Approve a project and ensure it is executed well are two separate responsibilities, although in most boards they are treated as if they were the same. The first responsibility falls to the management team. The second responsibility falls to the board — and in data or artificial intelligence projects, it is the one that is most often mismanaged, not because of a lack of interest, but because there is no simple framework for knowing what to ask.
Advise and monitor: two roles in front of the same project
A management team that evaluates a data or AI project naturally focuses on execution: how well the business case is structured, how realistic the timeframe is, and how qualified the vendor or internal team that will build it is. It is a role that is close to the technical details and the person who will ultimately make the decision to move forward.
The board of directors operates at a different level. It rarely looks at the technical details of the project—it sees a presentation, an executive summary, and a recommendation already validated by the team that proposes it. Its role is not to redo that analysis. It is to verify that the analysis was done with the rigor required by the capital committed, before approving it.
Both roles use the same vocabulary — ROI, predictive model, business case, risk — which makes it easy to confuse them. But one builds the answer to the business question, and the other has to verify that the question was asked correctly before an answer can be built.
What a management team asks—and what a board should ask
When faced with a data or AI project, a management team generally evaluates three things: whether the project will work, how much it costs, and how long it takes. These are legitimate and necessary questions—they determine whether the project progresses operationally.
One piece of advice should be evaluating something different: how we will know if the project worked, who is responsible if it doesn’t work, and how reliable the data on which the recommendation that is about to be approved is based is. This third question is the one that is asked the least often, and it is the one that most determines whether the project has a solid foundation — because a well-built model based on data without governance produces a recommendation that appears rigorous when it really isn’t.
¿Why most advice doesn't ask the third question?
The most common reason is a lack of interest on the part of the advisors. It’s a lack of vocabulary. Most members of a board don’t have a technical background, and in most organizations there is no simple framework that allows them to evaluate a data or AI proposal without needing to understand the model line by line.
Given that lack of vocabulary, the most common reaction to a technical proposal is divided into two: approving it based on the confidence generated by the person presenting it, or rejecting it out of instinct without being able to technically justify the reason. Neither of these constitutes effective corporate governance—the first amounts to delegating oversight entirely, and the second generates friction without a foundation that the management team can address.
There is a third option, less common in practice but more useful: establishing a reduced set of non-technical questions that allow assessing whether the background work was done rigorously, without requiring experience in data science.
The three questions a board should ask before approving
1. What specific business decision will change with this project, and who is responsible for making that change? Not what technology will be implemented, but what decision, made by whom, will be different as a result of the project.
2. How will we know, with evidence, that the decision actually changed and that the change was worth it? If the team presenting cannot name the metric that will confirm it, the project is not yet ready to be approved — it is an intention with an assigned budget.
3. How reliable are the data on which this recommendation was based? A consultant does not need to audit the data line by line. He needs to confirm that someone already did it, and what they found when they did it.
These three questions do not replace the technical work of the team that proposes the project. They complement it at the level that is appropriate for a board: not to build the answer, but to verify that the question was formulated rigorously before committing capital.
¿How to institutionalize these questions without technifying the entire board?
There is no need to redesign the corporate governance structure to implement this framework. A simple agreement is enough: no data or AI project is presented to the board without the proposing team being able to answer the three questions in writing before the meeting.
That change changes the dynamics of the meeting. Instead of the board evaluating a presentation in real time and deciding based on how convincing it was, it arrives with the already documented responses, and the discussion focuses on whether those responses are sufficient—the level of oversight that a board is entitled to, without needing to go into technical detail.
It is also not necessary for all members of the board to have a technical background. It is sufficient for one person—an independent advisor, the chairman of the audit committee, or someone with explicit authority—to be responsible for asking these three questions in every relevant proposal, until they become a regular part of how the board evaluates any data or AI project.
¿What happens when these questions are not asked??
The absence of these three questions does not always translate into a project that fails visibly. More often, it translates into a project that is approved, executed, and produces ambiguous results—neither an obvious failure that forces its discontinuation, nor a clear success that justifies its approval without the questions in the first place.
That ambiguous outcome is, in practice, the most costly scenario for a board. A project that fails clearly generates a clear lesson and a course correction. A project that produces ambiguous results consumes budget, management team time, and political capital within the organization, without anyone being able to determine with certainty whether it was worth it — because from the beginning it was never defined how success was going to be measured or how reliable the source data were.
Over time, that pattern erodes something more difficult to recover than the budget of a single project: the board’s confidence in its own ability to evaluate data or AI proposals. When that happens, the most common reaction is not to demand better questions—it is to demand fewer projects, which slows the adoption of technology that could generate value, simply because the oversight mechanism was never clearly established.
Frequently asked questions
- Do these questions apply only to large AI projects? No. They apply to any data or AI project that requires board approval, regardless of the size of the budget — the third question, about the reliability of the data, is often just as relevant in a small project as in one of several million dollars.
- Who should be responsible for asking these questions within the council? It doesn't have to be a technical profile. It can be any advisor who is explicitly assigned that responsibility — the important thing is that there is someone with the mandate to do so in every proposal, not that it is taken for granted that someone else has already asked for it.
- What happens if the team proposing the project cannot answer the three questions? That doesn't mean the project should be rejected. It means that it's not yet ready to be approved with the full capital — it's a signal to ask that those answers be resolved before the next session, or to approve a more limited initial phase while it's being built.
- Do these questions delay the approval of a project? They shouldn’t be asked for in writing before the session and not during it. The goal is not to add an additional filter at the end of the process, but for the team proposing the project to arrive with those already-resolved answers—which, in most cases, speeds up the conversation rather than prolonging it.
Next step
At MaIA by Scanda, we work with board members and management committees that need a simple framework to evaluate data or AI projects without relying on a technical profile within the board. If your organization is about to present or approve a project of this type, let's talk.
Español 
English