An AI assistant for a law firm: what can be automated in 2–3 weeks

A law firm has as much routine as any other company: tasks from calls and meetings, email, time tracking and invoices, articles for the website, answers to clients’ first questions. There is also a particularity: the work constantly involves information covered by attorney–client privilege and clients’ personal data. So the question is not only what to automate, but how to do it without creating a new risk.

Below is what we can usually launch as a first step in 2–3 weeks, what is better postponed, and which rules we put in place from day one.

Why start with one task?

Because the first result should appear quickly, and people should start using it. An “AI for the whole firm” project drags on for months, and over that time both the tasks and the AI services themselves change. One task that noticeably gets in the way of work and can be solved quickly gives a clear result and shows, on your own example, how to move on.

A convenient way to choose: write down everything that irritates partners and lawyers, and rate each item from 1 to 10 on two axes: how much it hurts and how hard it is to change. Start with the item where the pain is high and the difficulty is low.

What can be launched in 2–3 weeks?

These are the tasks that usually suit a first step.

• A partner’s personal assistant. The partner dictates a voice message in Telegram, and the assistant sets a task and a reminder, adds the meeting to the calendar and sends a summary for the day in the morning. Email triage can be set up separately: what is urgent and what to answer today.
• Articles and case studies for the website. A lawyer dictates a few sentences about a case or topic, and AI prepares a draft article or case study in the firm’s style and asks clarifying questions if facts are missing: dates, amounts, outcome. A person publishes the material after review.
• First-line inquiries in a messenger. A bot replies immediately, including at night, clarifies the nature of the question and contact details, and hands the lawyer a ready request. The bot does not give legal advice: its job is to make sure the inquiry is not lost.
• Draft invoices and certificates of completion. Based on time-tracking data, AI prepares a draft invoice or certificate, and a lawyer or accountant checks and confirms it. Invoices go out on time, and there is less manual copying of figures.

The team does not need to learn new software: everything works through the messenger, email and calendar they already use.

What is better left for the next stage?

There are tasks that often make firms think about AI in the first place, but they do not work as a first step. These are document analysis across the whole archive in a closed environment, quality control of employees’ work, analysis of phone calls and corporate correspondence, and daily reporting to management across the whole firm. These involve more data, more people and more legal questions, including under the Labor Code if it concerns monitoring employees. It makes sense to take on such tasks once the first one is already working and you can see how the team uses AI.

How to stay within attorney–client privilege and 152-FZ?

This is the main question for a law firm, and it has to be settled before launch, not after. We put several rules in place.

1. Two data environments. We determine in advance what can be processed in external AI services and what cannot. Information covered by attorney–client privilege and other client data does not go to external services: it is processed on the firm’s own servers or anonymized before sending.
2. Data in Russia. Personal data is stored and processed as Federal Law No. 152-FZ (Russia’s personal data law) requires, including on servers in Russia. Voice messages can be recognized locally, without sending the recording to an external service.
3. A person confirms everything outgoing. A letter to a client, a publication or an invoice goes out only after a person has checked it.
4. An action log. You can see what the assistant did and when, so an error is easy to find and review.
5. The AI service can be replaced. If a service becomes unavailable in Russia, it is swapped for another and the solution keeps working.

As a result, the firm is left with an AI usage policy: who uses it, what may be given to services and who checks the result.

How do these 2–3 weeks go?

The work goes in stages, and so does payment.

1. Analysis. We work out what exactly takes the time and choose one task. Often it is not what the conversation started with.
2. Development. We build a solution for that task and show how it works on your examples.
3. Rollout. We launch it together with the team so that people start using the solution in their everyday work.

After launch, support begins: we make sure everything works and, if needed, replace AI services.

How much does it cost?

The task review is free. The first solution starts from 50 000 ₽, support from 10 000 ₽ per month. We give the exact price after the review, once the task is clear.

If you would like to discuss where to start in your firm, write to us on Telegram.