Take a firm of ten fee earners, half in residential conveyancing and half in litigation. On plausible assumptions each of them gives two undertakings in a working month, which is twenty across the firm and around two hundred and forty across a year. Perhaps a third are outstanding at any given moment because the trigger has not arrived or the other side has gone quiet. The figures are illustrative and every firm differs. Ask the managing partner to name the eighty or so undertakings live on the firm's files, and the honest answer is that nobody can.
Undertakings do not arrive with a warning. They are given in the flow of work, in the paragraph of an email confirming what was agreed on the telephone an hour earlier, or in a DX letter promising to hold deeds to the other side's order. The fee earner giving one is usually solving a problem in front of them, moving a completion forward or persuading an opponent to release a document. The binding promise is a by-product of getting the work done, and it lands in the correspondence file rather than in any field the firm can search.
Discharge then happens quietly, or it does not happen at all. A mortgage is redeemed and the redemption statement lands, and nobody goes back to the email that promised it to record that the promise is now spent. The file closes because the client's matter has finished, which tells you nothing about whether the promises made along the way were performed. Closing a file closes the retainer. It does not close the undertaking.
Where the risk sits
An undertaking binds the individual who gave it and the firm that employed them, and it survives the archiving of the file and the departure of the fee earner. That is why the risk crystallises so late. The call comes eighteen months on from a lender asking why the charge has not been discharged, the person who gave the undertaking left in the spring, and the only record of what was promised sits in an email chain in an archived matter nobody has opened since. By then you are dealing with an enforceable personal obligation on terms you have to reconstruct before you can decide whether you have performed it.
The regulatory position sharpens the problem without solving it. The SRA Standards and Regulations expect a solicitor to perform an undertaking within an agreed timescale, or within a reasonable period where none was agreed, and they expect firms to run effective systems and controls and to supervise their people properly. Neither expectation is satisfied by good intentions spread across two hundred matters. If you cannot produce a list of what your firm has promised and to whom, you have no system, whatever your risk register says.
What a real register looks like
Building one is unglamorous and entirely within reach. Every undertaking needs the exact wording as given recorded against it, not a summary of what somebody thinks was meant, because the scope of the obligation lives in the words. Then you need who gave it, who it was given to, what event or date triggers performance, and the date by which performance falls due. A partner owns the review, not the fee earner who gave the promise, and it runs on a fixed cycle rather than when someone remembers.
The discipline that makes the difference is retrospective. A register populated only from today forward leaves the whole back catalogue untouched, and the back catalogue is where the danger sits. Somebody has to read back across the correspondence on live and recently closed matters and pull out the language that binds, because relying on people to fill in a form at the moment they give an undertaking assumes they noticed they were giving one.
Reading back across every matter by hand is where this stalls, and that is the gap the Bracton AI Assistant is built for. The assistant is built into Bracton, the case and client management platform, and it is also sold on its own as an add-on that plugs into the case management or CRM system your firm already runs, whether that is LEAP, Clio, Proclaim or another. It works through every live matter overnight, correspondence included, and surfaces language that reads as an undertaking wherever it was given. Outstanding items go onto a list a partner sees in the morning, with the ones whose trigger date has passed flagged for attention. The assistant surfaces and flags. It does not decide that an undertaking has been discharged, because that judgement belongs to a qualified fee earner, and everything it produces goes to one for sign off. The audit trail records what was done and by whom, the hosting is in the UK, and your client data never trains a model.
The test is what it finds rather than how it sounds in the abstract. The demonstration shows the compliance registers of an invented firm keeping themselves, and the written exercise on your own records starts with the client account: book a client account review.