Skip to content

AI at Al-Arjan

AI agents that work inside your systems, not beside them

They understand your customers and staff in Arabic and Iraqi dialect, search your real data and get the task done — then hand over to a human when they should.

AI Agent
Customer message
Understand
Search your data
Take action
Reply & log

See it work

One conversation — and everything behind it

Pick a business type and follow the conversation on one side and the steps the agent runs in your systems on the other.

Al-Arjan assistant

online

What the agent does behind the scenes

  1. Waiting for the customer’s message…
A pre-scripted illustration of how it works — not a live conversation; prices and numbers are illustrative only.

Bot or agent?

Choose the level your business needs

Traditional bot vs AI agent vs staff member

  • Understands free-form questions

    Menu bot:
    Weak
    AI agent:
    Strong
    Staff member:
    Strong
  • Performs tasks in systems

    Menu bot:
    Very limited
    AI agent:
    Yes, within permissions
    Staff member:
    Yes
  • Availability

    Menu bot:
    24/7
    AI agent:
    24/7
    Staff member:
    Working hours
  • Complex & sensitive cases

    Menu bot:
    No
    AI agent:
    Hands off to a human
    Staff member:
    Yes
  • Cost per extra conversation

    Menu bot:
    Very low
    AI agent:
    Low
    Staff member:
    High

In detail: AI agent vs chatbot

Safety first

Six principles that make an agent trustworthy

AI without limits is a risk. These limits are designed into every agent we deliver.

Answers only from your data

Prices, policies and availability come live from your systems and documents, not model memory.

Scoped permissions

Specific tools: read stock, create a booking. No open database access.

Human approval for sensitive actions

Refunds, discounts and data changes wait for staff approval.

Human handoff

On ambiguity, frustration or request — with a full conversation summary.

Your data stays private

Models configured not to train on your data per provider terms, or open models on your servers.

Logs and review

Every conversation and action logged, with regular answer audits.

Channels and models

Where your customers talk — with the right model for the job

Channels

  • WhatsApp
  • Telegram
  • Messenger
  • Instagram
  • Your website
  • Your app & internal tools

Language models

  • Claude
  • Gemini
  • GPT (OpenAI)
  • Llama
  • Mistral
  • Ollama

We’re not tied to one provider; we pick the model by language, accuracy, cost and data sensitivity.

How we work

From one task to a trusted agent in 6 stages

  1. 1 week

    Pick the task

    We choose one repetitive task eating your team’s time, and its success metrics.

    Written task and metrics

  2. 1–2 weeks

    Prepare knowledge and tools

    We collect real questions from your chats, clean documents and define tools and permissions.

    Knowledge base and tool map

  3. 1–2 weeks

    Prototype

    An agent running on your real data, tried internally by your team.

    Internal pilot agent

  4. 1 week

    Test on real conversations

    We measure accuracy and failure cases and iterate until you sign off.

    Signed-off accuracy report

  5. 2–4 weeks

    Limited launch

    Launched to a share of customers with daily monitoring and fast handoff.

    Live agent with real metrics

  6. Ongoing

    Scale and improve

    New tasks and channels, with monthly review of unanswered questions and accuracy.

    Monthly report and improvements

How we measure success

Numbers you review monthly — not promises

  • Resolved without staff
  • Handoff rate and reasons
  • Answer accuracy in audits
  • Average response time
  • Bookings and orders completed
  • Customer rating

Frequently asked questions

Questions about AI agents and bots

Does the agent understand Iraqi dialect?

Modern models handle Iraqi dialect well in most cases, and we test on your customers’ real conversations before launch to fix weak spots.

Can the agent make mistakes?

Yes, like any system — so we constrain it to your data and specific tools, require human approval for sensitive actions, and audit conversation samples regularly.

Does the agent replace customer service staff?

It works best taking the repetitive questions so staff can focus on cases needing human judgement and relationships.

Which channels does it work on?

WhatsApp, Telegram, Messenger, Instagram, your website, your app and internal staff tools.

Can we use our current number?

Usually yes after migrating it to the API, but it then won’t work in the regular app the same way; we plan the move with you.

Do numbers get banned because of bots?

Bans usually come from unofficial tools or bulk-messaging people who didn’t opt in. We use the official API and follow Meta’s policies.

How long does a WhatsApp bot take?

A simple menu bot can take around two weeks after verification completes; an AI bot connected to your systems 3 to 6 weeks for a pilot.

Insights

Read before you decide

AI agent vs chatbot: a clear guide for decision makers

A menu bot replies from fixed options, an AI chatbot understands free-form questions and answers from information, and an AI agent understands, decides and uses real tools to complete the task inside your systems: booking, logging orders, checking stock. Choose an agent when you want tasks done, not just answered.

Read article

WhatsApp bots for businesses in Iraq: the complete guide from idea to launch

To launch a WhatsApp bot in Iraq: use the official WhatsApp Business API via an approved provider, verify your business with Meta, prepare templates for business-initiated messages, then build a menu or AI bot connected to your systems with human handoff. Avoid unofficial tools and bulk messaging non-opted-in contacts so your number isn’t banned.

Read article

How much does a chatbot or AI agent cost? The cost breakdown before you decide

A chatbot or AI agent costs four things: build and system integration (one-off), language-model usage by conversation volume and length (monthly), channel fees such as WhatsApp (monthly), and maintenance plus knowledge-base updates. Compare it with the staff time saved and the sales it closes outside working hours.

Read article

What is RAG? The technique that makes bots answer from your data instead of inventing

RAG, retrieval-augmented generation, makes a language model answer from your documents and data: for each question, your knowledge base is searched for the most relevant passages first, then the model is told to answer only from them. The result: up-to-date answers specific to your company, source citations, and far less fabrication.

Read article

Bots and Iraqi dialect: building a bot that understands how Iraqis actually write

Modern language models understand Iraqi dialect well in most everyday conversations, but the challenge is in the details: non-standard spelling, Arabic mixed with English and numbers, and local product and area names. The fix: test on hundreds of real customer conversations, add a glossary of your products and areas, and hand ambiguous messages to staff.

Read article

Making an AI agent safe: permissions, boundaries and human handoff

A safe agent gets the least privilege its task needs, requires human approval for any sensitive action like refunds, discounts or data changes, sees only the data it must, doesn’t obey instructions smuggled into user messages or documents, logs everything it does for review, and hands off to a human when in doubt.

Read article

12 practical uses of AI in Iraqi businesses today

The most successful AI uses in Iraqi businesses today are repetitive tasks whose data you already hold: answering customers on WhatsApp, qualifying ad leads, booking appointments, tracking orders, helping staff from internal documents, extracting invoice data and summarising reports. Start with the one eating most of your team’s time.

Read article

Let’s start with one agent for one task — and measure it

Tell us the task eating most of your team’s time, and we’ll propose a limited pilot with clear metrics before any big commitment.

Our team is available for replies and support 24/7