QubiLEX: the AI that reads court documents, identifies the client, files the case and drafts the letter; 98% handled without intervention, the lawyer reviews and sends.

Overview

QubiLEX is an application developed by Qubitech that automates the sorting and processing of court documents received by a law firm. This case study covers a six-month deployment, from January to June 2026, still ongoing, in a general-practice firm of five lawyers in Geneva.

Every working day, around thirty documents arrive from courts and registries: statements of claim, defence briefs, judgments, hearing notices, enforcement orders, notices of appeal, requests for evidence, and more. Before QubiLEX, the firm's secretaries scanned incoming mail and forwarded it to the lawyer handling the case. That lawyer then had to open the document, determine which case file it belonged to, file it manually and notify the client before starting the legal work. QubiLEX now takes over this administrative chain from end to end.

As soon as an email arrives from an authorised sender, the system reads the attachment, extracts the case reference, the parties and any deadlines, recognises the document type, identifies the client from the parties named, files the document in its OneDrive folder and drafts a notification letter. The lawyer reviews this draft before sending it.

The system only processes emails from approved addresses. For this firm, everything runs in French, with a single preference setting both the interface and the language of the drafted letters (QubiLEX is bilingual: French or English). It runs inside the firm's Microsoft 365 tenant, has never been granted permission to send mail, and can also be deployed entirely on premises for firms that require it.

What is automated processing of court documents?

Automated processing of court documents means using artificial intelligence to analyse documents received from the courts, extract the important information, match them to the right client file and prepare the letter intended for the client. The lawyer then keeps the decision to review and send that letter.

QubiLEX applies this principle to a law firm's incoming documents. Unlike a simple sorting rule, it understands the content of the document. It distinguishes the different types of legal act, spots deadlines, extracts the essential information and identifies the right client from the parties named in the document, rather than relying only on the sender of the email.

How does an AI extract data from a court document?

Turning a scanned judgment or a statement of claim into structured information does not rely on a single technology. It is a set of steps that let the AI read, understand and organise a legal document. This is the type of process QubiLEX applies to handle a firm's legal correspondence.

1. Digitisation and text recognition. The first step is to turn the document into usable text. When a document arrives as a scan or a photograph, optical character recognition, known as OCR, extracts the words. The computer then works on textual content it can analyse, rather than on a plain image.
2. Understanding the document. The AI then analyses the page structure, the layout of the elements and the vocabulary used in order to identify the nature of the document. It can recognise whether it is a statement of claim, a judgment, a hearing notice or an order, while locating the important information.
3. Information extraction. The AI automatically identifies the essential data, such as the case reference, the names of the parties, the important dates and the deadlines mentioned in the document. It analyses the text to find this information, even when its wording varies from one document to another.
4. Data normalisation. The extracted information is then turned into directly usable data. For example, a sentence stating that a party has a deadline of thirty days after notification can be converted into a precise date in the firm's calendar. The information becomes operational data rather than a plain passage of text.
5. Verification and human control. Each extracted item is associated with a confidence level. When the AI detects uncertainty, it flags the item so that a staff member can check it. The system therefore does not replace human analysis but helps to quickly identify the points that need particular attention.

This method lets QubiLEX handle the variety of documents received by firms. Formats can vary between jurisdictions, from scanned PDFs to official letters and court decisions. Rather than using a different rule for each document template, the AI analyses the content and adapts to variations while keeping human control over the important information.

What was the firm's situation?

The firm concerned is a general-practice office of five lawyers based in Geneva and working mainly in French. Every working day, the office receives several dozen documents from courts and registries. These include statements of claim, defence briefs, judgments, interim measures, hearing notices, notices of appeal and enforcement orders.

Before QubiLEX was integrated, the processing of these documents rested essentially on the lawyers and the secretaries. Incoming mail was first scanned and then forwarded to the lawyer in charge of the case. That lawyer then had to open the document, read it, identify the case concerned, file it in the right place and notify the client of its arrival.

Each document took about eight minutes of manual processing on average. This task was carried out by professionals whose main added value lies in legal analysis and client advice. With about 600 documents received each month, the firm therefore spent a significant amount of time on repetitive administrative work, representing the equivalent of a half-time position spread across several staff members.

What problem needed to be solved?

The challenge did not come from a lack of skill in the teams but simply from the repetitive nature of certain tasks. Identifying the client concerned, filing a document, checking the main information and preparing an initial notification are necessary operations that draw little on the legal expertise of the lawyers.

This organisation also carried an operational risk. Misreading a deadline, assigning a document to the wrong case file or passing on information too late can have significant consequences in legal proceedings.

The goal was therefore not to replace human involvement but to better share the work. An AI that would automatically write or send letters to clients without approval would represent a risk for a law firm. On the contrary, the solution sought had to take on the upstream administrative tasks while leaving the lawyer responsible for the important decisions.

QubiLEX's role was thus to automate the receipt, analysis and organisation of documents, while maintaining human control over sensitive actions, in particular any communication intended for clients or authorities.

How does QubiLEX work, step by step?

The firm configures the tool once. After that, the processing of newly received documents happens automatically each time an email matching the defined criteria arrives.

1. Connection to Microsoft 365. The firm connects its professional mailbox to QubiLEX. The tool receives only the permissions required for its operation, in particular reading the profile, reading emails, creating drafts and accessing OneDrive storage. The ability to write to the mailbox is used only to place a draft in the Drafts folder. QubiLEX never has permission to send an email on behalf of the user.
2. Definition of authorised senders. The firm selects the email addresses considered trustworthy, for example those of courts, registries or other regular correspondents. The system only analyses messages from these senders, in order to avoid processing irrelevant documents.
3. Selection of monitored folders. The firm chooses the mail folders to monitor, such as the inbox dedicated to court documents. As soon as a new email arrives in one of these spaces, QubiLEX can automatically start its analysis.
4. Registration of clients and parties. The firm adds its clients, whether individuals or organisations. For each client, it can record one or more email addresses, the possible spellings of the name in court documents and the location of the corresponding folder in OneDrive. This information can also be imported in bulk using a prepared template.
5. Reading and information extraction. When an email from an authorised sender is received, QubiLEX analyses the attachments. The tool identifies the document type and extracts the important information such as the case reference, the parties concerned, the key dates and any deadlines to observe.
6. Case identification and automatic filing. The information extracted from the document is compared with the data stored in the system. QubiLEX matches the names of the parties to known clients, identifies the case concerned and then automatically files the document in the corresponding OneDrive folder.
7. Creation of the notification draft:
   Once the document has been analysed and filed, QubiLEX prepares a draft letter to inform the client. The message is created in the language and format chosen by the firm, then placed in the Drafts folder of the lawyer concerned. The lawyer keeps final control, reviews the content and decides whether to send it.

Why does the lawyer always stay in control?

QubiLEX prepares the actions but never sends a letter on the lawyer's behalf. Each message intended for the client is created as a draft in the lawyer's mailbox, where it can be reviewed, edited and sent at the lawyer's discretion. This is a key design choice: in a firm, responsibility for client communication must remain human. QubiLEX also holds no technical permission to send emails.

How is the client identified?

The client is identified from the parties named in the legal document, and not only from the sender of the email. Because the same name can appear in different forms across documents, the firm records the known spellings of party names for each client. QubiLEX can therefore correctly match a statement of claim or a judgment to the right case file, even when the wording differs.

Is QubiLEX secure, and can it run locally?

Security is part of the design of QubiLEX. Its operation rests on three principles that are essential to protecting the firm's data.

Access limited to what is strictly necessary. QubiLEX requests only the access needed to operate: reading the profile, reading emails, creating a draft and accessing OneDrive to file documents. It holds no send permission, which technically prevents any automatic sending of messages to clients.

A controlled scope. The system only processes emails from senders the firm has explicitly authorised. Messages from other addresses are neither downloaded, nor stored, nor analysed. If no sender is configured, no processing is started.

Data stays under the control of the firm. For this firm, QubiLEX runs in its own Microsoft 365 environment. The documents and the extracted data remain in an infrastructure already controlled by the firm, with encryption of data in transit and at rest. QubiLEX does not create an external copy of the case files on a separate infrastructure.

For firms with stricter requirements on confidentiality or data location, the same process can be deployed entirely on premises or in a private cloud controlled by the firm. In this configuration, the documents never leave the chosen infrastructure: reading, extraction, filing and draft creation are all carried out within the firm environment.

Where is the data processed, and what happens to it?

QubiLEX applies a simple principle when processing the firm's data. The tool keeps only the information it needs and processes data as close as possible to its original environment. Three points explain how the documents and the extracted information are processed, stored and protected.

• Reading is performed by Microsoft AI. Reading and information extraction rely on the Microsoft Azure document analysis service, within the same Microsoft environment as the firm's mailbox and OneDrive. Documents are not sent to a third-party AI provider, and the firm's data is not used to train a model.
• Documents are processed and then deleted. QubiLEX keeps only the elements needed for its operation, such as authorised senders, registered clients, party names and folder paths. The content of the documents and the extracted information are kept only for the duration of processing, then deleted. The durable copy remains in the firm's storage, in particular in OneDrive, and not in a parallel database.
• Data stays in Switzerland. Processing and temporary storage of the data take place in the Swiss regions of Microsoft Azure. For firms with particular confidentiality requirements, the whole process can also be deployed on premises or in a private cloud controlled by the firm. In this configuration, the documents never leave the chosen environment.

What does the firm gain from QubiLEX?

After six months of use, from January to June 2026, QubiLEX has allowed this firm of five lawyers to recover the time spent on manually processing incoming documents. The results below reflect the firm's observations over about 3,600 court documents, around 30 documents per working day and 600 per month.

Before QubiLEX, each document required about eight minutes of manual processing to be opened, read, matched to the right client, filed in the right folder and accompanied by a first notification. For 600 documents a month, this represented nearly 80 hours of work per month spent on administrative tasks before the legal work even began.

With QubiLEX, these same steps are automated. Documents are analysed, filed and a notification draft is prepared automatically. The lawyer only steps in to review the message before sending it, which represents about one minute per document.

Over the observed period, 98% of documents were processed end to end without manual intervention, 96% were automatically filed under the right client, and documents containing a deadline, about 40% of cases, had this information identified and turned into a usable date. Documents showing uncertainty, about 2%, were flagged for human review.

The time spent on administrative tasks thus fell to about 10 hours of review per month, allowing the firm to recover nearly 70 hours each month, about 420 hours over six months.

The two tables below present the results observed over these six months, first the volume of documents processed at each stage, then the comparison between manual processing and operation with QubiLEX.

In concrete terms, the roughly 420 hours recovered over six months represent more than ten weeks of work previously spent on the administrative processing of documents. This time can now be devoted to legal analysis and case follow-up. Because the decision to send a letter always stays in the lawyer's hands, this time saving never comes at the expense of human control.