AI use case · Public sector
AI that classifies Arabic correspondence and routes it to the right department.
For government entities and large organisations whose registry reads every incoming letter and decides where it goes. AI reads each letter or email, suggests its topic, urgency and owning department, and a registry officer confirms before anything moves.
What is AI correspondence classification?
AI correspondence classification reads incoming letters, emails and scanned documents in Arabic or English. It suggests a topic, an urgency level and the department that should handle each one, with the reason shown. A registry officer confirms or corrects every suggestion. Only confirmed items are registered and routed in your correspondence or document system.
The problem
Every letter waits for someone to read it and decide.
Government entities and large organisations in the UAE receive correspondence through many doors: official letters, emails, scanned attachments, portal submissions and documents delivered to the counter. Most of it is in Arabic, some in English and some in both.
A registry team reads each item, registers it and decides which department owns it. Routing rules are often known by experience rather than written down. When an experienced officer is away, work slows and misrouted letters move between departments.
Urgency is hard to see at intake. A letter with a deadline, or one that refers to an earlier case, looks like any other item in the queue until someone opens and reads it in full.
Manual registration
Sender, reference number, date and subject are typed into the correspondence system for every item.
Routing by experience
Which department owns a topic is known by a few people, not set down as a written rule.
Hidden urgency
Deadlines and links to earlier correspondence are found only when someone reads the full letter.
How it works
One letter, from intake to the right department.
An illustrative workflow. The intake channels, topics, departments and target system are agreed for each project.
Intake
Letters, emails and scanned documents arrive from the agreed channels, such as a registry mailbox, a scanning station or an upload folder.
Read
AI reads the text in Arabic, English or both, and extracts sender, reference number, date, subject and any deadline. Each value links to where it appears.
Classify
It suggests a topic from your agreed list, an urgency level and the owning department, and shows the wording or rule behind each suggestion.
Link
It looks for earlier correspondence with the same reference or sender and suggests the related file.
Confirm
A registry officer reviews the suggestion beside the original document, then confirms it, corrects it or escalates it.
Route
Only the confirmed item is registered and routed in your correspondence or document system, and every correction is logged.
Who does what
What the AI does · What a person checks.
What the AI does
Reads letters, emails and scans in Arabic and English
Extracts sender, reference, date, subject and deadlines
Suggests topic, urgency and owning department, with the reason
Finds earlier correspondence with the same reference
Marks items it is unsure about instead of guessing
What a person checks
Every suggestion, before the item is routed
Items marked confidential or addressed to senior officials
Low-confidence results and unreadable scans
Any change to the topic list or routing rules
First pilot
A bounded first pilot on your own correspondence.
We test on your documents before anyone talks about results. Scope, timeline and price are confirmed in a written proposal.
One intake point
Start with one channel, such as the main registry mailbox, before adding scanning or portal submissions.
A short topic list
Agree a limited set of topics and departments that covers most of the volume, with a catch-all category for the rest.
A real sample
Past correspondence, shared safely after the first call with sensitive items excluded or redacted, split into a development set and a held-out set.
Acceptance agreed first
Agree what counts as a correct topic, department and urgency, and how disagreements are scored, before the evaluation runs.
Data, hosting and Arabic
Where the data goes, and how Arabic is handled.
Hosting
The application and stored data can be hosted on Azure UAE North, on private servers managed by Vloud, or in your own cloud account. The AI model is used through a provider agreed per project with your information security team, and its processing region may be outside the UAE.
Access
Access follows your registry rules. Officers see the correspondence their role allows, and confidential categories can be kept out of the AI workflow entirely.
Arabic
Arabic, English and mixed documents are tested as separate samples. Handwritten Arabic and poor scans are measured on their own, not averaged in with clean documents.
Retention and audit
What the workflow stores, for how long and who can see it follows your records policy. Each suggestion, confirmation and correction is logged.
Honest limits
Where it does not work yet.
- Handwritten letters, stamps over text and poor scans often need a person to read them.
- The AI suggests; it does not decide. Routing happens only after a registry officer confirms.
- We do not quote accuracy or time savings before testing on your correspondence. Results are measured on a held-out sample.
- Topics that depend on policy judgement, rather than on what the letter says, stay with your staff.
- If your correspondence system already offers classification that covers the need, configuring it may be the better first step, and we will say so.
What drives cost and timeline. The number of intake channels, topics and departments, the share of scanned and handwritten documents, languages, the correspondence system's integration route, confidentiality rules and hosting requirements. We don't publish a price list; scope, timeline and price come in a written proposal after the discovery call.
Evidence
What we can show you.
Every example is labelled: delivered work, our own product, a demonstrator, or an illustration with synthetic data.
- Illustrative
AI document automation in the UAE
Where AI document automation works, where it struggles and how to run a safe first project. It is an article, not a client project.
- Illustrative
How to test AI document extraction
The evaluation method we apply in a pilot: a development set, a held-out set and field-level results reported with the review rate.
- Synthetic data
The illustrative AI workflow
The same read, check and review pattern, shown step by step with a synthetic invoice. It is not correspondence and not a live client connection.
Before you start
Questions buyers ask.
Does it read handwritten Arabic?
Sometimes, depending on the handwriting and the scan. Handwritten items are tested as their own sample in the pilot, and anything the system cannot read with confidence goes to a registry officer.
Can it work with our correspondence system?
If the system offers a supported API or an agreed import route, confirmed items can be registered there directly. We review your system and version before confirming the integration scope.
Who decides where a letter goes?
A registry officer. The AI suggests a topic, urgency and department with the reason shown, and nothing is routed until a person confirms it.
How are confidential items handled?
You decide which senders, categories or markings keep an item out of the AI workflow. Those items follow your existing manual route.
Can the topics and departments change later?
Yes. Topics and routing rules are kept as a list your team owns. Changes are tested on a sample before they go live.
Where is the data hosted?
The application and stored data can be hosted on Azure UAE North, on private servers managed by Vloud, or in your own cloud account. The AI model is used through a provider agreed per project, and its processing region may be outside the UAE.
Who owns the code?
You own the source code and IP of the custom work we build for you. Our own products stay ours, and any licence terms are agreed separately.
Related
Related AI use cases.
Rule-based design compliance checking
Check submitted designs against codified rules, with every finding linked to the rule and the drawing, and a reviewer making each decision. Rule-based.Emirates ID, trade licence and passport data extraction
Read Emirates IDs, trade licences and passports into your onboarding system, with field checks, role-based access and a person reviewing every uncertain value.HR and policy assistant in Arabic and English
Staff ask HR and policy questions in Arabic or English and get answers only from your approved documents, with the passage shown and access set by role.Government and public sector
Workflows for UAE public-service teams: requests, routing and traceable decisions.AI document processing
The service page for document extraction, checks and review.
Next step
Test it on your own documents.
Tell us what the work looks like today. We reply with the questions we need answered before a pilot can be scoped.
