An AI readiness checklist for Qatari organisations: five steps from idea to pilot

Many AI initiatives stall between an impressive demo and a system people rely on. The cause is rarely the model. It is usually an unclear problem, data that is harder to reach than expected, or no agreed way to decide whether the pilot worked. These five steps close that gap.

1. Choose a problem worth solving

Good first use cases are frequent, time-consuming and easy to measure: answering the same customer questions, extracting data from documents, routing requests, searching internal knowledge. Write down how the work is done today, how long it takes and what it costs. That baseline is what the pilot will be measured against. Look for:

• Frequency: it happens daily or weekly, not once a quarter.
• A clear output: an answer, a decision or a completed form, something you can check.
• Tolerable errors: a mistake can be caught and corrected, at least during the pilot.
• An owner: one person in the business who wants it to work.

2. Check the data before promising anything

AI is only as useful as the information it can reach. For the chosen use case, find out where the data lives, who owns it, how it can be accessed (an API, an export, a shared folder), how current it is, and how much of it is in Arabic, English or both. Scanned documents, inconsistent formats and missing history are normal; what matters is knowing about them before the timeline is set.

3. Agree the guardrails early

Decide at the start what data the system may use, where it will be processed, which actions need human approval and what will be logged. Where personal data is involved, the questions in our article on Qatar's data protection law apply. Guardrails agreed early speed up approvals later, because security and legal teams review a design rather than a finished system.

4. Build a small pilot, with a test set

A pilot should take weeks, not months. Keep the scope narrow: one channel, one document type, one team. Before building, collect a test set of real examples with the correct answers, and agree the success measure, for example “answers 70% of order-status questions correctly on its own and hands the rest to staff”. The test set lets you compare models and instructions objectively, and catch regressions later.

Involve the people who will use the system from the first week. Their feedback on tone, edge cases and workflow is worth more than any benchmark.

5. Decide with evidence

At the end, compare the results with the baseline from step 1 and the measure agreed in step 4. There are three good outcomes: scale it, adjust it and run another short cycle, or stop. Stopping a pilot that missed its measure is the process working, not failing: it protects the larger investment.

If you scale, plan for what production needs and a pilot does not: monitoring, regular evaluation against the test set, user training, and a clear owner for the system after launch.

Frameworks that help

For a more formal structure, the NIST AI Risk Management Framework (AI RMF 1.0, published in January 2023) organises AI risk work into four functions: govern, map, measure and manage. In Qatar, the National Cyber Security Agency's guidelines on the secure adoption and use of AI and, for financial institutions, Qatar Central Bank's AI Guideline add local expectations.

See where you stand

Our free AI Readiness Check takes a few minutes: ten questions give an indicative score across five areas, the main gaps and a suggested first step.