How can AI help a law firm run internal file reviews?

AI can take most of the reading out of a file review. It can summarise what a matter is about, list what is on the file, point at where key dates and figures came from, and flag what appears to be missing. What it cannot do is decide whether the fee earner exercised sound judgement, and that is the part of a file review that matters most.

What a file review is actually trying to catch

Most firms run two kinds of review and confuse them at their peril.

The first is a compliance check: is there a client care letter, is the identity evidence on file, is the retainer scope recorded, is the file being billed and closed properly, are key dates diarised. This is largely a question of presence or absence. It is tedious, it is objective, and it is exactly the sort of work AI is good at supporting.

The second is a quality check: was the advice sensible, was the strategy proportionate to the value at stake, did the fee earner spot the issue that was not in the client's instructions, should the matter have been escalated three months ago. This is judgement. No tool substitutes for a senior person reading the correspondence and forming a view.

If you use AI for the first and free up supervisor time for the second, file reviews get better. If you use it to produce a tidy report that nobody experienced ever reads behind, they get worse.

Where AI genuinely helps

A few tasks come up in almost every review and take a disproportionate share of the time:

• Building the chronology. Reconstructing what happened and when, from correspondence that arrived out of order, is slow manual work. A tool that reads the file and produces a dated sequence, with a page reference against each entry, gives the reviewer a starting point in minutes.
• Finding the retainer scope. What did the firm actually agree to do? The answer is usually in one paragraph of a letter sent eighteen months ago. Being able to ask the file that question directly saves a hunt.
• Checking the dates. Whether limitation, a contractual deadline or a directions timetable, a reviewer wants to see the calculation and the rule behind each step, not just a date in a diary. Working shown is what makes the check meaningful.
• Spotting gaps. "Is there anything on this file that records the client's instructions on settlement?" is a useful question. So is "does the file contain a signed engagement letter?" An honest answer of "the papers do not say" is a finding in itself.
• Comparing across files. If you are reviewing six matters handled by the same fee earner, the same questions asked of each produce a comparable picture rather than six differently shaped notes.

Where it does not help, and can mislead

Be clear about the limits before you build a process around it.

A summary reflects what is on the file. If the fee earner had a long telephone conversation and never made an attendance note, no tool will surface it. AI-assisted review will therefore reward files that are well documented and penalise ones that are not. That is arguably the right incentive, but do not mistake a thin summary for thin work.

AI has no view on whether advice was correct. It can tell you that a letter advised the client to accept an offer. It cannot tell you whether that was the right call on the facts, or whether the fee earner should have taken counsel's opinion first. Do not ask it to grade the work.

And it cannot judge tone or client handling. A file can be procedurally immaculate and still show a client who was left in the dark for four months. That is something a person notices by reading the correspondence.

A process that holds up

For a firm of four to fifty fee earners, something like this works:

1. Fix the questions. Write the list of questions every reviewed file gets asked, and use the same list every time. Consistency is what makes a review programme evidence of anything.
2. Run the factual pass first. Chronology, scope, dates, documents present and absent. Save the output.
3. Have the reviewer read the file. Not all of it, necessarily, but the advice letters, the attendance notes and the last three months of correspondence. The factual pass tells them where to look.
4. Record what was checked and how. If AI was used to prepare the review, say so in the review note. A file review record that does not disclose its own method is worth less if it is ever scrutinised.
5. Feed findings back. Reviews only improve a firm if the pattern across files reaches the people who can change it: supervision arrangements, precedent letters, diary practice.

One more point. File reviews involve client confidential material, often across several clients at once. Whatever tool you use for them has to sit inside your existing confidentiality and data protection position, not alongside it. Check where the data goes, who inside the firm can see which matters, and whether anything is retained or reused. A reviewer with a firmwide remit still should not be able to open matters they have no business seeing.

Where Alesis fits

Alesis answers questions about a matter from that matter's own papers and names the page each answer came from; 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. Inside a firm, people see only the matters they are on, and seniority alone grants no view. It assists qualified professionals and does not replace them.