AI readiness assessment checklist for UAE teams

AI readiness assessment · UAE

An AI readiness assessment asks whether a workflow has a clear outcome, usable source material and an acceptable way to handle mistakes. For a UAE team deciding what to automate, I would start with one recurring task and a written decision about its next step.

You can work through this checklist yourself, with the person who does the work and the person who owns the outcome. If you would rather have me run it, my AI workflow readiness assessment uses the same method on one workflow. A workflow audit can end with a simpler process, ordinary automation, a limited AI proof or a decision to keep the work manual.

The one-workflow checklist

Check each line for one workflow. Any line you cannot tick is the next thing to fix before you spend money on a build.

You can name the trigger, inputs, responsible person and finished outcome.
You have timed ordinary and difficult runs, and counted corrections and reopened cases.
You have compared a simpler process and ordinary automation with the AI-assisted version.
You know where each input comes from, who may use it and which provider would receive it.
The records are complete, current and consistent enough for someone to check the output against them.
If the team works in Arabic and English, the test cases include both.
Permissions are written as actions, with a named approver, a fallback and stop conditions.
Test cases and pass criteria were fixed before anyone saw model output, with a few cases held back.
You have written down what happens if the result is late, wrong or sent to the wrong person, and how you would contain it.
The AED comparison includes review, correction and maintenance time.
The decision is written down with its reasons and the person who owns the next step.

Use the AI readiness assessment template with the operator and sponsor, or compare the completed synthetic example.

Choose a workflow you can observe

“Use AI in operations” is too broad to test. “Prepare the weekly delivery-exception report from approved exports, ready for the operations lead to check” has a beginning and an end.

Write down the trigger, the inputs, the person responsible and the completed outcome. Include what happens after the draft is produced. If a person still needs to correct figures, chase missing records or send the report, that work belongs in the assessment.

For a team in Dubai or elsewhere in the UAE, record costs in the currency used for the decision. If the team works in both Arabic and English, include representative cases in both languages. Test the files and terminology the operator actually encounters.

Measure the current work

Observe several ordinary runs and some difficult ones. Record active handling time separately from waiting time. Count corrections and reopened cases. Ask the operator which steps the written procedure leaves out.

Record these inputs before comparing alternatives.
Input	What to record
Volume	Cases per week, peaks and the observation period
Time	Preparation, checking, corrections and handoffs per case
Cost	An agreed hourly cost in AED and current software costs
Quality	Errors, missing information and cases that need escalation
Consequence	What happens if the result is late, wrong or sent to the wrong person

Reclaimed staff time is capacity. It becomes a cash saving only if expenditure actually falls. Keep those two claims separate in the decision.

Compare the simplest alternatives

Try removing an unnecessary step first. A spreadsheet formula or a fixed rule may solve a reconciliation problem more reliably than a language model. AI may help when the work involves varied text or interpretation, provided someone can check the result.

Compare the current process, a simpler process and a limited AI-assisted version using the same cases. Include review time and maintenance in the comparison. A quick first draft can still create more work downstream.

The workflow automation guide works through which steps can use rules, where a model might help and when an agent would need more freedom.

Map data and authority

List each input, where it comes from, who may use it and which provider would receive it. Use synthetic data or expressly approved non-production examples for the first proof. A local model still needs access controls and a retention decision; “local” alone does not settle how data is handled.

Then check data readiness for the task itself. Data quality is a practical question here: are the records complete, current and consistent enough for someone to check the output against them? Note missing fields, duplicates and conflicting sources in the baseline, because any later run on real records will meet them too.

Write permissions as actions. In a reporting proof, the agent may read approved exports and draft a report. It may not change source records or send the report. Name the person who approves the final output, the fallback if the tool fails, and the conditions that stop the run.

This is an operational assessment of one workflow, not a company-wide AI governance programme. Questions requiring legal or regulatory sign-off need the appropriate qualified reviewer.

Freeze the test before running it

Choose the cases and acceptance criteria before seeing the model's output. Include missing inputs, duplicates, conflicting figures and instructions embedded in source text. Keep a few cases out of prompt development so the final check is not just a repeat of the examples used to tune it.

Every reported number must trace to an approved source row.
Missing or conflicting source material must be flagged for a person.
The agent must stay inside its read-and-draft permissions.
Total preparation, checking and correction time must meet the agreed target.

These are example criteria for a report, not universal thresholds. Agree the actual thresholds with the workflow owner. Preserve failures alongside successful outputs. A small test can justify another controlled trial; it cannot prove that rare failures will never occur.

Work through the economics in AED

Suppose a weekly report takes three hours to prepare and one hour to check. At an assumed staff cost of AED 100 per hour and four runs per planning month, the current labour cost is 4 × 4 × 100 = AED 1,600 per month.

Now suppose an assisted workflow takes one hour to prepare and one hour to check each week, plus two hours of maintenance per month. Its labour cost would be (2 × 4 + 2) × 100 = AED 1,000. Add an assumed AED 200 in monthly tool costs: the total is AED 1,200.

The difference is AED 400 per planning month before one-off setup costs and unexpected rework. With an illustrative AED 4,000 setup cost, simple payback would be ten months if those assumptions held. One additional hour of corrections each week would consume the entire AED 400 difference.

That sensitivity is the point of the exercise. Measure the correction burden before treating the headline time reduction as a reason to build. See the worked reporting example for the permissions and test failures that go with this calculation.

Write a decision someone can act on

Keep missing information visible in the recommendation.
Decision	What supports it
Proceed	The agreed tests pass, the economics support the next stage, and named owners accept the controls. Production still needs its own approval.
Proceed with limits	Only a defined subset works. Record the excluded cases, human checks and test results needed before expanding.
Do not automate	The simpler option wins, costs outweigh the benefit, or a material failure cannot be contained.
Assessment incomplete	The assessment cannot support a decision yet. Name the missing input and the next check; do not count it as a pass.

The handoff should contain the baseline, test cases, outputs, failures, cost assumptions and the person responsible for the next decision. Keep the approved scope beside the results so a later implementation cannot widen it.

Use the AI readiness assessment worksheet with your team

Download the workflow-readiness worksheet (Markdown). Fill it in with the operator who does the work and the sponsor who owns the outcome. It includes the baseline, alternatives, permissions, evaluation record and decision.

For UAE teams already using AI that want help applying this method, my ten-working-day Verified AI Operations Sprint assesses one workflow and tests a small non-production proof. Implementation is separately scoped after the decision.

A completed AI readiness assessment example

Download the completed synthetic assessment beside the blank readiness worksheet. The example records an incomplete decision rather than treating planned tests as measured passes.

Its task is the weekly delivery-exception report. The operator has a candidate export and an agreed read-and-draft boundary, but has not timed representative runs or checked the holdout cases. The next decision is to collect that records. The example's AED assumptions illustrate the calculation already shown above; they do not establish savings.

Is this an organisation-wide AI maturity assessment?

No. This template assesses one recurring workflow. Company-wide strategy, infrastructure, procurement and governance may need separate work. Narrowing the first decision makes its records easier to inspect.

What if the data is missing?

Record which input is absent, who can supply it and how the absence affects the decision. An incomplete assessment is not a failed business and not permission to automate.

For the implementation estimate, use the AED cost worksheet. For behavioural records, use the bilingual agent-testing guide.

Suhail Mohebi · published 31 October 2026 · the cover is a generated illustration