How a small law firm can start with AI without a big project
By the ChamberOne team 5 minute read
A principal in a ten-person firm has read enough headlines to know that she should be doing something with AI, and not enough to know what. She has no technology team and no time for a project. Good news: she does not need either. A small firm can start with two modest uses, a page of rules and a month of paying attention. Here is how, and what to leave for later.
Key points
- Start with frequent, tedious, easy-to-check tasks rather than research or advice.
- Choose two uses and one approved tool, and settle data rules before anyone starts.
- One page of review rules, with a lawyer reading everything that leaves the firm, does most of the safety work.
- Run a month with a small group including a sceptic, and log time saved and corrections needed.
- Postpone client-facing automation, tools with unclear data handling and firm-wide rollouts.
Begin with the problem, not the product
Ask the team what costs them time that does not need much judgment. Typical answers are writing the first version of routine letters, summarising long documents, tidying notes into file records, and drafting meeting follow-ups. Pick tasks that are frequent, tedious and easy to check.
Avoid starting with the exciting uses. Legal research and advice drafting are where mistakes are costly. They are better tackled once the firm has built the habit of reviewing output.
Write the list on a whiteboard and give each task two scores from one to five, one for how often it happens and one for how easy it is to check. Start with the tasks that score high on both. Anything that scores low on checkability goes to the bottom, however tempting.
Low checkability goes to the bottom of the list, however tempting.
Start here: tasks that score high on both.
Bottom of the list.
Not first: start with tasks that score high on both.
Choose two uses, no more
Two is enough to learn from and few enough to supervise. A sensible pair for many firms is internal summarising of documents the firm already holds, and first drafts of routine correspondence that a lawyer reviews before it is sent.
Both have a clear person checking the result, both involve material the lawyer already understands, and neither sends anything unreviewed to a client or a court.
Suppose a three-partner conveyancing firm picks summaries of title documents for internal use and first drafts of routine chasing letters. Neither is exciting. Both happen daily, and in both the fee earner can tell at a glance when the output is wrong.
Decide the tool and the data rules first
Before anyone starts, settle which tool is approved and what may go into it. Work through retention, training use, hosting and access with the supplier and write the answers down. Check the guidance from your regulator or bar on confidentiality and on AI.
A simple rule works well to begin with: no client names, identifying details or privileged material in any tool that has not been approved, and nothing that has not been approved is used for client work at all.
Tell the support staff as well as the lawyers. Reception, accounts and secretarial staff handle client information every day and often have the most to gain from small time-savers, and the least guidance on what is safe.
Write a one-page review rule
Most of the safety comes from one habit: a person reads everything before it leaves the firm. Put it on a page that everyone signs.
- AI may be used for the two named purposes, with the approved tool only.
- A lawyer reads and takes responsibility for anything sent to a client or a court.
- Citations and quotations are checked against the source before use.
- The file records when AI was used, in a line.
- If in doubt, ask the supervising partner before sending.
Run it with a small group for a month
Choose two or three willing people, ideally including one sceptic. Ask them to keep a simple log: what they used the tool for, how long it took compared with the usual way, and what they had to fix. A short weekly conversation is enough.
The point is to learn what it does well in your firm's work, not to prove that it is brilliant. A sceptic will find the failures early, which is exactly what you want.
Hold the first review after one week, not four. Early problems, such as a tool that invents a date or an instruction that produces vague drafts, are cheap to fix and give the group confidence that someone is paying attention.
Ask each person to keep one example of a good result and one example of a bad one. Those two examples, shown at the review meeting, teach the rest of the firm more than any summary of numbers.
Measure what matters
Resist elaborate metrics. A few observations tell you most of what you need.
Record the numbers in a shared note rather than a system. The aim is a judgment about whether the use is worth keeping, not a report for anyone outside the firm.
- Time saved on the named tasks, estimated honestly by the people doing them.
- Number and type of corrections needed before the output could be used.
- Any near misses, such as an invented citation or a wrong date caught in review.
- Whether the people using the tool want to keep using it.
What to postpone
Leave these until the first two uses are stable and the review habit is real.
- Anything that goes directly to a client or a court without a lawyer reading it.
- Automated decisions about people, such as case acceptance or staff performance.
- Tools for which you cannot get clear answers on data handling.
- Firm-wide rollouts. Expand by team once one team has run it successfully.
- Custom builds and integrations, which tie up time before you know what you need.
- Frequent, tedious tasks that are easy to check.
- Internal summaries of documents the firm already holds.
- First drafts of routine correspondence a lawyer reviews.
- Anything sent to a client or court without a lawyer reading it.
- Automated decisions about people.
- Firm-wide rollouts and custom builds.
Deciding at the end of the month
After four weeks, meet and decide. Keep, change or stop each use. If the time saved was small or corrections were heavy, stopping is a fine result. You have spent little and learned something. If it worked, add one use, with the same review rule, and run the next month the same way.
Share the result with the whole firm, including what did not work. A short note saying what was tried, what was found and what happens next does more to calm worries and build sensible interest than any announcement of a new strategy.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
A 90-day AI pilot plan for a law firm
A week-by-week plan to choose a use case, set measures, run a small trial, review errors and decide.
5 minute readChoosing AI models and firm-level controls: what a practice should decide
Model choice, an off switch and a usage record are three decisions a firm should make deliberately.
4 minute readMeasuring AI return on investment in a law firm
Measure per task against a baseline, count review time, and claim only what you can trace.
4 minute read