Can AI tell us what is missing from a file?

Partly, and the distinction matters. An AI assistant can tell you what is missing from the papers it has been given: a letter referred to in a reply but not on file, a gap of four months in the correspondence, a figure quoted in a schedule with no invoice behind it. It cannot tell you what was never sent to the firm in the first place, and it cannot tell you what a file ought to contain in a matter of this type unless you tell it.

Three different kinds of missing

Fee earners use the word "missing" to mean at least three things, and they call for different work.

Documents referred to but not present. A statement mentions an exhibit that is not there. A contract refers to a schedule nobody has seen. An email chain begins halfway through. This is the easiest kind to find, because the evidence of the gap is inside the papers you already have.

Gaps in the sequence of events. The correspondence runs to March and resumes in August. A payment appears with nothing explaining it. The file shows an instruction and then an outcome with no step between. These gaps are visible once the material is arranged in date order, which is exactly the sort of arranging a machine does patiently and a tired fee earner does not.

Gaps in the support for a point you need to make. You want to say the notice was served and served in time. Nothing in the papers proves service. This is the kind of gap that matters most and the kind that is easiest to miss, because the absence of a document announces itself only when you go looking for the document.

What AI is genuinely good at here

Reading everything is the part humans skimp on, and it is the part machines do without complaint. If a tool has read a file page by page, it can be asked direct questions: which documents are referred to in the papers but not included; what is the longest unexplained gap in the correspondence; is there anything on the file showing when this letter was received.

The second thing it is good at is refusing to fill in blanks. A useful assistant, asked when a notice was served, should say the papers do not record it rather than infer a date from the surrounding material. That answer is worth having. It converts a vague unease about the file into a specific item for a list.

The third is scale. On a file of several thousand pages, a request for every reference to a particular document, agreement or person is something you can act on in minutes rather than devoting a weekend to it.

What it cannot do

It cannot know the shape of your matter unless you give it. A machine does not know, unaided, that a file of this kind should contain a signed retainer, a specific statutory notice, a particular consent or a set of accounts. Give it your own checklist and ask it to go through the papers against that list, and you get something useful. Ask it an open question and you get what the papers happen to suggest.

It cannot see outside the file. If the client never sent the bank statements, the papers may contain no clue that bank statements exist. A gap only becomes visible when something on the file points at it.

It cannot judge what matters. A missing covering email may be irrelevant. A missing proof of posting may decide the case. Weighing that is legal judgement and belongs to the fee earner.

And it cannot be trusted blindly on absence. "Nothing on this file shows X" is a claim you should test on anything important, by searching yourself and by asking the person who built the file. Absence is harder to verify than presence, which is precisely why it deserves a second look.

How to use it in practice

A workable sequence on a file you are getting on top of:

1. Make sure the tool actually has everything, including scanned and emailed material, and check that it has flagged any page it could not read.
2. Ask for a dated sequence of the main events, with the page each one came from.
3. Ask what documents are referred to in the papers but not present.
4. Give it your own list of what this type of matter should contain, and ask it to work through the list item by item.
5. For each point you intend to assert, ask what on the file supports it, and treat "nothing" as a task, not an answer.
6. Turn the output into a request list for the client or the other side, and record on the file what you asked for and when.

Done this way, the exercise is a file review with the reading done faster. The chasing, the judgement about what is material and the decision about what to do next remain yours.

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. It assists qualified professionals and does not replace them, and it does not provide legal advice.