AI automation for business operations in Dubai and the UAE

AI automation consultant in Dubai with five AI systems in daily production, not pilots. I am Head of IT of a UAE plastics manufacturing group, where systems of my design run every day: document OCR that reads supplier bills into the ERP, automated document verification, a 24 hour sales agent on WhatsApp and the web, a voice agent on the corporate telephony, and one chat platform for every channel. I help UAE companies pick the first process and design a pipeline that survives daily use.

Deployments

1. Document OCR into the ERP: supplier bills and expense claims read by vision-capable models and created as records in Odoo for a person to approve. Finance stopped retyping invoices.
2. Automated certificate and document verification that runs every time, leaves a record, and flags rather than decides.
3. A 24 hour AI sales agent on WhatsApp and the group websites that answers, qualifies and creates the lead in the Odoo CRM pipeline.
4. An AI voice agent on Grandstream telephony, built on a phone platform inside my own IT estate rather than a third-party cloud.
5. One AI chat platform running several WhatsApp numbers and the websites from a single place, so adding a brand is configuration rather than a project.
6. Underneath all five: one Odoo ERP across six sites in five countries, on PostgreSQL 9.5 to 16 with no unplanned downtime, with biometric attendance and geo-verified patrol integrated with the same ERP.

What I build: AI bots, AI calling, agentic AI and document AI

Four kinds of system, all in daily production. AI bots: conversational agents on WhatsApp and websites that answer, qualify and write the lead straight into the CRM, plus one chat platform running several numbers and brands from a single place. AI calling systems: a voice AI agent that answers real inbound calls on Grandstream telephony, hosted inside the company's own estate rather than a third-party cloud, so the call recordings and the customer data stay put. Agentic AI: goal-driven systems that pick the next step from a small tool set, with a catalogue they cannot contradict, a hand-off rule for anything about money, and an agreed stopping rule. Document AI: vision-capable models reading supplier bills, expense claims and certificates into the ERP with strict JSON and a human review queue. If you are looking for an AI chatbot, an AI calling agent or an agentic workflow, these are the same four patterns under different names.

Where AI pays back first

Repetitive work with a paper trail: supplier bills and expense claims typed into the system, documents checked by eye, enquiries answered at midnight, calls that only need a fact. Those are the processes where a model plus a validation layer removes hours without removing control. Strategy decks and chatbots on the home page are not where it pays back.

The design that survives production

Capture, extraction with a vision-capable model returning strict JSON, validation rules that reject anything that does not add up, a confidence threshold, a human review queue for everything below it, idempotency so the same bill cannot post twice, an audit log, and a way to measure accuracy honestly every month. Agents get a small set of tools, a catalogue they cannot contradict, and a hand-off rule for anything about money or complaints. Every agent I run creates records for a person to check; none of them post or pay.

What I have run

Document OCR into the ERP in daily production. Automated certificate and document verification. A WhatsApp and web sales agent that qualifies enquiries into the CRM. An AI voice agent on Grandstream telephony. One chat platform running several WhatsApp numbers and the group websites. Plus, on my own venture, an autonomous search and content agent that ships work hourly inside guardrails I set.

How an engagement runs

An AI readiness review picks the processes worth automating and designs the pipeline, the validation rules and the measurement. Then design oversight through build and the first month in production, with a switch-off rule agreed up front. Remote across the GCC or on-site in the UAE. Fees are quoted in writing before anything is paid.

Read the evidence

• Case study: five AI systems in daily production at a UAE plastics manufacturer
• Case study: document OCR into the ERP
• Case study: AI sales agent on WhatsApp and the web
• Article: five AI systems in production, what breaks when nobody is watching
• Tutorial: reading supplier bills into an ERP with an LLM OCR pipeline
• Service: AI in operations readiness review

Questions people ask

Which AI models do you use?

Vision-capable models such as Gemini and Claude for document work, and current language models for agents, chosen per task on accuracy and cost. The model is the least important part of the design; the validation rules and the review queue are what make it safe.

Will AI replace our accounts or sales staff?

No. It removes the typing and the waiting. A person still approves every bill and takes over every conversation about money. The staff spend their time on exceptions and customers instead of data entry.

What does a first project look like?

Usually supplier bills or expense claims into the ERP, because the paper trail makes accuracy measurable. Four to eight weeks from review to the first month in production, with a labelled sample re-checked monthly.

Do you build agentic AI, or just chatbots?

Both, and the difference matters. A chatbot answers; an agent decides and acts. The agents I run pick the next step from a small tool set, write records into the ERP or CRM, hand off to a person for anything about money or a complaint, and stop when a rule says stop. On my own venture an autonomous search and content agent ships work hourly inside guardrails I set. Agentic AI without a stopping rule, an audit log and a human review queue is not a system, it is a liability.

Can you build an AI calling system for our phones?

Yes, and I run one. A voice AI agent on Grandstream telephony that answers inbound calls, handles what it can from a catalogue it cannot contradict, and hands the rest to a person. It sits inside the company's own telephony estate, which keeps call data in your control and avoids per-minute pricing from a third-party voice cloud. The same pattern works on other SIP platforms.

Can you work with our existing ERP?

Yes. The pipeline writes into the ERP through its API. I run this on Odoo and the pattern is the same for other systems with an API.

Talk to me

Email with a few lines about your situation, or message me on LinkedIn. I reply within one working day, Dubai time. Fees are quoted in writing before anything is paid.