing workflows, conflicts checks, intake processing, deadline tracking. These are places where AI and automation reduce friction without touching professional judgment. The risk is lower and the upside is real, so this is typically where a firm should start.

UseWho checks the outputRiskGood first project?
Intake, conflicts, billing follow-upStaff who run the process todayLowYes, once the process is written down
Precedent and internal document searchThe attorney using the resultLowYes
Document review and redliningReviewing attorneyModerateYes, with review built in
Research memosAttorney verifies every citationModerate to highAs a first draft only
Client letters and status updatesAttorney, before anything is sentHighNot until the firm has experience
Jurisdiction-specific answers from a general modelHard to check without redoing the workHighNo

Where do law firms get burned with AI?

Sending AI output to clients without review. Any AI output that goes to a client (a letter, a summary, a status update) needs a human in the loop before it leaves the firm. I've seen firms automate client communications to save associate time, then spend that saved time managing the fallout. Clients notice when something feels off, and in a relationship business, "felt off" is enough to start them looking elsewhere.

Trusting general-purpose models on jurisdiction-specific questions. General-purpose AI tools weren't trained to be right about the law in your state. They were trained to sound coherent and produce plausible output. Those are different things. Where jurisdictional accuracy is the whole ballgame (local court rules, a specific regulatory framework, recent case law that shifted the standard), relying on a general model without verified sources is how you get blindsided.

Automating before the process is documented. This is the one I saw most often, and it isn't unique to law firms. A firm decides to automate intake, or conflicts, or billing follow-up. The process has never been written down; it lives in the head of the paralegal who's done it for twelve years. The automation gets built and immediately starts producing exceptions nobody knows how to handle. Now the paralegal manages the automation instead of doing the work. You haven't saved time. You've moved where the confusion lives. We cover this pattern in more depth in why small business AI projects fail before they start.


What do the ethics rules say about lawyers using AI?

The ABA's first formal ethics opinion on generative AI, Formal Opinion 512 (July 2024), doesn't create new rules. It applies the existing ones: lawyers need a reasonable understanding of the tools they use, must protect client confidentiality, communicate with clients about AI use when the situation calls for it, supervise the people and tools doing the work, and charge reasonable fees. Your state bar may have its own guidance on top of that.

The practical questions follow from there. Which account type is approved? Does the vendor train on your data? Who verifies the output? We answer the common ones in our FAQ on whether Claude is safe for legal work. If associates are already using personal AI accounts, that's a separate problem, and blocking them rarely solves it; see why IT can't be the department of no on shadow AI.


Will clients trust a firm that uses AI?

Law is a trust business. Clients hire you because they believe you'll protect their interests, exercise sound judgment, and tell them the truth when the truth is inconvenient. AI doesn't replace any of that. But it can erode it, quickly and quietly, if it sits somewhere clients can feel it operating.

The question I'd ask about any legal AI deployment is whether it's invisible or explicitly endorsed. If a client knows you're using AI to work more efficiently on their matter and they've agreed to that, you're fine. If a client can tell from the tone of a letter that something was automated, and they didn't sign up for that, you've created a trust problem no efficiency gain is worth.

AI a client never notices, used where accuracy is verifiable: that's good AI for law firms. Anything else needs more caution than most vendors will tell you.


How should a small law firm start with AI?

The safest way for a small law firm to start with AI is from the inside out. Begin with operational work that never touches a client: intake, conflicts checks, billing follow-up, deadline tracking and search across your own past work product. Before you automate any of it, write down how the process actually runs today, including the exceptions, and name the person who owns it. Next, add AI to work an attorney already checks, such as document review, contract redlining and first drafts of research memos, and make verification a required step rather than a good intention. Use business accounts configured not to train on your data, and set a plain rule for which client information may go in. Only after the firm has real experience should AI get near client communications, and then always with attorney review. That sequence keeps you inside your ethical duties and builds the confidence that makes adoption stick.

That's what the firms that get this right do. They build confidence with small internal wins before AI affects how clients experience the firm. One law firm we worked with had documents, email and its practice management system each living apart. We built one place to file, find and work with all of it, connected to the software the firm already runs, with AI assistants that respect each person's permissions. Nothing client-facing changed on day one. The firm's own knowledge just got easier to use.

These firms also have someone in their corner who understands both the technology and the professional responsibility constraints. Vendors know their product. They don't know what your state bar has said about AI, and they won't tell you when their tool is a bad fit for your practice. That's the role our AI strategy and fractional AI leadership work fills.


Mosaic has worked on legal AI adoption with firms from boutique practices to larger regional firms, across Cedar Rapids, Iowa City and beyond. If you want someone who will tell you what not to do as well as what to try, start with our Advise, Educate & Lead page.

Frequently Asked Questions

What is the safest way for a law firm to start using AI?

Start inside the firm, on work an attorney can check. Good first uses are document and contract review, first drafts of research memos, search across your own past work product, and operational tasks like intake, conflicts and billing follow-up. Write down how each process actually works before you automate it, and keep anything that reaches a client behind a human review.

Can lawyers use ChatGPT or Claude for client work?

They can, with care. The ABA's Formal Opinion 512 says existing duties apply: competence, confidentiality, communication, supervision and reasonable fees. In practice that means a business account configured not to train on your data, a clear rule about what client information may go in, and an attorney who verifies every citation and conclusion before it is relied on.

Do lawyers have to tell clients they use AI?

It depends on the task and your jurisdiction. ABA Formal Opinion 512 says some uses call for telling the client or getting informed consent, such as putting confidential information into a tool that may learn from it. Check your state bar's guidance as well. A good practical test: if a client would be surprised to learn AI was involved, tell them.

Why does legal AI get case law wrong?

General-purpose AI models are trained to produce fluent, plausible text, not to be right about the law in your state. They can invent citations or miss a recent case that shifted the standard, and they sound just as confident when they are wrong. Use them as a starting point, check every authority, and prefer tools that cite sources you can open.

What should a law firm avoid automating with AI?

Avoid sending AI output to clients without attorney review, relying on a general-purpose model for jurisdiction-specific questions, and automating any process that has never been written down. That last one is the most common mistake. The automation produces exceptions nobody knows how to handle, and the person who knew the process ends up managing the tool instead of doing the work.

Drew Bloom, founder of Mosaic Solutions

Drew Bloom

Founder, Mosaic Solutions — Business Systems, Software & Automation

Drew is the founder of Mosaic Solutions. He spent several years as an automation consultant and Innovation Lead for Affinity, where he worked with law firms and enterprise clients on AI strategy, change management, and custom software. He works with small and mid-sized businesses in Iowa and beyond to fix the systems that slow them down — with better processes, software, and AI that actually works.

More about Mosaic LinkedIn Work with Drew

Bring Us the Decision in Front of You.

AI strategy, policy, vendor choices and fractional technology leadership, scoped first and billed for the time the work takes.