---
title: "AI for law firm intake and review | Mirai360"
description: "Where an AI system fits in a law firm: new-matter intake, conflict searches, bundle summaries and precedent drafts, each checked by a lawyer."
canonical: https://mirai360.ai/blog/agentic-ai-for-law-firms/
---

Blog

# What AI can prepare for your lawyers

New enquiries, conflict searches, bundles and drafts need preparing before a lawyer can use them. An AI system prepares them; a lawyer signs off.

Shivang Patel, Mirai360 · 19 July 2026

An enquiry comes in through the website on a Tuesday afternoon. A company director is in a dispute with a supplier over unpaid invoices and wants to know whether the firm can help. Before a lawyer can answer, the firm has to work out who the parties are, search its records for each of them and their group companies, and gather enough facts for a first view. **If that takes days, the director may have instructed another firm before anyone calls back**, and nothing in the firm's systems records the matter it lost.

Many lawyers have tried a chatbot and come away sceptical. What we build works inside the firm's own systems instead of a chat window. It uses a large language model to read material, works within limits the firm sets, and prepares the next step: an updated matter record, a first version for a lawyer to review, or a query passed to the right fee earner.

## Four places it fits

At intake, the system takes the enquiry from the web form, an email or a call note, and collects the facts a first assessment needs. It runs the conflict search for every party it finds and lists the identity documents the firm will need for client due diligence. The lawyer receives a short intake note with the conflict results attached and decides whether the firm acts. The enquirer gets a same-day acknowledgement that gives no advice, and the lawyer's reply follows the decision.

Given a bundle of contracts and correspondence, the system produces a chronology, a summary of each document and a list of clauses flagged against a checklist the firm writes. Every point in the summary links to the passage it came from, so a reviewer can check it against the source in one click.

The firm already drafts from its own precedents, and the system can make the first assembly. It picks the right precedent, inserts the details of the matter and marks every clause that departs from the standard wording, for a lawyer to review and change.

Client care letters, key-date reminders, the undertakings register, time-recording prompts and billing narratives are routine tasks the system can prepare for approval. Of the four, this one involves the least judgement, and it is often where support staff spend the most hours.

## Confidentiality comes first

For a law firm, the technology question and the confidentiality question are the same question. Client material is confidential and often privileged, and the SRA Code of Conduct requires solicitors to keep their clients' affairs confidential. Pasting a client's documents into a consumer AI service with unclear terms puts both confidentiality and privilege at risk.

A firm should insist on three things.

**We build and run the system in the firm's own cloud, under the firm's access controls.** Before work starts, we agree in writing which outside services can see any client data, and the firm approves that agreement on the advice of its data protection officer or compliance officer.

Every document read, every draft produced and every record updated is logged with the matter, the time and the user. The conflict search looks only at party names across the firm's records and passes any hit to the people who clear conflicts, without opening the matter behind it. For everything else, the firm's information barriers apply to the system as they do to staff, so while it works on one matter it cannot read the documents of a matter on the other side of a barrier.

Limits on what leaves the system are built into it. Nothing goes to a client without a lawyer's approval, and every factual point in a summary carries a link to its source.

## Checking its work

Language models can write confident text that is wrong. In June 2025 the Divisional Court, in [Ayinde v London Borough of Haringey](https://caselaw.nationalarchives.gov.uk/ewhc/admin/2025/1383) , dealt with two cases in which lawyers had put citations to cases that did not exist before the court, along with others that did not say what was claimed. The judges said the use of AI needs an appropriate degree of oversight. They added that the profession can expect the court to ask whether heads of chambers and managing partners have met their leadership responsibilities.

The system works only from the firm's documents and sources the firm approves, rather than from what the model remembers. Each point links to its source, and before the system touches a live matter we test it on the firm's own past matters against answers the firm knows are right. **No system removes a lawyer's duty to check work that goes out under the firm's name, and checking any case law stays with the lawyer.**

## What partners see each week

Partners get a figure they already track, such as the days from enquiry to a decision on the matter, or the hours of preparation inside a fixed-fee matter. We would measure it on recent matters before building anything, then report it weekly once the system is live. A routine cost that is measured makes fixed fees easier to price. [What we build for law firms](https://mirai360.ai/industries/legal/)

If new enquiries at your firm wait days for a conflict search, take the ten-question survey to see whether AI fits your matter intake. [Find your bottleneck](https://mirai360.ai/tools/bottleneck-finder/)

[All posts](https://mirai360.ai/blog/) [Add Mirai360 as a preferred source on Google](https://www.google.com/preferences/source?q=mirai360.ai)
