AI automation in the UAE: what to automate first

Most teams have work that comes back every week in the same shape: a report to compile, invoices to chase, a first draft to write before anyone can edit it. AI automation gives that part to software and keeps a person on the decision. I design and build these systems, and I measure them in hours given back to the team.

At a company I work with (under NDA), one business unit had 30 people. The same work is now done by 11, and the plan is 5. The people freed work on other projects.

What was automated

The client is under NDA, so I stay general. The work fell into four families:

• Content production. The system writes the first version and a person approves it.
• Billing follow-up. The system tracks what is due and which client needs a reminder.
• Invoicing. The system prepares the invoice data and a person checks it before it is sent.
• Monthly reports for clients. The system gathers the figures and writes the first draft.

In each case a person still signs off. Their hours now go to reviewing and deciding. Nobody types the same document by hand every month.

What to automate first

I start with the tasks that pass three tests.

1. The task comes back every week or every month in the same shape. A monthly report qualifies. A one-off negotiation does not.
2. A person can check the output in a few minutes. Checking a drafted report is quick; writing it from zero is the slow part.
3. The data already sits in your tools: accounting, CRM, analytics, shared drives. If the knowledge lives in people’s heads, start with a company brain so the system has something to read.

I leave two things for later: decisions with legal or financial weight that nobody reviews, and processes that nobody has written down. When a process is unclear, the first weeks go into writing it down with the people who do it.

How a project runs

1. In writing first: which tasks, how many hours a week they take today, and what a correct output looks like.
2. A pilot on one task, measured in hours per week before and after, with your team checking every output during the pilot.
3. Production on your own accounts. A person approves anything that leaves the company, and you own the code and the data.

Invoices and client reports carry personal data. The UAE PDPL guide covers what that changes for an AI system.

What it will not do

It does not run without the people who know the work. Someone approves every output that goes to a client, and someone owns each automated task when the process changes.

Automation also does not fix messy data. If the same client appears under different names in different tools, the project starts by cleaning that up.

The case studies below show how I check AI output in production.