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

Related on The Signal Room: From AI Strategy to Execution with Brian Sutherland. Related from HDSC: Healthcare AI Consulting: What It Takes to Make AI Work in Healthcare.A member of a health system board reads about a hospital being sued due to an erroneous AI-assisted diagnosis. At the following board meeting, she directly asks the CEO, how many AI tools are we using in our clinics, and who is overseeing these tools? The CEO asks the chief information officer, who asks the chief medical officer. None of them know what the answer is. Then another question is asked, what is our exposure if one of these tools causes a harm event? Silence. The question is on the table, and the institution is not prepared.

Media reports, examples from peer institutions, and an embedded awareness of AI tools in the clinical operational environment have opened the floodgates. Hospital and health system board members are asking about AI. Almost all of these boards confront a timing problem. AI tools are already in operational use and actively transforming the clinical environment before the board has had the opportunity to ask these questions. Centralized reporting, documentation, and operational oversight needed to provide the board answers to this critical line of inquiry do not yet exist.

According to research from Deloitte, which analyzes the queries posed in board meetings at health systems regarding AI, the predominant questions concentrate on three areas: costs and ROI, competitive AI tool usage, and high-level descriptions of AI tool usage. These questions are valid, and while they show some board engagement, they should be considered the very basics rather than the greater complexity the current AI landscape calls for.

AHA research shows that the questions boards are most frequently not asking are the questions related to clinical AI and patient safety, legal and liability risk regarding AI-assisted clinical decisions, workforce effects of AI on the clinical staff, post-deployment monitoring, performance management, and AI risk management structures. A wide gap exists between what boards are asking and what they should be asking, and it is the obligation of executive leadership to close it.

Health system boards operate on a predetermined timeline, on a monthly or quarterly basis. In contrast, the implementation of AI tools is done on a continuous basis by the operational and clinical leadership in response to vendor offers, clinical requirements, or competitive pressure. Therefore, by the time a board member inquires about AI, the organization under discussion may have already implemented numerous tools without any information being provided to the board regarding any of them.

Health Affairs studied the length of time that health systems took to implement AI tools and the associated time frames to update the board. In the majority of the cases studied, board members were only apprised of the specifics of AI implementation after the fact, and only when the circumstances warranted an inquiry from the board. These circumstances included scrutiny of AI implementations at competitor institutions, lawsuits pertaining to AI-supported care, and governmental studies probing the use of AI within health systems.

McKinsey has documented the gap in preparedness for board-level engagement regarding the use of AI in health care, where boards of health systems are demonstrably less prepared to respond to questions involving the use of AI than their counterparts in financial services, technology, and manufacturing sectors. According to their review, the rapid adoption of AI in health care and the complexity of proprietary AI technologies, combined with the absence of a uniform reporting structure, are key contributory factors.

Given the importance of board assessments and in anticipation of questions from the board, the briefing should incorporate several key elements.

First, an enterprise-wide inventory of all AI systems and tools used in the clinical and operational domains, with the vendor name, clinical use case, deployment date, and designated performance owner included for each.

Second, an overview of potential risk to patient safety, legal, and operational risks of each AI tool and the risk mitigation strategies in place.

Third, reporting on performance metrics, such as accuracy, consistency, clinical appropriateness, and general performance. It will be a report on the functional performance of deployed AI tools compared to the defined parameters. Also include any performance drift that has been identified since the last board briefing and what actions have been taken.

Fourth, workforce impact explains how AI tools have affected the clinical workflow, clinician workload, and employee satisfaction. Joint Commission reports that health systems should assess the impact of clinical AI on the workforce, and this should be highlighted at the board level.

Fifth, a forward-looking description of remaining AI risks, as well as the planned controls and resource requirements needed to manage the risks of the growing portfolio of AI technologies. KLAS has studied how health systems report AI strategy to their boards. According to KLAS, the most effective briefings include a forward view so that board members can better understand what is coming and how they can provide the needed resources.

It is a failure of executive leadership to wait for the board to ask. Board members are not familiar enough with the operational detail of clinical AI tools to formulate the right questions without the necessary briefing. That responsibility falls to the executive team.

AHA research suggests that where health systems implement AI technologies, executive management must customize their AI reporting to boards according to their typical executive ownership. Leadership decides what information should go to the board, how often, and in what format the board can fulfill its fiduciary and oversight duties.

Health system leaders must begin creating board briefs, even before the board requests them. Set the reporting cadence, catalog the enterprise AI, define the metrics, compile the risk summary, and present it to the board as a routine agenda item rather than as a result of a specific incident.

Institutions that take this proactive approach will have boards that are informed, engaged, and ready to support the complexities of clinical AI adoption. Those that wait will have boards that are surprised, reactive, and far less willing to provide the backing that an AI strategy requires.

Deloitte has researched board meeting questions about AI at health systems. AHA has documented the gaps in board inquiries about clinical AI. Health Affairs has studied the gap between board awareness and AI implementation. McKinsey has analyzed the readiness gap for AI in healthcare at boards. Joint Commission has recommended monitoring clinical AI impact on the workforce. KLAS has focused on the reporting of AI strategies to board leadership in health systems. This issue connects with institutional leadership themes in Editions AH, AK, and AB of this newsletter.

Christopher Hutchins
Founder & CEO, Hutchins Data Strategy Consultants