How can AI help a law firm build a chronology of a matter?

AI is genuinely useful for chronologies because most of the work is mechanical: finding every dated event in a large set of papers, putting them in order, and recording where each one came from. A good tool does that quickly and points at the page for every line. What it cannot do is decide which events matter, which are disputed, and what the silences in the file mean; that is the fee earner's judgement and it is the part that makes a chronology worth having.

Why chronologies stall

Most firms know a chronology would help. Most matters do not have one, or have one that stopped being updated eighteen months ago.

The reasons are always the same:

• The first pass is slow. Someone has to read several hundred pages before writing a single line.
• It is unbilled or hard to bill, so it slips behind the things with deadlines.
• Once it falls behind, catching up feels worse than starting again.
• The person who built it leaves, and nobody else trusts it enough to add to it.

AI changes the economics of the first pass and of catching up. It does not change the need for someone to own the document.

What a chronology tool should actually give you

Before you use AI for this, be clear about what a usable output looks like. A list of dates with no provenance is worse than nothing, because someone will rely on it in a hearing and be unable to say where a line came from.

Insist on:

1. A page reference for every entry. Not "the correspondence bundle", but the document and the page. If you cannot get from a line in the chronology to the paper behind it in a few seconds, the chronology will not survive contact with a judge or an opponent.
2. The event described in neutral terms. A chronology entry should say what happened, not what it proves. Argument belongs in the note that sits alongside it.
3. A record of unreadable material. If a page was a poor scan or a photograph of a handwritten note, you need to be told, not to have it quietly dropped.
4. A clear distinction between dated and undated material. An undated letter that probably sits in a particular window is useful, but it must be flagged as an inference rather than a fact.
5. The source type. An event recorded in a contemporaneous email is not the same evidential animal as an event recalled in a witness statement drafted three years later.

A working method

A method that holds up in practice looks something like this.

Start by defining the period and the question. "Everything in the file" produces a chronology nobody reads. "Every communication between the parties from the tender to termination" produces one that gets used.

Run the first pass over the core documents only: the contract or the key instrument, the correspondence, the attendance notes. Leave the exhibits and the peripheral material for a second pass, or you will drown the important events in noise.

Read the output against the papers. Not every line, but a proper sample, weighted towards the entries you would actually rely on. If the sample is clean, spot-check the rest. If it is not, find out why before you go further.

Then do the part only a fee earner can do. Mark the entries that are disputed. Mark the entries where the only source is one party's account. Note what you expected to find and did not: the meeting with no attendance note, the month with no correspondence, the version of the document that nobody has produced. Those gaps are frequently the most valuable output of the whole exercise, and they are invisible to a tool that is only listing what exists.

Finally, decide who owns updating it, and make updating it part of the routine when new documents arrive rather than a separate project.

The checks that matter

Two failure modes are worth watching for specifically.

The first is date confusion. Documents mix formats, refer to "last Tuesday", carry a header date that differs from the date of the event described, and get scanned out of order. Any entry that drives a deadline or a limitation argument should be checked against the document itself, every time, without exception.

The second is quiet omission. A chronology that looks complete is more dangerous than one that is obviously partial, because nobody goes looking for what is missing. Ask what was searched, ask what could not be read, and satisfy yourself that the answer is honest rather than tidy.

And treat the chronology as a working tool, not as a finished product. If it goes into a bundle or a skeleton, it needs the same review as anything else that leaves the firm with your name on it.

Where Alesis fits

Alesis answers questions about a matter from the matter's own papers and names the page each answer came from; if the papers do not say, it says so. Documents are read page by page, so citations point at pages, and any page it could not read is flagged rather than skipped. When it cannot find support for a point it says what is missing instead of guessing, and where dates and figures are counted it shows the working, with the firm's diary system still governing. It prepares drafts for a qualified person to review and sign off.