Your Board Will Ask About AI. The Question Will Come Too Late.

The Board will eventually want to know what AI is operating within the organization. This is all but inevitable. What will most likely be the case by the time they ask the question is the answers will be beyond our grasp. The Board typically comes to the discussion last, after the system has been operational for a significant amount of time, and the timing of this discussion determines the extent of the ability of the Board to take meaningful action.

The question does not typically come during procurement, when a system is being selected and the associated terms can be negotiated. It is not asked during the pilot of the system, when the system is small and easy to modify. By the time they ask the question, the system is operational, and the Board has very little ability to meaningfully influence the situation.

The pattern is very clear

The sequence is almost universally the same across organizations. A system is introduced to generate process efficiencies and otherwise lessen a burden, and the system is effective enough that no one questions it. The pilot that was (or should have been) contained to a single department expands to additional departments, and clinicians modify their behavior until the system is just part of how work is accomplished. None of this is brought to the Board, for no one suspects any of this is of Board concern.

Then something bubbles to the surface. A complaint from a patient. Maybe an external question. It could be a piece of press or notification that an authority is examining the practice. Only then does the AI make it to the board agenda. The board treats this as something outside of their control, causing them the most frustration at the most inconvenient time. The board meets the system for the very first time at this critical juncture when it is most entrenched and least malleable.

The question that lands the hardest

When it finally surfaces, the question that does the most damage is also the simplest. Why are we hearing about this now? It is a fair question that has no comfortable answer. The honest answer is the system was never considered to be something the board needed to see until it became a risk.

By the time that question is asked, the tool has already formed real care and real documentation for months. Decisions were made on its output. Patients were affected by it. The board is being asked to give their opinion on something that has already done the majority of its work, and the time lag between when it could have helped the most and when it was finally presented with the issue is exactly the time that the most difficult questions arise.

Why Boards are Last to Know

This is usually not an issue involving any one person. It is simply a function of the way information disseminates. Usually in a hierarchy, risk is dealt with at the level it is encountered, and the unaddressed risk is passed along in a more palatable form. By the time an issue is brought to the board, it is the result of several escalations and assurances, each well-founded, and what remains is a calm and retrospective picture. The board is briefed on what has already occurred and has already been managed, rather than what is in the process of being formed.

This system is generally acceptable for known and slow-moving risks. It falls apart for something like AI, which moves and changes while it runs and, while crossing the clinical and legal boundaries, fails to be captured by the normal reporting frameworks. The most concerning aspect about AI is the element which, paradoxically, is the most difficult to capture by the current reporting framework.

An Embedded System Cannot be Unraveled

Part of the reason being late is so costly is that these systems cannot be easily removed once they have been embedded. By the time the board learns of the introduction of a new system, the surrounding consent language and the accompanying clinician practices have been established. Removing the system means redoing all of that, which is disruptive and poses its own risk.

The board, typically arriving last, is not faced with a decision between a fix and the status quo. Rather, it is faced with making a decision between a series of bad choices. This is because easy choices were made unavailable, while the system was quietly reaching a load-bearing level of stress. Late engagement limits the influence of the board, and it transforms the role from one of setting the direction, to a containment role for which the scope is much smaller, and thus far less valuable.

What the board should see early on

The solution is to put AI to the board while it is still in the early decision making stage. This ensures that the board sees AI at the procurement stage, and has the ability to influence the contract terms and conditions, as well as the use case. The board should also be able to influence AI while it is still at the pilot stage, which is small enough to meaningfully influence or stop without a major cost. The intent is to engage up to the point of influence, and not thereafter.

This has a few requirements that are underlying conditions. Someone has to own this, along with a role that includes deciding which issues come to the board so that the filtering does not drown out important signals. Also, there have to be known, and agreed upon, limits for what an organization considers as a board-level question pertaining to AI, in order that the decision to escalate is not made by the chance observer. In this context, AI oversight is integrated into what the board considers as part of the enterprise infrastructure, rather than an oversight that becomes a surprise the board encounters as they deal with the fallout.

The board will ask, regardless

There is no scenario that passes without the board asking about AI. The implications of the technology coupled with the significant exposure creates a situation that makes this a certainty. What an organization does control is when the discussion takes place.

A board that sees AI early on is able to, as boards do, weigh the risks and the benefits and give direction while it is still possible to do so. The same board meeting AI is eventually brought on, after an incident, can ask very little other than the reason it was not informed earlier. The priority of work should be to ensure the answer to that question is one the organization is prepared to give.

Christopher Hutchins Founder and CEO, Hutchins Data Strategy Consultants