AI in law firms: what can be automated today and what can't

What can be automated today

Receiving and sorting email

Identifying who is writing, which client and matter each email belongs to, and whether it opens a new matter or continues one under way. It's reading and sorting work, where AI is very reliable and the time saving is immediate.

Client documents

Requesting documents, receiving them, checking they're complete and legible, filing them in the case file and reminding the client of what's missing. In a firm with volume, this alone frees up hours every day.

Extracting data from documents

Reading an employment contract, a deed or a judgment and extracting the relevant data (dates, parties, amounts, clauses) to fill in the case file or a template.

Drafts of recurring documents

Dismissal letters, formal requests, procedural documents or client communications that follow a known structure, generated from the firm's template and the case file data, ready for review.

Calculations with rules

Severance, holiday counts, deadlines or tax settlements. There's an important nuance here: the calculation shouldn't be done by the language model but by programmed, tested rules. The AI decides which calculation is needed and with what data; the system calculates it.

Deadline control and follow-up

Recording each matter's deadlines, alerting before they expire and following up on what's pending with clients and third parties.

Questions about a matter's status

Answering the client who asks where their file stands, with real information, and passing to the lawyer whatever needs their input.

What shouldn't be delegated

• Strategy and advice. Deciding what to do in a matter is the lawyer's job.
• Signature and responsibility. No document should leave the firm without a lawyer having reviewed it.
• Unverified legal research. Models can invent plausible-looking judgments or articles. In the United States, in Mata v. Avianca (2023), a court sanctioned several lawyers for filing fake citations generated by an AI assistant.
• Decisions affecting people without human involvement. Besides being an ethical issue, the GDPR recognises the right not to be subject to decisions based solely on automated processing with significant effects.

Professional secrecy and data protection

A law firm handles especially sensitive information. Before automating, insist on:

• A data processing agreement with every provider that accesses the data, as the GDPR requires.
• Guarantees about where the data is processed and, if it leaves the European Economic Area, about international transfers.
• Data not being used to train models. The business APIs of the main providers exclude it by default.
• Isolation and access control: each firm sees only its own data, and each person and system only what it needs.
• A log of every action, so you can reconstruct what was done, by whom and when.

As for the EU Artificial Intelligence Act, most law firm uses (sorting email, preparing documents, drafting) aren't among the high-risk systems. Obligations such as telling people when they're interacting with an AI and training the team that uses it do apply. We go into it in AI agents in business: risks, human oversight, GDPR and the AI Act.

How to start

1. Pick a flow with volume. Almost always incoming email or client documents.
2. Measure the time it takes today. Without that figure you can't prove the savings.
3. Define what a lawyer must always approve. That list is the basis of the design.
4. Start with close supervision and widen it when the results are as expected.

That's how the digital employees we build for law firms work, today in employment law and estates.