AI Consulting for Saudi Government

A government AI strategy earns its value when a service owner can explain which decision will improve, what evidence supports the change, and who remains accountable to the beneficiary. A collection of demonstrations cannot answer those questions. For a Saudi ministry, authority or public service organisation, the consulting brief should connect public outcomes, administrative authority and delivery readiness before it selects a platform.

This guide is for digital transformation leaders, service owners, procurement teams and AI office leads preparing that brief. It focuses on the decisions an advisory engagement should resolve: which services deserve attention, how authority is preserved, what evidence procurement needs, and when a proposal should stop. A useful first outcome may be a decision to redesign a form or clarify a policy before introducing an agent.

Begin with a public-service problem that has an owner

Describe the beneficiary journey in operational language. “Improve government efficiency” is too broad to test. “Reduce repeated requests for missing documents without changing eligibility decisions” identifies a specific problem and a boundary. Name the unit that controls the instructions, the team that receives incomplete submissions, and the authority that can amend the process. Include assisted and offline channels where relevant; an easier digital journey can still move work onto a contact centre.

Interview frontline staff and review a representative sample of completed, delayed and contested cases. Look for avoidable handoffs, inconsistent policy versions and requests that have no accountable destination. The AI use case discovery framework helps compare opportunities, but the government-specific question is whether the proposed intervention improves a service outcome within the entity’s remit.

Record the baseline before a vendor demonstration changes expectations. Useful measures include complete submissions at first attempt, elapsed time under the entity’s control, repeat contacts, unresolved exceptions and accessibility failures. Each measure needs a denominator and an owner. A reduction in handling time is incomplete evidence if appeals or abandoned applications increase.

Separate assistance from the exercise of authority

An agent can assemble information, recommend a next step or take an action. Those are different permissions. A consulting assessment should trace the proposed output to its real effect: does it merely draft an internal summary, communicate an official position, change a record, reject a request, or trigger payment? Labels such as “assistant” do not settle the distinction.

The Digital Government Authority describes national enterprise architecture as a government-sector reference for aligning business and technology. Use that existing architecture discipline to place AI proposals alongside service processes, applications and data ownership. This does not mean every entity has identical approval requirements; its legal and governance teams should identify the instruments that apply.

Service decision brief to complete before selecting an AI solution

Field	Question to resolve	Evidence to retain
Public outcome	Which beneficiary problem will change?	Journey sample and baseline measure
Service authority	Who may change the process or communicate a decision?	Named owner and applicable delegation
Agent boundary	What may be prepared, recommended or executed?	Allowed and prohibited action list
Source authority	Which policy version and records are authoritative?	Document owner and update process
Beneficiary protection	How can an incorrect result be challenged or corrected?	Human escalation and correction path
Release decision	What evidence permits a bounded trial?	Acceptance criteria and accountable sign-off

Use one brief per service decision, rather than a single sheet for an entire ministry. If two departments disagree about the authoritative policy, record the dispute and its resolver. Do not bury it as a technical dependency. An AI project cannot confer administrative authority that the organisation has not granted.

Make readiness a set of evidence requests

Ask whether Arabic and English service information describe the same conditions, whether superseded circulars are marked, and whether staff can identify the source behind an answer. Review permission boundaries between units and contracted operators. Determine how long records and outputs should be retained through the entity’s applicable policies. Avoid collecting live personal records merely to make a discovery workshop more realistic; redacted samples often reveal the decision structure.

An AI readiness assessment should expose work required before a trial, with a person responsible for each item. “Data quality is weak” is not a usable finding. “The service owner must resolve conflicting document lists before the test set is approved” is. Separate blockers from improvements that can run alongside a restricted evaluation.

Government AI governance also requires an applicability review. The advisory team should map relevant entity policies, data obligations and security requirements to the proposed activity, and distinguish binding requirements from recommended practice. The Saudi AI governance guide provides the broader structure; the entity retains responsibility for confirming its own legal and regulatory position.

Write a procurement brief that competing suppliers can answer

Specify the problem, available evidence, authority boundary and acceptance method before asking for a solution architecture. Ask bidders to state assumptions, exclusions, access requirements and what would make their recommendation change. Require the advisory deliverables to be usable by another implementation team: a prioritised portfolio, decision records, source inventory, evaluation plan and sequenced roadmap should have explicit acceptance criteria.

Separate an advisor’s recommendation from any commercial interest in subsequent implementation. A supplier that can both advise and build may offer continuity, but the buyer should retain the right to compare alternatives and procure delivery separately. Request disclosure of vendor relationships and proposed subcontractors. Use AI vendor due diligence to examine evidence rather than accepting broad claims of government compliance or Arabic capability.

Do not make an impressive demonstration the sole selection event. Give shortlisted suppliers the same redacted service scenario, including an ambiguous request and an outdated instruction. Evaluate whether they identify missing authority and evidence, not simply whether they produce a polished answer. Agree how confidential procurement information will be handled during this exercise.

Illustrative choice: incomplete permit applications

Consider a hypothetical public entity receiving permit applications. Its initial proposal is an agent that answers questions and determines whether applicants qualify. Discovery finds that the largest avoidable delay occurs earlier: applicants receive inconsistent instructions about supporting documents. Eligibility decisions involve exceptions that the entity has not documented consistently.

The advisory recommendation separates the work. First, the service owner reconciles the document requirements and assigns version ownership. Next, a bounded assistant can help staff identify missing items from an approved checklist. Eligibility judgement remains with authorised personnel. A separate proposal for decision support requires its own evidence and review; it does not inherit approval from the checklist assistant.

Suppose an illustrative sample contains 200 applications, of which 50 required an additional document request. The baseline is 50 ÷ 200 = 25%. If a later comparable sample records 30 of 200, the difference is 10 percentage points. It is not automatically an AI benefit: changes in application mix, policy and staff support must be considered. The consulting output should define this measurement design before implementation, including how excluded and abandoned cases are counted.

Review the effect on beneficiaries who need another route

Assess the service for people who cannot complete the standard digital journey, need clarification in different wording or submit unusual but legitimate evidence. This is a proposed evaluation practice: choose cases that reveal the consequence of an incorrect assumption, rather than treating a high average completion score as sufficient. Ask the service owner how these cases should be assisted and who can correct an inappropriate response.

Compare the information available to a beneficiary with the information available to a staff reviewer. If internal policy contains an exception that the public guidance omits, the assessment should identify the publication or process issue. An agent should not quietly create two inconsistent accounts of the same service. Where Arabic and English are offered, evaluate substantive meaning and escalation routes, not merely translation fluency.

Include the cost of maintaining these routes in the proposed operating model. Someone must own revised instructions, recurring misunderstandings and corrections following an incident. The advisory acceptance meeting should ask the service owner to explain how a beneficiary reaches a human, how a disputed answer is reviewed and how that finding improves the source material. These questions connect governance to everyday service quality.

Decide what the next funding gate purchases

A roadmap should specify evidence gained at each stage. Funding a trial buys a test of service quality and operational feasibility. Funding production buys a supported service with ownership, monitoring, incident handling and a fallback. The pilot-to-production playbook addresses that transition. A roadmap that lists dates without these decisions leaves the sponsor unable to stop weak proposals.

Bring the decision to the appropriate leadership forum with alternatives: proceed within the defined boundary, repair a prerequisite, run a narrower evaluation, or stop. The guide to presenting AI to your board can help structure the evidence for senior oversight, adapted to the entity’s actual governance arrangements.

Open for AI’s agentic AI consulting can help frame the assessment and roadmap, with deliverables agreed for the engagement. To discuss a government service advisory brief, prepare a redacted journey map, current process ownership and the decision you need to make. These are more useful starting points than a list of preferred models.