We partnered with a mid-size law firm based in Riyadh with offices in Jeddah and Dubai, having over 800 active cases in Saudi and UAE courts to that effect. The firm dealt with commercial disputes, labour cases, real estate litigation, and corporate matters in several different court systems, each with different filing rules, different portal behaviours, and different deadline logic. Their lawyers and legal assistants were terrific, but the business model was still held together by spreadsheets, manual portal monitoring, and relentless follow-up.
That approach was working as far as it could. They each represented 30-50 cases. For hours, legal assistants were also signing into Najiz, Dubai Courts, DIFC Courts, and Abu Dhabi court systems to know if hearing dates had been rescheduled, new orders had been released, or a filing deadline had been initiated. The firm figured that around 5-8% of due dates were always in danger of falling through the cracks somewhere in the system, and two to three real deadline misses were occurring each year. It’s not an inconvenience in the courtroom. It’s a professional and commercial menace.
At aTeam Soft Solutions, we created an AI agent that perpetually monitored court portals, parsed deadlines out of Arabic legal orders, drafted submissions, verified formatting compliance, and even filed electronically when allowed. The outcome was easy to articulate but proved to be tremendously beneficial in reality: no missed deadlines in 18 months, significantly quicker document production, fewer filing errors, improved client reporting, and a much stronger operating system for the whole firm.
Before we designed anything, we went to work understanding how the firm’s work really did flow between hearings, between filings, and between courts. What we saw was a very common law firm operational issue: amazing legal talent surrounded by a process layer that had become too manual for the complexity of the practice.
This wasn’t a law firm that was overseeing a single court system. It was managing many. Saudi General Courts, Commercial Courts, Labor Courts, the Board of Grievances, as well as UAE-side practice through DIFC Courts and Dubai and Abu Dhabi courts. Different rules apply at each forum. Each portal is unique. Each type of document has expectations. Some issues were resolved exclusively through electronic filing. Others still involved a mixture of physical processes. Some orders were clean digital entry. Some (orders) were scanned or handwritten in official legal Arabic. Some deadlines run from hearing dates, others run from service dates, or filing dates, or from the date an order is issued.
Within the firm, the lawyers and legal assistants were burdened with much of the practical work.
Every lawyer had 30 to 50 active matters each. The legal assistants had a massive Excel spreadsheet with over 800 rows, which included case number, court, parties, next hearing, filing deadlines, responsible lawyer, and procedural notes. Whenever a hearing concluded or a new filing came in, somebody had to log in and manually update that spreadsheet. Whenever a court order was entered, somebody needed to read it, determine whether it required a response, do the math on the deadline, and plug that information into the tracker. If a hearing date was moved, the spreadsheet needed to be updated again. If an opposing party filed something, someone needed to spot it.
The daily surveillance workload was severe.
Legal assistants were signing into multiple court portals every day. In the Saudi context, that means checking into Najiz and some other related court-service platforms. On the UAE side, it involves Dubai Courts, DIFC e-Registry, Abu Dhabi court system, and others that have their own similar platforms. They weren’t logging in to do legal work. They logged in merely to check and see if anything had happened.
Now, that may seem just a routine administration, but in a litigation practice, it is control of the risk. When a court issues a new direction, changes a hearing, requests a memorandum, or triggers a response time limit, failing to receive that notice can adversely affect a client’s position immediately.
The problem was not only quantity. It was an interpretation.
Court orders are not always written in plain procedural language. Orders of the Saudi court may be in classic informal legal Arabic using abbreviations, references to provisions, etc., and written in a manner that presumes a legal background. Deadlines aren’t always given in a straightforward manner. An order can read “15 days from service to respond,” which is not the same as saying “15 days from publication.” In the UAE, the DIFC procedure differs in absolute contrast with that of Saudi courts. Weekend rules differ. Submission procedures differ. Holiday periods vary. A practice direction of one court is irrelevant to another.
That meant that the spreadsheet was only as good as the last human who got the last procedural event right.
Preparing documents was another significant operational strain.
As a deadline neared, the team assigned to the file had to read through the file, figure out what the court was asking for, pull relevant models, develop the draft, format it correctly for the relevant court, check the attachments, and then file. In Saudi cases, the main drafting language was Arabic legal prose. In DIFC issues, it was a question of English drafting and compliance with the DIFC Rules that mattered. There was no question of treating the formatting loosely. Using the wrong cover page, the party naming format, a missing annex, the wrong case reference, incorrect fees, or a missing power-of-attorney attachment can lead to rejection or delay.
The people at the company were handling it all by effort and memory, but they could sense the strain. They calculated that 5 to 8 percent of deadlines were “at risk” at any given time, not because anyone wanted to miss them, but because the system required perpetual manual monitoring. With another 2-3 real deadline misses occurring each year. Any one of them could lead to client injury, reputational harm, and internal stress far beyond the level of the administrative blunder that triggered it.
When we tackled the problem at aTeam Soft Solutions, that’s just not how we saw it as a reminder-system problem. We saw it as a legal operations issue. The firm wanted an AI agent that could constantly monitor the courts, interpret procedural events, anticipate the next steps, and leave the ultimate legal judgment where it belongs: with the lawyer.
The firm had previously used practice management software, including a popular platform designed primarily for US and UK firms. It helped to organize matters and with overall legal management, but it didn’t practically know Saudi or UAE court procedure. It had no idea about the character of Najiz, how Najiz behaves, how Saudi Commercial Court deadlines are calculated, what the requirements are for Arabic documents, or how DIFC and Dubai Courts really differ operationally.
Calendar alerts weren’t good enough either. A calendar can inform you of a date you already have in mind. It can’t reliably find that date for you when the date is being set by a new court order, service confirmation, judicial instruction, or procedural rule that varies by forum. When a law firm has 800+ cases, each with multiple possible deadlines, a calendar is a heavy source of noise unless the dates contained therein are already correct. That was the deeper issue here.
And they also didn’t have a tool to constantly monitor the court portals of Saudi Arabia and the UAE and turn what was there into structured, reliable legal work. That left the real work still manual: monitoring portals, reading orders, doing the calculations on deadlines, and entering those duties into the tracker.
That’s why the fix had to be more than just a case database or a reminder layer. The company wanted a true court deadline tracking AI agent, one that could live in the midst of court updates, procedural rules, case documents, and lawyer workflow. This was a true AI legal case management problem shaped by the legal culture of Saudi Arabia and the UAE, not a generic law office software issue.
We developed the system in stages because the legal teams are rightly cautious. There isn’t a serious law firm that wants a system filing or drafting autonomously on its own from day one. So the workflow is designed to first increase visibility, then support drafting, and then manage, when safe, permit filing, and finally add case intelligence.
The first stage concentrated on the most dangerous issue: the firm discovering it was making too many procedural changes manually.
We integrated the AI agent with the relevant court environments: Najiz and related Saudi workflows, Dubai Courts, DIFC Courts, and Abu Dhabi court systems. System performed daily checks on all open matters to identify any updates to hearings, recently issued orders, submissions from the opposing side, and response windows triggered by services and procedural instructions.
That sounds straightforward until you keep in mind that court events don’t come in neat packages of task objects. They come as portal posts, scanned orders, handwritten Arabic notes, status updates, or hearing transcripts that must be interpreted. So the monitoring layer was not simply “portal scraping.” It was a legal event extraction.
Whenever the AI agent discovered a new court order or status event, it parsed the content, determined if a deadline was established, and computed the due date using the appropriate procedural logic for that court. Those rules included weekend differences, holiday exclusions, service-date dependence, and other forum-related timing rules. A reply period in Saudi litigation is not calculated like a deadline under DIFC rules. The system had to recognize that.
This stage also marked the beginning of a cascading-escalation mechanism. After a real deadline had been set, the AI agent would not simply store it in a database. It ran it through a timed legal risk workflow: advance notice to the assigned lawyer, escalation to the supervising partner, and emergency escalation where needed. The objective was not to receive more notifications. The aim was a reliable legal clock.
Now this was the first point where the company really felt the value of operational. Rather than legal assistants dedicating hours just to verify whether something had changed, the system continuously presented what really mattered.
The second stage addressed the next operational bottleneck: preparation of the document.
As a deadline for filing neared, the AI agent would examine the case file, the most recent court direction, earlier submissions, and the applicable litigation posture and would write a first draft of the necessary document. In matters of Saudi Arabia, it was written in Arabic and aligned with the hierarchy of the relevant court. In DIFC matters, it drafted the English-language documents in accordance with DIFC procedure.
That doesn’t mean the system was choosing legal strategies for the lawyer. What it was doing was 60 to 70 percent of the structured drafting that normally consumes so much of the team’s time. The designated attorney subsequently examined, revised, strengthened, and completed the document.
We also formed a formatting compliance layer around this. And it turned out to be surprisingly valuable. Before a document could be filed, the AI agent verified if the file complied with the target court’s formal rules: proper case reference, proper party designation, necessary attachments, cover sheet structure, language layout, page numbering, and other court-specific expectations. It cut down on frustrating but common rejections of filings, not because the legal argument was weak, but because the packaging was wrong.
This stage altered the work process in a very practical way. Lawyers weren’t opening a blank page quite as often. They were starting a more structured version of the first draft with the appropriate framework already set up.
At Team Soft Solutions, we believe that is the role of an AI agent in legal drafting is speed, structure, and preparation, and not the removal of lawyer judgment.
Once monitoring and drafting became more stable, we enabled the system to join filing workflows when the court environment allowed it.
For the electronic filing portals, such as Najiz and DIFC e-Registry, the AI agent can take the final lawyer-approved document, submit it via the appropriate portal, capture filing confirmation, download the receipts or stamped acknowledgments, and automatically file those documents back into the matter file.
If the court still required physical submission or hybrid processing, the system prepared the package to be filed instead: the correct number of copies, the fee expectation, the filing checklist, and the task instructions for the runner or legal assistant who would take the document to the court. So even where there was no ability to file fully digitally, the administrative preparation work still became much faster and more uniform.
This stage also enhanced control after filing. The AI agent was monitoring confirmations and acknowledgments, so the firm no longer needed to manually track whether a filing had successfully been received in the court system.
The last big phase extended the system beyond deadline safety and then further to litigation intelligence.
We leveraged the firm’s historic matter results and publicly available case patterns to equip the AI agent with trend identification: durations of cases by forum, arguing patterns with stronger outcomes, typical settlement behavior, and lawyer workload pressure. It was in that context that the system was used for strategy discussions, workload management, and client reporting.
That didn’t make for a simplistic predictiveness in litigation. Courts aren’t spreadsheets. And it did provide the firm with a bit more structured insight about timing, anticipated procedural paths, and case management pressure.
We automated client reporting as well. Clients could see a summary of the status of a matter each month, including what had been done, what hearings were upcoming, what major next steps were planned, and what the timeline expectations were. That built client confidence because the firm was not depending on partners or associates to manually pull together each and every status report from scratch.
This is what made the project more than a deadline system. It turned into a true agentic AI law firm operating layer in Saudi Arabia for litigation and dispute management.
We developed the platform in Python, with FastAPI acting as the orchestration layer, Celery and Redis as the continuous monitoring and scheduled processing mechanisms, PostgreSQL as the storage for structured data related to matter, deadline, and filing, and Elasticsearch for precedent, filing, and case searches.
The Playwright dealt with the court-portal automation across Najiz, DIFC Courts, Dubai Courts, and related environments. As the court portals will change, the automation layer is designed to be robust and monitored, rather than just being a fragile script. Where there were handwritten Arabic court orders, we applied an OCR and document-intelligence pipeline that reads Arabic handwriting, then passes that interpretation through Claude and our legal-language rules layer.
Claude Opus managed an Arabic reading of the legal document, first-draft generation, interpretation of court order, and summarization of procedures. But we also developed a custom Arabic legal NLP module, as legal Arabic, as used in Saudi court practice, does not behave like an ordinary Arabic text. That distinction made all the difference in extraction quality.
The React dashboard was now the work surface for lawyers, legal assistants, and partners. Active matters, upcoming deadlines, court status updates, escalation flags, filing status, and workload pressure were all visible in one operational view. Rather than working from a giant spreadsheet and scattered around on the portals, the firm now had a real-time litigation command layer.
And most importantly, each and every AI-generated draft was clearly flagged as draft material that needed to be reviewed by a lawyer. No content could advance to filing without the express approval of a named lawyer. That safeguard was not transitory. It was part of the permanent architecture.
One of the most challenging aspects of the project was Arabic handwritten court orders. Saudi court records are not always just plain, clean typed text. OCR is always challenging on handwritten Arabic notes of any kind, more so when they contain abbreviations or legal shorthand. We solved this with a multi-stage pipeline: firstly, Arabic handwriting recognition; secondly, domain-specific legal interpretation; and thirdly, procedural-rule validation. If confidence were still low, the system explicitly routed the item to a human reviewer rather than pretending certainty.
Procedural rules also posed a significant challenge. There was not a single “court rules” model that would cover everything. The procedures of the Saudi Commercial Court are not the same as those of the Labour Court. DIFC practice differs from that of the Dubai Courts. Even within a forum, individual practice directions and filing habits matter. We collaborated with senior lawyers at the firm to capture hundreds of procedural rules and encode them into a rules layer that the AI agent could reference.
Risk was also generated by service dates. In some Saudi cases, the date that starts the response clock is not the same as the date the document first appears on Najiz. We thus incorporated service-date tracking in the workflow so the deadline engine could rely on the date of actual service confirmation instead of a misleading portal timestamp.
And, as expected, the lawyers were cautious of AI drafting. We have agreed with that caution. Legal drafting is not a thing that should go straight from model output into a court filing. It is up to the design to ensure that clear legal responsibility is maintained at every step.
The most significant outcome was straightforward: there were zero missed deadlines over the course of 18 months compared to 2-3 per year previously. For a litigation practice, that one change is worth a tremendous amount in protecting clients, professional confidence, and risk control management.
Monitoring deadlines themselves became completely automated across 800+ matters and multiple court systems. The firm didn’t need legal assistants to spend four hours a day just checking the portals. That time was cut and then reallocated to more substantive assistance work.
Drafting time also declined significantly. Lawyers said their time spent preparing documents dropped by approximately 55% because they were reviewing and editing structured drafts instead of writing standard procedural documents from the beginning. That did not diminish the quality of the law. It also reduced the organizational hassle.
Filing errors declined significantly. Incorrect formatting, missing attachments, the wrong fees, and similar kinds of issues fell from about 12% to 1%. Rejection rates for court filings declined from 8% to 0.5%, mostly because the formatting and compliance checker caught preventable problems before they got to the court.
Each lawyer could effectively manage about 20% more active matters because much of the deadline tracking, portal checking, and first-draft assembly had been shifted over to the AI agent. Eight legal assistants were reassigned from manual tracking work to more substantive legal research and case support.
Client satisfaction increased as well, moving from around 3.8/5 to 4.6/5. But the reason was that it wasn’t just speed. The clients noticed when the firm became more aggressive, more organized, and more assured in its communications on the substance of its matters.
The company also landed three big corporate clients following the system’s demonstration during pitches. This is worth pointing out because predictable operations have come to distinguish legal markets, particularly for business clients with substantial litigation exposure in Saudi Arabia and the UAE.
We develop the system using Python with FastAPI, Celery and Redis for asynchronous case monitoring and workflow orchestration, PostgreSQL for case and deadline storage, Elasticsearch for precedent and filing search, Playwright for court portal automation, Claude Opus for Arabic legal analysis and first-draft generation, custom Arabic legal NLP layers, React.js for the case management dashboard, and AWS EC2, RDS, and S3 for infrastructure, storage, and document processing.
The biggest takeaway was that lawyers don’t need or want “automatic legal drafting.” They want to work faster and better without losing control.
That is why we approached the system as a first draft and an operations engine, instead of an autonomous legal author. Every paragraph that is AI-generated remained clearly labeled as draft content that needs to be reviewed by a lawyer. That framing made for a much easier adoption and is aligned with the professional reality of practicing law.
We also came to understand that the true value of legal operations is not just in drafting. It’s about the certainty of the procedure. The worst failures in legal operations frequently have nothing to do with the quality of the argument. They derive from dates that have been missed, from updates on portals that have been missed, from formatting that isn’t quite right, or from responses to court directions that are a day too late. Fixing that layer creates immediate risk reduction.
And lastly, we found out that everything is about local procedural knowledge. A legal AI application designed for London or New York does not become relevant for Riyadh or Dubai simply by including Arabic. It must be familiar with local courts, local filing culture, local rules, and local language forms. This is the reason why aTeam Soft Solutions considers legal document automation Middle East solutions as a very contextual challenge rather than a generic layer of productivity.
Some law firms in Saudi Arabia, the UAE, and the rest of the Middle East tend to view deadline risk as an inherent part of the complexity of litigation. But in reality, much of that risk, though, is the result of manual monitoring of portals, disparate case tracking, and drafting workflows that require too much time for administration.
At aTeam Soft Solutions, we design and develop systems in which an AI agent is constantly monitoring the courts, extracting procedural obligations, drafting filing preparations, and even assisting lawyers without substituting their judgment. This is how a litigation practice becomes safer, quicker, and more scalable — with the result that the lawyer is left firmly in charge of the legal decision.
