How to use AI in a law firm: what to automate first
In a law firm, AI helps with preparation work: client intake, a first pass on contracts, search over the firm's precedents, legal research with the source cited, and deadlines. The lawyer checks and signs. The first decision is where data may go; the first pilot is usually a contract first pass on a single document type.
Lawyers already use AI. What they do not do is use it on the files that matter: in the Romanian National Bar's 2026 survey, 74% of responding lawyers used AI tools, but 86% of those users avoided entering sensitive or identifiable data, and 59% avoided AI entirely for confidential work (UNBR, 2026; 457 complete responses, self-selected respondents). This guide is about closing that gap: what a law firm or legal department can automate, with the data kept safe.
How do you use AI in a law firm?
You use AI on preparation, not on the decision: client intake, a first pass through contracts, searching your own precedents, legal research with sources and deadline administration. The lawyer checks, decides and signs. If you want the advisory view first (what AI-use policy the firm adopts, how to train the team), Vlad Tudor covers it on his page on AI for law firms and legal departments.
| Problem | What AI can do today | What stays with the lawyer |
|---|---|---|
| Client intake and conflict checks | Extracts parties and facts from the email, searches the client base for matches | The conflict decision |
| First pass on contracts and due diligence | Compares clauses with the firm's positions, flags deviations, builds the table | Negotiation and the opinion |
| The firm's precedents and templates | Searches memos, opinions and templates, citing the paragraph | Choosing the argument |
| Legal research | Finds the text in force and cites the official source | Interpretation |
| Deadlines, court files, time entries | Extracts deadlines, tracks case files, drafts time entries | Confirming every deadline |
1. Client intake and conflict checks
A new client sends a long email with attachments. Someone extracts the parties, the facts and the request, then checks whether the firm has ever acted for the other side. AI can build the intake sheet from the email and search the client and matter database for the parties' names, including variants and groups of companies. The output is a list of possible matches for the responsible lawyer.
What it cannot do: decide that there is no conflict. It can miss a link that is not in the data; the decision stays with a person.
2. First pass on contracts and due diligence
In a due diligence or a review of a client's contracts, most of the time goes on finding the clauses: change of control, termination, liability, governing law. AI does the first pass: it compares each contract with the firm's positions (the playbook), flags deviations, cites the clause and puts everything in a table. We have worked on this kind of problem in a contract-analysis project that identifies critical clauses.
What it cannot do: say whether a deviation is acceptable for the client. That is legal and commercial judgement.
3. The firm's precedents and templates
Every firm has years of opinions, memos and templates that only their authors can still find. An assistant built on the firm's documents (RAG) answers questions like "have we argued this before?" and points to the exact document and paragraph. How such a system is built is described in how to build a chatbot on your company data.
What it cannot do: fix a disorganised document system. And access rights have to be mirrored exactly: a lawyer must not find, through the assistant, a file they have no access to.
4. Legal research, with the source cited
General-purpose models sometimes invent articles or quote repealed versions. A good legal assistant answers only from verified texts and shows the source. That is how we built ai-aflat.ro: an assistant over 220,000+ Romanian legislative acts, updated daily, which cites the law, links the official source and deliberately refuses to give verdicts.
What it cannot do: replace interpretation. The answer is a verifiable starting point, not a legal opinion.
5. Deadlines, court files and time entries
Deadlines in summonses and judgments, tracking case files on the courts' portal and recording time are administrative work that eats lawyer hours. AI can extract deadlines from documents and propose them in the calendar, track changes in case files, and draft time entries from emails and the calendar.
What it cannot do: be the only deadline system. Every deadline AI proposes is confirmed by a person. A missed deadline costs more than the whole project.
What should a law firm automate first?
Start with the first pass on contracts for a single document type, for example the supply contracts in a due diligence, or with search over the firm's precedents. Both have volume, a lawyer who checks the result anyway, and a cost that is easy to measure: hours per contract, hours per search. Legal research is a good candidate too, especially if the firm works a lot with Romanian legislation.
A hypothetical example: a firm with an M&A practice would take a completed due diligence, run the first pass on the same contracts and compare the AI table with the one the team produced. That is the only way to learn the real gain: measured in your own pilot, on a file you already know. The steps are in how to run an AI pilot.
Data and compliance: professional secrecy, GDPR, the AI Act
A few things to know before the pilot. This is not legal advice; your firm makes the call.
- Professional secrecy: the first decision is where data is allowed to go. There are three options: a cloud service with EU processing and a contract that rules out training on your data, a private deployment (on-premise or in a cloud the firm controls), or no AI processing at all for certain files. Write the decision down before the pilot, not after.
- GDPR: the AI supplier is a processor and needs an Article 28 contract. Access rights from the document system must carry over into the assistant. More in how to stay GDPR-compliant when using AI.
- AI Act: tools a law firm uses for research, contract review or administration are not, as a rule, high-risk. Annex III targets systems used by or on behalf of judicial authorities (Regulation (EU) 2024/1689). Since February 2025, Article 4 requires measures that support the AI literacy of the people who operate the systems.
- An assistant that talks directly to clients has to say it is AI (Article 50, applicable since 2 August 2026) and must not give legal advice in the lawyer's place.
How do I start?
- Tech Call, free: a short conversation where you tell us which process you want to automate and where data is allowed to live. Book it here.
- Tech Audit: a short, paid audit on real documents (anonymised where needed) and the firm's systems. You get in writing the use case, where the system runs, what data it uses and what "working" means in numbers.
- Pilot: one flow, on a file the team knows, with lawyers checking every result.
- Production: integration with the document system and its access rights, an audit trail, monitoring. The code and documentation stay with you.
What we build for companies is on the AI agents and AI automation pages, and our published projects are on the projects page.
Sources
- UNBR (Romanian National Union of Bars), pre-Congress survey 2026, published June 2026 (1,364 started, 457 completed; self-selected respondents; in Romanian)
- Regulation (EU) 2024/1689 (AI Act), Articles 4 and 50 and Annex III
74% of lawyers responding to the Romanian National Bar's 2026 survey used AI tools, but 86% of those users avoided entering sensitive or identifiable data, and 59% avoided AI for confidential work (457 complete responses).
Frequently asked questions
Can I use AI on confidential client documents?
Yes, if you decide in advance where the system runs and what data it receives. The options are a cloud service with EU processing and no training on your data, a private deployment the firm controls, or excluding certain files. Write the decision down before the pilot. This is not legal advice; the firm applies its professional-secrecy rules.
Does AI make up legal citations?
General-purpose models can invent articles or quote repealed versions. That is why a legal assistant has to answer only from verified texts and show the source. That is how ai-aflat.ro works, built by Sapio over 220,000+ Romanian legislative acts updated daily: it cites the law and links the official source.
Is an AI tool for lawyers high-risk under the AI Act?
As a rule, no. Annex III targets systems used by or on behalf of judicial authorities. Research, contract review and administration in a law firm generally fall outside it. Article 4 (AI literacy) and Article 50 (transparency towards clients) still apply. The exact classification depends on the specific system.
What should a legal department automate first?
A first pass on contracts for a single document type, or search over the firm's precedents and templates. Both have volume, a lawyer who checks the result anyway and a cost that is easy to measure. Run the pilot on a closed file and compare the AI output with the team's.
Can AI run conflict checks?
It can prepare them: it extracts the parties from the email and searches the client base for matches, including name variants and groups of companies. It cannot decide that there is no conflict, because it can miss a link that is not in the data. The decision stays with the responsible lawyer.
Want to discuss a project?
Book a free discovery call with the Sapio team.