ChamberOne

Paralegal workflows: delegating tasks with AI support

By the ChamberOne team 5 minute read

A solicitor asks a paralegal to summarise a bundle. The paralegal asks an AI tool to do the first pass. The solicitor reads the summary, trusts it because it came from the paralegal, and quotes it to the client. Three people looked at the work, and nobody checked it against the documents. This is how errors travel through firms: not through carelessness, but through assumptions about who had verified what.

Key points

  • Write a short card for each repeated task: inputs, output, method and review gate.
  • Note where AI contributed so the reviewer knows what kind of error to look for.
  • Make review specific: trace facts to sources, check for omissions and set a clear sampling rule.
  • Keep some tasks done by hand so juniors still build judgment.

Delegation was always about three things

Good delegation has always needed a clear task, a clear standard and a clear point of review. AI adds a new participant but does not change that. What it changes is how easy it is to produce something that looks finished. A fast, fluent first draft hides the question of who has actually checked it.

So the answer is to say, in writing, who checks what. It sounds bureaucratic until the first time it saves you.

Define each task with four parts

For any repeated task, write a short card with four headings. Keep it to half a page.

  • Inputs: which documents or information the person starts from, and where they are.
  • Output: what exactly is delivered, in what form, by when.
  • Method: which steps, including whether and how AI may be used.
  • Review gate: who checks the output, against what, and what they sign off.

Tasks that suit this treatment include document summaries, chronology building, first-draft letters, disclosure review, bundle indexing, form completion and client update drafts.

An example card

Suppose a four-lawyer litigation firm wants a standard process for chronologies. The card might read as follows.

  • Inputs: the documents in the matter folder marked as received from the client, plus the client's own timeline if one exists.
  • Output: a table of date, event, source document and page, with unknowns marked clearly, delivered within two working days.
  • Method: the paralegal may use the firm's approved AI tool to extract dates and events from each document, then must check every entry against the source page.
  • Review gate: the supervising solicitor checks a sample of entries against the source and every entry that is marked as important to the case.

Make the AI step visible

A reviewer needs to know whether AI contributed, because it changes how they read. A human summary tends to make mistakes of omission and emphasis. An AI summary tends to make mistakes of invention and smoothing. Ask the person delegating to note on the work product which parts started with AI. A single line is enough: first pass drafted with the firm's tool, checked line by line against documents.

This is not about blame. It is about telling the next reader where to look.

Review gates that actually work

A gate is only useful if it checks something specific. Review the whole thing is not a gate. These are.

  • Every date, name and amount traced back to a source document.
  • Every legal statement traced to a primary source by the person who relies on it.
  • A check for omissions: is there anything in the documents that should be here and is not?
  • A read for tone and fit with this client.

Two levels often work well. The doer checks everything against sources. The supervisor checks a sample, and everything that matters most. Make the sampling rule explicit, so that a busy supervisor is not tempted to skim.

Do
  • Trace every date, name and amount to a source document
  • Trace every legal statement to a primary source
  • Check for omissions
  • Make the sampling rule explicit
Don't
  • Review the whole thing without saying what to check
  • Let a busy supervisor skim
Making review specific

Protect the learning

Paralegals and trainees learn by doing the tasks that AI now speeds up. If every chronology starts with a tool, they may never build the instinct for what a document means. Build in some deliberate practice. For example, have the person do the first ten entries by hand before using the tool, or review a colleague's AI-assisted work as an exercise. Supervisors should explain why they changed something, not only that they did.

A second example: first-draft client letters

Take a different task. A paralegal drafts routine update letters to clients on a family of conveyancing matters. The card might say that the inputs are the matter notes and the last three file entries, that the output is a letter of no more than one page in plain English, and that the method allows the firm's approved tool to produce a first version from the notes. The review gate is the solicitor, who checks that every date and figure in the letter matches the file and that nothing promises an outcome the firm cannot control.

Notice how little the card says about the tool and how much it says about the checks. That is deliberate. Tools change every few months, while the question of who has checked the facts stays the same.

Handling mistakes

Mistakes will happen, and the way a team responds matters more than the mistake. Agree that anyone who spots an AI error says so, and that nobody is blamed for reporting one. Keep a short shared note of what went wrong and the check that would have caught it, then update the task cards. After a few months, the cards will carry a good deal of practical wisdom specific to your firm.

  1. 1Someone spots an AI error and says so
  2. 2Nobody is blamed for reporting it
  3. 3Note what went wrong and the check that would have caught it
  4. 4Update the task cards
  5. 5Use the improved cards on the next task
  6. Then back to step 1.
Learning from AI mistakes

Questions a team leader should ask monthly

  • Which task cards were used this month, and which were ignored?
  • Where did reviewers find errors, and were they in the AI step or elsewhere?
  • Is any review gate being skipped because of time pressure?
  • Has anyone found a task that the tool handles badly, and should the card now forbid it?

Where a matter system helps

Task cards work best when they live where the work lives. If tasks, documents and notes sit on the matter, the card can be attached to the task and the review recorded alongside it. If they live in a separate document that nobody opens, they fade. Whatever you use, make the review sign-off something that leaves a trace, such as a note or a status change, so that it is possible to see later who checked what.

Tasks and review on the matter in ChamberOne

  • The matter type sets up phases and default tasks.
  • Documents, tasks, time and notes are added on one matter page.
  • An AI brief or suggested tasks can be requested when needed, and each output stays a draft until someone accepts it.
  • A lawyer decides: AI output is not accepted automatically.

What it does not do

  • AI output can be wrong or incomplete. It does not replace a lawyer's review.
  • The software does not decide who should check what. The firm sets its own review rules.

This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.

Read next

Paralegal workflows: delegating tasks with AI support