Which tasks in a law firm are worth giving to AI?

The tasks worth giving to AI are the ones where the source material already exists and the answer can be checked against that source in less time than it would take to do the work yourself. Finding what a bundle says, building a chronology, working out a date, producing a first draft from papers you have: those repay the effort. Anything that depends on judgement, tactics, or a relationship with a client or an opponent does not.

The test to apply

Before handing a task over, ask three questions.

1. Is the source material available? If the answer depends on documents the firm holds, on legislation, or on published guidance or decisions, the work has something to be measured against. If it depends on what a client said on the phone last month, or on what a particular district judge tends to do, there is nothing to check against.
2. Can I verify the output faster than I could produce it? Checking a chronology against the pages it cites is quicker than building one from scratch. Checking a piece of strategic advice is not quicker, because verifying it means forming the view yourself anyway.
3. What happens if it is wrong and nobody notices? Some errors surface immediately; a mis-numbered paragraph in a draft will be caught at review. Others sit quietly on a file until a deadline passes. Tasks in the second category still suit AI, but only with a verification step that is written down and actually performed.

If a task passes all three, it is a candidate. If it fails the first, it is not a task for AI at all.

Tasks that usually repay the effort

These are the areas where firms of four to fifty fee earners tend to see the clearest return.

• Finding what the papers say. Questions such as what date a contract was signed, whether a particular term appears, or what the other side said about a point in correspondence. The value is in the citation: an answer that points at a page can be confirmed in seconds.
• Chronologies and schedules. Assembling events from documents is slow, mechanical and error-prone when done at speed by a tired person.
• Date and figure calculations. Counting periods, applying a formula, running interest. The working matters more than the number, because the working is what a supervisor checks.
• First drafts from material you already hold. A letter, a statement, a summary of a position. A draft that needs heavy editing is still faster than a blank page, provided the fee earner is editing rather than adopting.
• Locating the relevant provision or guidance. Not deciding what the law means, but getting quickly to the section, the guidance page or the published decision so you can read it yourself.
• Internal, non-billable work. File reviews, checklists, summaries for a supervision meeting. The stakes are lower and the habit of checking can be built safely.

Tasks to keep away from it

Some work should stay with the fee earner regardless of how capable the tool is.

• Advice on what the client should do. Strategy, risk appetite, whether to settle, whether to litigate. This is the work the client is paying a qualified person for.
• Anything where the source is not on the file. If the papers are incomplete, the honest answer is that the papers do not say. An answer that fills the gap is worse than no answer.
• Judgement about a tribunal, an opponent or a witness. These depend on experience that is not written down anywhere.
• Sending, filing or serving. A person should remain the last link before anything leaves the firm.
• Difficult conversations. Complaints, bad news, fee disputes. AI can help you prepare, but the conversation is yours.
• Decisions about people. Recruitment, performance, supervision outcomes.

How to decide in your own firm

Do not start with the tool. Start with a week of your own work.

Ask two or three fee earners to list, for one week, the tasks that felt slow and mechanical. Most lists look similar: reading into a file, finding a document, checking a date, pulling together a summary someone else needs. Those are your candidates.

Then apply the test above to each one. Pick a single task, agree how the output will be checked and by whom, and run it on real matters for a few weeks. Ask afterwards whether the checking was quick and whether the output was reliable enough to use. If checking took as long as doing, the task was the wrong one, not the tool.

Resist the temptation to roll out everything at once. A firm that does one task well and knows exactly how it is verified is in a far better position, with its regulator and its insurer, than a firm using AI broadly and informally.

Whatever you choose, the professional obligations do not move. The fee earner who signs the work is answerable for it, and supervision arrangements should reflect that in writing.

Where Alesis fits

Alesis is an AI assistant for UK law firms, made by L25 Limited, used through the web browser, with one conversation for a matter. It answers questions about a matter from the matter's own papers and names the page each answer came from, and where the papers do not say, it says so. It counts key dates and figures with the working shown, each step carrying the rule that allows it, and says the firm's diary system still governs. It prepares drafts for a qualified person to review and sign off; it does not file, serve or email anything.