How UAE government entities scope AI work before the tender.

What to settle before a UAE government AI tender goes out.

An AI idea reaches procurement either naming a process, a team and a number to move, or naming a technology and asking the market what to do with it. The second kind draws proposals that answer different questions, hard to compare and harder to hold to account. Abu Dhabi's procurement FAQ gives one more reason to settle the scope early: it may change only before the tender closes, and a changed scope after that means publishing the tender again (question 22).

The five settlements below turn an idea into a scope, and none needs a vendor in the room. Our AI consulting in the UAE work starts at the same point: before the tender is written, while the scope is still easy to change.

One workflow, one owner.

Describe one workflow the way the people who run it would: what arrives, who touches it, what leaves, where it waits. "Correspondence triage in the licensing department" is a scope. "AI for customer service" is a direction.

Then record today's numbers for that workflow: volume per week, time per item, rework, backlog. They become the baseline, cheap to collect now and expensive to reconstruct after go-live.

Name two owners: the business owner who judges whether the workflow improved, and the unit that keeps the system running after handover. A system with no named keeper drifts as models change.

The data the workflow touches, and where it may be processed.

List the data classes the workflow reads and writes. Government data sits outside the federal personal data law: Article 2 of Federal Decree-Law No. 45 of 2021 (the PDPL) excludes government data and government entities from its scope. The classification scheme to look for is then the emirate's or the entity's own.

In Dubai, Resolution No. (2) of 2017 approves the policies document on classification, dissemination, exchange and protection of data in the Emirate of Dubai. Under its Article 15, an entity classifies a data set as Open Data where its dissemination is in the public interest and meets listed conditions; Article 16 splits Shared Data into Confidential, Sensitive and Secret. Outside Dubai, ask the entity's data office which scheme applies and write its answer into the scope.

The processing location follows from the classes: which may reach a hosted model, in which region, under which setting. Our post on AI data residency in the UAE covers the cloud side: a UAE region fixes where data rests, and a deployment setting decides where a prompt is processed.

The rules that already apply.

Federal entities buy under Federal Law No. (11) of 2023 Concerning Procurement in the Federal Government and its executive regulation, Cabinet Resolution No. (122) of 2024, in force since 14 May 2025. Article 54 of that resolution repeals Cabinet Resolution No. (4) of 2019, which older guides still cite. Article 1 of Resolution No. (122) defines a procurement strategy as a document prepared by the procurement unit that outlines the procurement methods, expected timelines, evaluation criteria and risks, and its Article 31 has a tender awarded on the evaluation criteria stated in the tender announcement, after technical and financial evaluation.

Dubai government entities buy under Law No. (12) of 2020 Concerning Contracts and Warehouse Management in the Government of Dubai, in force since 1 January 2021; its Article 4 lists the entities and contracts it does not cover. Article 21 lists five procurement methods: public tender, limited tender, negotiation, direct agreement and competitive tender, and Article 23 calls public tender the primary one. Article 5 has entities follow the procurement templates and policies adopted by the Department of Finance, and Article 14 has each entity set technical and financial principles, including that purchases meet its actual needs and the prescribed specifications.

Abu Dhabi entities work to the Abu Dhabi Standards for Government Procurement, which the Department of Government Support launched in April 2021 for use by all government entities. The Government Procurement Office describes the framework as four documents: the Procurement Standards, the Procurement Charter, the Procurement Delegation of Authority and the Procurement Manual. Its published FAQ says a tender's technical and commercial weightings are set in the evaluation methodology, when the sourcing strategy or solicitation documents are created.

AI principles sit beside these. The UAE Charter for the Development and Use of Artificial Intelligence, published in July 2024, lists twelve principles, among them human oversight, data privacy, transparency, and governance and accountability, and names the public sector in its audience. Dubai's AI Ethics Guidelines describe themselves as non-binding; guideline 1.2.7 says operator organisations should understand the AI systems they use well enough to assess their suitability for the use case.

Which of these binds a purchase is for the procurement unit and counsel. For scoping, list them so every bidder answers the same ones.

Success measures a vendor can be held to.

Write measures in the workflow's own units: items per week, time from arrival to decision, the share of outputs a reviewer accepts without edits. Put the measurement method next to each, or vendor and entity will measure different things.

Then acceptance tests: real cases, redacted or synthetic where the data class requires it, each with its expected output, run before acceptance and after every model change. Where the workflow mixes Arabic and English, the set covers both; we test models on real documents in both languages before recommending one.

Then human review: which outputs a named person signs off, and what the system records. The Charter's human-oversight principle and Dubai's traceability guideline (1.3.1) both point at this step; the scope is where it gets a name and a role.

What the vendor gets access to, and what stays with the entity.

Decide access before award, because it shapes every proposal: the accounts a vendor's engineers will hold, in which environments, with which data classes, approved by whom.

Then ownership. The scope can state that code, prompts, evaluation sets and model configurations live in the entity's own repository and cloud tenancy from the first commit, and that the vendor works inside them. The handover is then a change of people with nothing to migrate. Our forward-deployed engineers build in the client's own environment.

Our post on how embedded AI engineers work inside enterprise teams covers the client side: guest accounts, code review and the exit.

Evaluation criteria that follow from the scope.

The evaluation criteria follow from the five settlements and read the same to every bidder:

• Workflow fit. How the proposal changes the named workflow, and how it reports against the baseline.
• Data handling. Which data classes reach which component, where each is processed, and which settings fix that.
• Texts and principles. How the design answers the procurement terms and the AI principles listed in the scope.
• Measures and acceptance. Whether the vendor accepts the measures, the test set and the review steps as written.
• Access and handover. The accounts requested, the environments used, and what the entity holds at contract end.

Do it yourself, or bring us in.

Every step above can be done in-house by a business owner, a data officer and a procurement lead in one room. The output is a short scope: a workflow, a data map, the applicable texts, measures and an access plan.

We come in before the tender: an audit inside the team that maps the workflow, tests candidate models on real documents in Arabic and English, and ends in a written, sequenced roadmap. We aren't lawyers, so the reading of the procurement texts stays with your counsel and procurement unit.

Frequently asked questions.

Keep reading.