Present a decision directors can take
How to Present AI to Your Board
Prepare an evidence-led board proposal with a clear resolution, alternatives, realistic economics and a bilingual decision boundary.
An effective AI board presentation makes a decision easier to take. It tells directors what management recommends, why the choice matters, what evidence supports it and what remains uncertain. The presentation should allow the board to approve, modify, defer or reject the request without needing to reconstruct the proposal from a product demonstration.
This guide provides a reusable structure for a Saudi executive sponsor, chief financial officer or transformation lead. Adapt the pack to the company's governance, language and information requirements. The board governance guide explains how to identify the appropriate approval route; this page focuses on communicating a particular decision.
Write the requested decision before building slides
Use one sentence that names the action, scope, funding boundary, owner and next review. For example: management requests approval to investigate a defined renewal-preparation workflow within an agreed budget, with no authority to send customer communications, and will return with comparative evidence before requesting operational deployment. Fill in the actual limits rather than leaving the request as support our AI ambition.
State whether the board is being asked to note information, endorse a direction, approve funding or accept a decision reserved to it. Those are different actions. Ask the company secretary to confirm the resolution format and how conditions will be recorded. A presentation can be persuasive and still fail operationally if nobody can tell what was approved.
An eight-part board pack
| Section | Main message | Evidence or exhibit |
|---|---|---|
| 1. Decision requested | What management recommends and what authority is sought | Proposed resolution, limits and next review |
| 2. Business problem | Which outcome is underperforming and who is affected | Baseline with population, period and source |
| 3. Options | Why this approach is preferable to credible alternatives | Comparable choices, including simpler process or existing-tool options |
| 4. Changed workflow | What the AI and people will actually do | Current and proposed steps, action boundaries and review points |
| 5. Evidence and uncertainty | What is known, tested, assumed and still unresolved | Evaluation summary, excluded cases and next tests |
| 6. Economics | What it costs and how value would be realised | Cash, capacity, assumptions, downside and funding stages |
| 7. Ownership and risk | Who is accountable and how exposure is controlled | Named owners, material risks, conditions and escalation |
| 8. Next commitment | What happens after the decision and when to stop | Milestones, acceptance evidence and return-to-board trigger |
Eight sections are a suggested structure, not a rule requiring eight slides. A complex proposal may need a short main pack with detailed appendices; a narrow decision may need fewer pages. Keep the recommendation, material uncertainty and consequences in the main narrative. Do not hide the evidence most likely to change the decision behind technical background.
Explain the workflow using actions and consequences
Show the current process and the proposed change in plain language. Identify what information enters, what the system prepares or decides, what a person reviews and what changes outside the system. A director should be able to distinguish a draft from an approved action after reading the exhibit.
A brief demonstration can help make the workflow concrete, but explain whether it uses representative data, a controlled example or a prototype. Include an exception or failure path relevant to the proposal. A successful performance on one prepared case should not be presented as proof that every intended case has been evaluated. Use the director briefing if the audience first needs a shared vocabulary.
A worked example that invites challenge
Consider a hypothetical service business preparing 500 contract renewals per month. Current information assembly takes 18 minutes per renewal. Management estimates that an assistive workflow could reduce assembly to 10 minutes, but adds four minutes of review per case. The proposed system prepares material only; it cannot change contract terms or send a renewal offer.
Current effort is 500 × 18 ÷ 60 = 150 hours monthly. Proposed assembly and review together require 500 × (10 + 4) ÷ 60 = 116.7 hours monthly, rounded to one decimal place. Calculate net capacity directly from the unrounded assumptions: 500 × (18 − 10 − 4) ÷ 60 = 33⅓ hours monthly, reported as 33.3 hours. These are illustrative assumptions, not measured results or a customer case.
Now suppose management requests a SAR 90,000 pilot. Directors should ask whether that commitment is proportionate to the plausible value, what simpler option was considered and which assumptions the pilot would resolve. The 33.3 hours are capacity, not automatically payroll savings. If management cannot explain how that capacity will improve outcomes, the board may choose a smaller investigation, revise the proposition or decline the request.
This is a stronger presentation because the review burden is visible and the board can challenge the economics. The purpose of the pack is a sound decision, including an unwelcome one. The business case framework and cost model provide more detailed support for the financial exhibit.
Label evidence so directors know how much weight to give it
Use clear labels: observed baseline, tested result, supplier representation, management estimate and unresolved assumption. State the sample, period and scope where relevant. If the team has only a small test, explain what decision it supports and what it cannot establish. Confidence should follow the evidence rather than the visual polish of the slides.
Separate the evidence for the technical task from the evidence for adoption and business value. A good extraction result does not establish that employees will use the workflow or that a downstream bottleneck will change. Present the remaining tests as planned decisions with owners, not as a vague list of risks that will be managed later.
Prepare answers to the questions that could change the recommendation
- Why now? Explain the cost of waiting and the evidence that makes the timing appropriate.
- Why this option? Compare the strongest credible alternative, not an intentionally weak one.
- What could make the investment fail? Name the assumptions with the largest effect on value or feasibility.
- Who owns the result? Identify business, technology, risk and operating responsibilities.
- What happens if it is wrong? Explain detection, human review, suspension and the consequences of actions already taken.
- What would cause us to stop? State the evidence or conditions that end or narrow the commitment.
Allocate answers to the people who can substantiate them. The technology lead should not invent a finance view, and the sponsor should not improvise a legal conclusion. Where an answer is not yet available, say what determination is needed and whether the requested decision can responsibly be taken before it is resolved.
Make a bilingual pack consistent at the decision level
Where directors use Arabic and English materials, align the proposed resolution, numbers, conditions and technical terms across both versions. Keep one controlled source of the financial assumptions and a glossary for ambiguous terms. Verify that prepare, recommend, approve and execute have not been translated into a single vague action.
Check what the official meeting record will preserve and who validates it. Language consistency matters most at the authority boundary: a proposal that sounds like assisted drafting in one version must not imply independent execution in the other. This is an editorial and governance preparation task to agree with the company secretary, not an assumption about every Saudi board's required language format.
Close with the resolution and the next evidence review
Return to the requested decision, show any changes made during discussion and identify conditions and owners. Record deferred questions and the route for resolving them. The next report should compare evidence with the approved conditions rather than introduce a different success story. The steering committee operating model can organise management follow-through.
If the board needs stronger challenge of a claim before deciding, use the assurance guide to frame the review. Agentic AI consulting can help organise the proposal and its evidence. Discuss an AI board decision pack with the requested resolution, current assumptions and the questions directors are most likely to challenge.
FAQ
Your questions, answered.
What should be on the first slide of an AI board presentation?
State the decision management requests, the scope and authority involved, the funding boundary and the next review. Explain why the decision is needed now. Directors should know whether they are being asked to note information, endorse direction or approve a commitment before they see technical details or a demonstration.
How many slides should an AI board proposal contain?
Use enough material to support the decision and keep detailed evidence in accessible appendices. The eight-part structure in this guide is a suggested narrative, not a required slide count. Material uncertainty, alternatives, economics and conditions belong in the main pack. A shorter presentation is not better if it conceals information that could change the decision.
Should we show a live AI demo to the board?
A demonstration can make the proposed workflow understandable if its limitations are clear. State whether it uses a prototype, prepared example or representative data, and show a relevant exception. Do not let a successful demo replace evaluation evidence, operating ownership or the business case. The board should understand what the demo establishes and what it does not.
How do we present AI time savings without overstating the financial return?
Show the baseline, affected volume, proposed effort and any new review or support work. Calculate net capacity before assigning financial value. Explain whether that capacity changes cash spending, improves throughput or remains unallocated. Use clearly labelled assumptions and avoid presenting every hour released as an automatic payroll saving.
What must be consistent in Arabic and English AI board materials?
Align the requested resolution, figures, conditions, authority boundaries and key terms. In particular, preserve the distinction between preparing, recommending, approving and executing an action. Use controlled financial assumptions and confirm the meeting-record process with the company secretary. Translation should not unintentionally expand the scope or authority being requested.
How should we respond when directors reject the proposed AI pilot?
Record the reasons and distinguish a request for better evidence from rejection of the underlying proposition. Reassess the scope, alternatives and economics before returning. Do not relabel the same request or change success criteria without explanation. A board presentation succeeds when it supports an informed decision, including a decision to defer or stop.