Skip to content
Efidis
Efidis

Applied AIUpdated 10 min read

AI for law firms: 10 tasks to hand off and 5 to keep

3D illustration of scales of justice with lilac pans, case files, a pen and an AI sparkle.
In this article

Most of the figures in this article come from the annual report by Clio, a software vendor for law firms that surveys legal professionals and also uses data from its own customers. They show where the industry is heading, not an average across all firms.

What the 2026 data says

85% of the professionals Clio surveyed say their firm uses AI to some degree, up from 79% in 2025. The same report projects the billable share of the workday rising from 39% to 42% in 2026, and 68% of AI users say it took tedious work off their plate (Clio, 2026). The billable-hours projection is Clio's own and doesn't prove AI is the cause.

Clients have opinions too. 82% of the consumers Clio surveyed want to know when their lawyer uses AI, 70% trust a lawyer more when they're told, and only 6% are comfortable with AI being involved in high-stakes personal matters (Clio, 2026).

The gap lies elsewhere. Many firms already use AI but haven't decided what it's for, what it isn't for, or which case files it can touch. A simple rule sorts that out.

The rule: draft yes, signature no

To decide what to hand off, split the work in two. The draft is everything that prepares a decision: reading, organizing, extracting, summarizing, proposing a first version. The signature is everything that makes the decision or communicates it under your name: the advice to the client, the brief that gets filed, the response to a court order, the power of attorney you grant.

AI can handle much of the draft. The signature doesn't get delegated, even if the tool is right almost every time, because professional responsibility is yours, not the software's. And every AI draft gets reviewed with the same care as a first-week intern's: in good faith, taking nothing for granted.

10 tasks you can hand off now (with review)

Each row says how serious a mistake would be and what to check before using the output. Low risk means a mistake shows up quickly and can be fixed without harm; medium risk means it could reach the client or the court if nobody reviews it.

TaskRiskWhat to check
1. Summarize a long case file for a first read (complaint, answer, evidence)MediumThat it doesn't leave out facts, dates or claims; read the original before giving advice
2. Pull deadlines and dates from an order or notice and propose them for the calendarMediumEach deadline against the document and the counting rules that apply; nothing goes on the calendar unreviewed
3. Build a timeline of facts from emails and documentsMediumThat every fact has a supporting document
4. Compare two versions of a contract and list the changesMediumConfirm each change in the document; look for the ones it missed
5. First draft of a brief or contract from the firm's templates and approved clausesMediumThat it uses your clause bank, not generic text; full review by the lawyer
6. Routine emails and letters to clients (case stage, missing documents)LowTone, names, dates and that it doesn't promise outcomes
7. Transcribe and summarize meetings with clients who agreed to itLowNames, figures and commitments made
8. List the documents still needed for a power of attorney, a complaint or a filingLowThat the list comes from your client records, not from what the AI assumes
9. Sort and name incoming documents using the firm's conventionsLowThe ones left unsorted or filed in the wrong case
10. Explain a stage of the case to the client in plain languageMediumTechnical accuracy, and that it doesn't read as advice for their specific case

All ten have something in common: there's a document or a record to check the output against. That's the practical test. If you can verify the result against a source you already have, the task can be delegated. If verifying it means doing the whole job yourself, the savings disappear.

An example of how much review weighs

This example is made up, with round numbers. A five-lawyer firm receives a 60-page case file to prepare an answer. Summarizing it by hand takes 90 minutes. With AI, the summary takes 5 and checking it against the original takes 25: the firm saved an hour. If the review means reading all 60 pages again, it saved nothing. Measure the review, not just the generation.

5 tasks you don't hand off

  1. The final advice or recommendation to the clientWhat to do, what risk to take, whether to sue or settle, how to structure a deal. AI can lay out options; deciding which one you recommend and why is your job and your responsibility.
  2. Citing statutes, cases or commentary without checking themAI tools can make up references that look real. In Mata v. Avianca, a federal court in New York fined two lawyers and their firm $5,000 for filing a brief with court decisions that ChatGPT had invented (Mata v. Avianca, 2023). Every citation gets opened and read in the official source.
  3. Filing documents and keeping final control of deadlinesWhat goes before a judge or an agency carries your signature, and a missed deadline can't be recovered. AI can propose dates and prepare the papers; a lawyer counts the deadline and files.
  4. Conflict checks and client intake decisionsThey require knowing the firm's history, its relationships and the context of each matter. A tool can search names in your client database; the decision to take or decline a client isn't handed over.
  5. Difficult conversationsDelivering bad news, discussing fees, negotiating with opposing counsel, judging whether a client is telling the whole truth. In those parts of the job, judgment and trust matter more than speed.

Confidentiality: what to ask before you upload anything

This isn't legal or ethics advice. Rules on attorney-client confidentiality vary by country and by the bar or court that regulates you. Before using AI with client information, check the rules and opinions that apply to your firm.

As a reference for how regulators are approaching this, the American Bar Association's ethics committee issued Formal Opinion 512 on generative AI in 2024. Among other things, it says that before entering client information into an AI tool, a lawyer must assess the risk of that information being exposed to people outside the firm. For tools that learn from what they're given, it calls for the client's informed consent and makes clear that boilerplate in an engagement letter isn't enough (ABA, 2024). It applies to U.S. lawyers, but the questions it raises are useful for any firm.

This checklist doesn't replace that review, but it organizes what to find out about each tool before you put a client document into it:

  • You know whether the vendor uses what you give it to train its models, and whether you can turn that off by contract, not just with a setting.
  • You know how long it keeps conversations and files, and whether you can delete them.
  • You know which country the data is stored in and whether that conflicts with any obligation of yours or your client's.
  • You use a firm account with a business contract and terms, not free personal accounts.
  • You know who inside the firm can see what others uploaded.
  • You remove names, ID numbers and data the task doesn't need.
  • You've checked whether your professional rules require informing the client or getting consent, and how to document it.
  • There's a written list of approved tools and what information can go into each one.

If you can't check the first three boxes, use that tool only with public information or documents with no client data. And put the rules in writing: there's a one-page AI use policy template to start from.

What about fees?

If you bill by the hour and AI saves you time, the client will ask. Opinion 512 also says a lawyer billing hourly can only bill the time actually spent, review included, not the time the tool saved (ABA, 2024).

What AI doesn't fix: scattered files and deadlines

An AI tool can summarize a case file in minutes, but it doesn't know a deadline falls on Thursday if that date lives in an associate's inbox, or how many hours a case has used if everyone logs them in their own spreadsheet. For AI to help run the firm, and not just with loose documents, the data has to be in one place first.

This is the minimum case tracking template. If today you build it from three different files, that's the problem to solve before any AI tool:

FieldWhat it's for
Client and caseOne record per case, with its number, the court and the opposing party
Responsible lawyer and teamWho's accountable and who works on it
Next deadline and next hearingThe date, who counted the deadline and from which notice
Case stageA fixed list (complaint, answer, evidence, closing arguments, judgment, appeal…)
Powers of attorney and client documentsWhich power is in force and what you've been waiting on from the client, and since when
Hours logged and budgetTo know whether the matter is within what was quoted
Last contactDate and channel; avoids clients who don't know where their case stands
Billed and collectedWhat was invoiced and what has come in

With the table up to date, a number shows up that almost no firm tracks: where inquiries come from (referrals, the website, WhatsApp) and how many end in a signed engagement. That's how we approach it on our page for law firms.

When it's not worth it

  • You handle very sensitive matters (criminal, family, health information) and don't yet have an approved tool that passes the checklist. Wait until you do.
  • Nobody has time to review. Without review, AI doesn't save work: it moves it to the day the mistake surfaces.
  • Volume is low. If you draft two contracts a month, learning a tool and keeping an eye on it can cost more than it saves.
  • Your clients asked you in writing not to use AI on their matters. That instruction wins.
  • Your case files, deadlines and hours live in three different places. Move it off spreadsheets first; AI works better on complete data.

How to start this week

Pick two low-risk tasks from the table, like routine emails and the list of missing documents. Use them for a month with a tool that passes the checklist, with no sensitive client information. Track how long it takes to review each result and how many mistakes you find. With those numbers, decide whether to add a medium-risk task, like case file summaries, or whether the tool isn't worth it. To help your team learn to do this within these rules, there's AI training.

Sources

  1. Clio, 2026 Legal Trends Report, press release, 2026 (opens in a new tab)
  2. American Bar Association, ABA Ethics Opinion on Generative AI Offers Useful Framework (on Formal Opinion 512), Business Law Today, 2024 (opens in a new tab)
  3. Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023) (opens in a new tab)
See how we work in law firms

Frequently asked questions

Case files and deadlines under control

Efidis builds your firm a web and mobile app shaped around how your lawyers work. On the call we'll look at how you track cases today and what to organize before adding AI.