Can AI help us run a high volume of similar low value files?

Yes, in parts. Volume work is where AI tends to pay off fastest, because the same handful of questions get asked on every file and the papers arrive in a predictable shape. The catch is that volume cuts both ways: a habit that saves two minutes a file also spreads an error across two hundred files before anyone notices.

What volume work looks like from the tool's side

Most volume practice, whether that is low value personal injury, debt recovery, residential conveyancing, small employment claims or consumer complaints, breaks down into a small set of repeated steps:

• read what the client or the other side has sent and work out what is in it
• pull out the fixed facts: names, dates, amounts, policy or account numbers, addresses
• work out the dates that matter and what they are driven by
• produce a letter or form that is ninety per cent the same as the last one
• notice when a file is not like the others

The first four are mechanical. The fifth is the one that justifies a fee earner. A tool that is good at the first four, and honest about the fifth, is worth having. A tool that quietly smooths over the unusual file is a liability.

The tasks worth handing over

Start with extraction rather than drafting. Asking a tool to read a bundle of papers and tell you the date of the incident, the date of the letter of claim, the sums claimed and where each of those appears is low risk and immediately checkable. You can verify it in seconds against the page it points at.

After that:

• Date calculations with the working shown. On volume files the date logic is usually the same every time, which is exactly why it gets done on autopilot. Having the steps written out, each with the rule behind it, makes the autopilot auditable. Your diary system still governs; the calculation is a cross-check, not a substitute.
• First drafts from the file. Not a precedent with the name swapped in, but a draft that has actually read the papers and filled in the specifics. The fee earner then does the part that needs judgement.
• Completeness checks. Ask what is missing: no signed authority, no medical evidence, no proof of loss, no response to the last chaser. This is dull work that humans skip when the pile is high.
• Triage at intake. A short, structured read of a new enquiry against your own acceptance criteria, so the files that do not fit get spotted before anyone spends time on them.

The tasks to keep off the list

Do not use AI to decide whether to settle, what a case is worth, or whether a limitation point is arguable. Those are advice, and they are the part of volume work where the money is actually lost.

Be careful too with anything that goes out without a human read. The temptation on low value files is to automate the whole loop: read, draft, send. Resist it. The margin on these files is thin enough that one complaint, one missed deadline or one letter sent to the wrong party wipes out the gain from a hundred efficient ones. Nothing should leave the firm without a named person having read it.

Supervision when the volume is the problem

Supervising one file is easy. Supervising four hundred is a sampling exercise, and AI changes what you should sample for.

A few practical habits:

1. Sample by task, not by file. If the tool is doing date extraction on every file, review the date extraction across a slice of files rather than reading one file end to end.
2. Look for repeated errors, not one-off ones. The question is not "is this file right" but "is anything going wrong the same way every time". That is the failure mode that volume creates.
3. Check the unusual files by hand. Whatever your triage flags as atypical should get a full human read, not a faster one.
4. Keep the sources visible. On volume work, the value of an answer that names the page it came from is that checking costs seconds. If checking costs minutes, nobody will do it at scale.
5. Write down who is accountable. One person who owns how the tool is used across the caseload, and who sees the error reports.

Cost and pricing

On fixed fee or capped work, the maths is simple: the cost per file has to be small and predictable. Watch for pricing that charges per user when only part of the team touches the tool, or that pushes you into a tier you do not need. Measure against something real, such as the fee earner time per file before and after, over a reasonable sample, rather than against an impression that things feel quicker.

Also budget for the checking. Efficiency on volume work comes from doing less of the mechanical part, not from doing less supervision.

Where Alesis fits

Alesis is an AI assistant for UK law firms, used in the browser with one conversation for each matter, so the file's own papers are what it answers from. It names the page behind each answer, flags any page it could not read rather than skipping it, and says so when the papers do not cover the point instead of guessing. It counts key dates and figures with the working shown, each step carrying the rule that allows it, while your diary system still governs, and it prepares drafts for a qualified person to review and sign off. It is funded by credit rather than a subscription, with no feature tiers, so a firm can size the cost against the caseload it actually runs.