Can AI help us keep track of the undertakings on a matter?

Yes, in the way that matters most in practice. AI can read a file end to end and produce a list of every place an undertaking appears to have been given or received, with a page reference for each, far faster than a fee earner can skim three lever arch files. What it cannot do is decide whether a form of words binds the firm, or confirm that an undertaking has actually been performed. It finds candidates; you exercise the judgement.

Why undertakings go missing in the first place

Undertakings rarely arrive labelled as such. They turn up in the fifth paragraph of a long letter, in a one line email sent at the end of a busy Friday, or in a reply confirming what was agreed on the telephone. Nobody sets out to hide them, but they do not always reach the register.

The usual failure points are familiar to every firm:

• Wording that creates an obligation without ever using the word undertaking.
• Undertakings given by a fee earner who then goes on holiday, leaves, or hands the matter on.
• Undertakings received from the other side, which need chasing just as much as those given.
• Conditional undertakings, where the trigger sits in a different document from the promise.
• Obligations that outlive the transaction, so the file is closed while the promise is still live.

The Solicitors Regulation Authority expects undertakings to be performed and expects firms to be able to show that they are managed rather than left to memory. A register that only records what someone remembered to record is a weak answer if something goes wrong.

What AI can usefully do

Given the papers on a matter, an AI assistant can:

1. Sweep the whole file for language that typically signals an undertaking: we undertake, we confirm that we will, on completion we shall, we hold to your order, we will forward on receipt.
2. Name the page each candidate came from, so a fee earner can read the surrounding paragraph rather than take the summary on trust.
3. Separate undertakings apparently given by the firm from those apparently given to it.
4. Pull out the working detail of each one: what is promised, to whom, by when, and on what condition.
5. Look elsewhere in the papers for anything that suggests the obligation has been discharged, such as a completion statement, a redemption receipt, a filed document or a release in later correspondence.
6. Flag pages it could not read, so a poor scan of a key letter is visible rather than silently skipped.
7. Produce a draft register entry for each candidate, ready for a qualified person to check, amend or reject.

That is a substantial amount of mechanical work removed from a task that firms often postpone because it is tedious.

What you must not hand over

Whether a particular form of words amounts to a binding undertaking is a question of regulatory and legal judgement. It depends on the context, the parties, the course of dealing and the seriousness with which the recipient was entitled to treat it. No AI tool should be given the final word on that, and you should not record an entry in the register on the strength of a summary alone.

Three other limits are worth stating plainly.

First, absence of evidence is not discharge. If the file does not show that a promise was performed, that is a prompt to check with the fee earner or the other side, not a conclusion.

Second, an undertaking given orally and never confirmed in writing will not appear in the papers at all. A file sweep cannot find what was never written down, which is one reason to ask the fee earner as well as the file.

Third, your register and your diary remain the systems of record. An AI review is a way of testing whether the register is complete. It is not a replacement for it.

A process that works in a small firm

Pick the moments where the risk is highest and run the sweep then, rather than trying to do everything at once:

• When a matter is inherited, on a departure, a long absence or a transfer between fee earners.
• At the file review stage, as one item on the reviewer's checklist.
• Before closure and archiving, to catch obligations that will outlive the retainer.
• On a sample of files across a department, so the supervisor can see whether register discipline is holding up in general.

Whatever you choose, make the follow through explicit. The output goes to a named person. That person reads the cited pages, decides what is genuinely an undertaking, updates the register, diarises anything outstanding and records on the file that the check was done and when. A list nobody acts on is worse than no list, because it looks like assurance.

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; where the papers do not say, it says so rather than guessing, and any page it could not read is flagged rather than skipped. It prepares drafts for a qualified person to review and sign off, and the firm's own register and diary system still govern. It assists qualified professionals and does not replace them.