Latest / The Clinical Realist / What Hospital Boards Get Wrong About Healthcare AI And Why It Matters to You

Transcript

What Hospital Boards Get Wrong About Healthcare AI And Why It Matters to You · The Clinical Realist

• 0:00
  SPEAKER_00: Health system boards are approving eight-figure AI investments without a governance framework that tells them what they're actually approving.
• 0:08
  SPEAKER_00: They're being given technology presentations.
• 0:10
  SPEAKER_00: What they need are risk assessments.
• 0:12
  SPEAKER_00: The difference between these two things will determine whether AI becomes a competitive advantage for their health system or their next public performance crisis.
• 0:19
  SPEAKER_00: I'm Dr.
• 0:20
  SPEAKER_00: Sarah Matt.
• 0:21
  SPEAKER_00: This is a show for physician executives, technologists, and health system leaders working at the real edge of AI strategy.
• 0:26
  SPEAKER_00: Today's episode is aimed at the board conversation, specifically what hospital boards are getting wrong about healthcare AI investment and what that means for every physician executive and clinical leader who's accountable for the downstream results of those decisions.
• 0:40
  SPEAKER_00: So I want to be direct about the audience.
• 0:42
  SPEAKER_00: If you're a CMO or CMIO, a CNIO, a CHIO, if you're a clinical leader, this episode gives you language for the board conversation you may need to initiate.
• 0:52
  SPEAKER_00: So if you're on a health system board, this episode is a framework for the questions you should already be asking your executive team.
• 0:58
  SPEAKER_00: Either way, the governance gaps at the board level are shaping your clinical reality.
• 1:02
  SPEAKER_00: Let's talk about them.
• 1:04
  SPEAKER_00: So the core argument here is one sentence.
• 1:07
  SPEAKER_00: Health system boards are approving capital expenditures and enterprise software contracts, not governance structures.
• 1:14
  SPEAKER_00: Those are not the same thing, and the difference matters enormously.
• 1:18
  SPEAKER_00: Here's what board AI approval typically looks like.
• 1:21
  SPEAKER_00: The CFO presents a business case with a three-year ROI projection.
• 1:24
  SPEAKER_00: Of course it does.
• 1:25
  SPEAKER_00: IT or a consulting firm presents a technology overview.
• 1:28
  SPEAKER_00: The board votes to approve the budget line.
• 1:31
  SPEAKER_00: That's what get called, that's what gets called board level AI governance in most health systems.
• 1:36
  SPEAKER_00: I mean, that's it.
• 1:37
  SPEAKER_00: What's absence from that process is an actual governance framework that specifies who's clinically accountable for each AI deployment, what the criteria are for pausing or stopping a tool when a safety signal appears, how performance will be monitored after go live, and what the board will be told and when if something goes wrong.
• 1:56
  SPEAKER_00: Boards are approving the spend.
• 1:58
  SPEAKER_00: They're not approving the governance.
• 2:00
  SPEAKER_00: That gap means that when a deployment underperforms or causes harm, there's no board level framework that explains what was supposed to happen.
• 2:07
  SPEAKER_00: So let me put that more concretely.
• 2:10
  SPEAKER_00: I attended a board presentation for a large health systems AI investment proposal.
• 2:14
  SPEAKER_00: It was well-structured, 60-minute session.
• 2:16
  SPEAKER_00: The CFO walked through the ROI model, the chief digital officer presented the vendor's capability demonstrations, and the CIO covered the integration timeline.
• 2:25
  SPEAKER_00: 90 minutes of airtight, airtime, and not one slide covered who would be the named clinical accountability owner for the deployment, what performance data the board would receive and on what cadence, or what organizational authority existed to pause the deployment if a safety signal appeared post-launch.
• 2:44
  SPEAKER_00: The board approved the investment.
• 2:46
  SPEAKER_00: The governance framework that would have answered those questions did not exist.
• 2:51
  SPEAKER_00: And no one on the board even asked for it.
• 2:54
  SPEAKER_00: So this is not a story about a bad board.
• 2:56
  SPEAKER_00: This is a story about a board that was given the information it was given and asked to make a decision with it.
• 3:02
  SPEAKER_00: What the board was not told is that approving the spend without approving the governance structure is in itself a governance failure.
• 3:09
  SPEAKER_00: And most boards don't know what they don't know here.
• 3:12
  SPEAKER_00: Again, it's new for most of them.
• 3:14
  SPEAKER_00: There's also a fiduciary dimension worth naming directly.
• 3:17
  SPEAKER_00: Board members have fiduciary responsibility to the institution.
• 3:21
  SPEAKER_00: In traditional capital investment decisions, fiduciary responsibility means understanding the financial risk.
• 3:28
  SPEAKER_00: In AI decisions, financial risk and clinical risk are inseparable.
• 3:33
  SPEAKER_00: And most board members are not equipped to evaluate clinical risk.
• 3:37
  SPEAKER_00: And that's not a criticism.
• 3:38
  SPEAKER_00: It's a structural gap that requires physician executive input or other clinician executive input at the board level, not just at the operational level.
• 3:46
  SPEAKER_00: So when a board approves an AI investment without a clinical executive's clinical risk assessment on the table, they're making a fiduciary decision with incomplete information.
• 3:56
  SPEAKER_00: Now, the clinical executives who bring that assessment, they're not being difficult.
• 4:00
  SPEAKER_00: They're fulfilling a gap in the board's decision-making capability.
• 4:04
  SPEAKER_00: So three specific misconceptions at the board level are shaping eye investment decisions in ways that create operational and clinical risk downstream.
• 4:11
  SPEAKER_00: And I want to name each one directly.
• 4:13
  SPEAKER_00: The first misconception is our vendor is accountable for performance.
• 4:18
  SPEAKER_00: So boards hear vendor performance guarantees at the business case stage.
• 4:21
  SPEAKER_00: I mean, this is SLAs on uptime, right?
• 4:24
  SPEAKER_00: What they don't hear is what those guarantees actually cover and what they explicitly do not cover.
• 4:30
  SPEAKER_00: Now, most AI vendor performance guarantees cover system uptime, right?
• 4:34
  SPEAKER_00: Technical specifications, they don't cover clinical outcomes.
• 4:38
  SPEAKER_00: So when a deployment underperforms clinically, the vendor is within the contract terms and the health system is holding the clinical accountability.
• 4:46
  SPEAKER_00: And that's not a vendor betrayal.
• 4:48
  SPEAKER_00: It's a construct in the contract.
• 4:50
  SPEAKER_00: And most boards are never seen, have never seen it explained to them very plainly.
• 4:54
  SPEAKER_00: So, you know, the board is approving the investment based on a vendor promise.
• 4:58
  SPEAKER_00: Now, the vendor delivered on what the contract actually said they would deliver.
• 5:01
  SPEAKER_00: And the gap between what the board understood they were buying and what the contract actually specified is where the institutional risk lives.
• 5:08
  SPEAKER_00: So a clinical executive who walks into a board presentation and says, I'd like to review the vendor's contract language on clinical accountability before this vote is not slowing down a good decision.
• 5:18
  SPEAKER_00: They're protecting the board from making an uninformed one.
• 5:22
  SPEAKER_00: Now, the second misconception is that physician buy-in is a change management problem we can handle after launch.
• 5:31
  SPEAKER_00: Insert whatever other clinicians you want there.
• 5:34
  SPEAKER_00: The nurse buy-in is a change management, the PT, the physics, whatever it is.
• 5:38
  SPEAKER_00: Boards see clinical staff resistance to AI tools as a communication and trading problem.
• 5:44
  SPEAKER_00: Something to solve with education campaigns, shiny stickers, and workflow champions post-launch.
• 5:51
  SPEAKER_00: What it actually is, in most cases, is a governance design signal.
• 5:55
  SPEAKER_00: So clinical staff resist tools that were selected without their leadership, having genuine authority in the process.
• 6:01
  SPEAKER_00: So that resistance is not irrational.
• 6:03
  SPEAKER_00: It's actually a rational response to being accountable for tools you did not have authority over.
• 6:08
  SPEAKER_00: So when a clinician knows, a physician, for example, that the CMO's role at a procurement cycle was to validate a decision already made rather than to shape the decision, they trust the resulting tool less.
• 6:21
  SPEAKER_00: Not because they're anti-technology, but because they know that clinical judgment was not structurally in the room when the tool was selected.
• 6:29
  SPEAKER_00: Now, boards tend to treat physician, nurse, pharmacist, et cetera, resistance as a change management problem.
• 6:35
  SPEAKER_00: And this will keep funding change management.
• 6:39
  SPEAKER_00: And this will keep funding change management campaigns.
• 6:45
  SPEAKER_00: Boards that treat physician resistance as a change management problem will keep funding change management campaigns and keep getting the same result.
• 6:53
  SPEAKER_00: The structural fix is not better communication about the tool.
• 6:57
  SPEAKER_00: The structural fix is genuine clinical authority in the selection process.
• 7:02
  SPEAKER_00: Third misconception is our AI governance committee handles governance.
• 7:06
  SPEAKER_00: First of all, that's hilarious.
• 7:08
  SPEAKER_00: Most health systems have stood up maybe an AI governance committee in response to institutional pressure to have a governance structure.
• 7:14
  SPEAKER_00: That's a reasonable response to a real pressure.
• 7:16
  SPEAKER_00: But committees are not governance.
• 7:19
  SPEAKER_00: A committee that meets quarterly to review vendor performance reports, that's not the same thing as named clinical accountability.
• 7:25
  SPEAKER_00: Define stopping criteria and explicit authority to halt a deployment, for example.
• 7:31
  SPEAKER_00: A committee can recommend, a committee can advise, a committee cannot, by its nature, act with the speed that a clinical safety signal sometimes requires.
• 7:41
  SPEAKER_00: So boards that equate committee existence with governance adequacy are significantly underestimating their institutional risk.
• 7:48
  SPEAKER_00: So the question to ask is not, do we have a committee?
• 7:51
  SPEAKER_00: Heaven help me.
• 7:52
  SPEAKER_00: The question is who has the named authority to stop a deployment if a safety signal appears?
• 7:57
  SPEAKER_00: And can they act without calling a committee meeting first?
• 8:03
  SPEAKER_00: So what boards should actually be asking their executive teams?
• 8:08
  SPEAKER_00: There's six questions.
• 8:09
  SPEAKER_00: So if your board asks these before every AI investment approval, the quality of AI governance in your health system would improve immediately.
• 8:16
  SPEAKER_00: So I'm going to give them to you in the order I think they matter most.
• 8:19
  SPEAKER_00: Question one: Who's the name clinical executive accountable for this deployment and what is the scope of their authority?
• 8:26
  SPEAKER_00: Not the CMO's office, a specific person by name with a defined authority scope.
• 8:31
  SPEAKER_00: If the answer is a committee, ask again.
• 8:34
  SPEAKER_00: Committees advise.
• 8:36
  SPEAKER_00: A deployment needs an accountable individual whose professional standing is connected to the outcomes because that connection is what ensures the monitoring and governance actually get done.
• 8:46
  SPEAKER_00: Question two.
• 8:47
  SPEAKER_00: What are the criteria for stopping or pausing this tool after deployment?
• 8:52
  SPEAKER_00: And who has the organizational authority to act on those criteria?
• 8:56
  SPEAKER_00: This needs to be in the actual deployment plan before the board approves the budget.
• 9:00
  SPEAKER_00: Not defined after a problem occurs.
• 9:02
  SPEAKER_00: The criteria for stopping deployment are always easier to find before you're in the middle of a vendor relationship, a capital investment, and organizational momentum behind expansion.
• 9:12
  SPEAKER_00: Question three.
• 9:14
  SPEAKER_00: What is the validated scope of this tool and does our planned deployment match that scope?
• 9:19
  SPEAKER_00: If the vendor validated the tool on a different patient population or in a different workflow context, the board is approving deployment outside the evidence base.
• 9:26
  SPEAKER_00: And that's a risk the board should understand before approving the spend, not discover after the deployment is underperforming.
• 9:35
  SPEAKER_00: Question four.
• 9:36
  SPEAKER_00: What will you tell us and when if this deployment is not performing as projected?
• 9:41
  SPEAKER_00: Define the performance reporting cadence and the threshold that triggers a board-level conversation before it becomes a crisis conversation.
• 9:49
  SPEAKER_00: Boards often receive AI performance updates only when things are going well.
• 9:52
  SPEAKER_00: The governance question is what triggers the conversation when they are not.
• 9:58
  SPEAKER_00: Question five.
• 9:59
  SPEAKER_00: So what does our vendor contract say about our clinical accountability if the tool produces an output that contributes to a patient harm event?
• 10:06
  SPEAKER_00: Your legal team should be able to answer this in plain language.
• 10:09
  SPEAKER_00: If they can't, that's information about the adequacy of the contract review process.
• 10:14
  SPEAKER_00: Boards that approve AI investments without a plain language summary of their clinical liability exposure in the event of adverse outcomes are making incomplete fiduciary decisions.
• 10:25
  SPEAKER_00: Question six, does the clinical or physician executive who will be accountable for this deployment have genuine authority to pause or stop it if they identify a safety signal?
• 10:35
  SPEAKER_00: Not ceremonial oversight, actual organizational authority with the ability to act without needing executive consensus.
• 10:42
  SPEAKER_00: If the clinical executive who is accountable for outcomes must seek approval from a committee before pausing a tool with an active safety signal, you have accountability without authority.
• 10:51
  SPEAKER_00: And that's the most dangerous governance structure in clinical AI, and it's the most common one too.
• 10:56
  SPEAKER_00: So getting governance right at the board level requires clinical executives to initiate the conversation, not wait to be asked.
• 11:03
  SPEAKER_00: And most are waiting.
• 11:05
  SPEAKER_00: Why?
• 11:06
  SPEAKER_00: Because physician, nurse, pharmacy executives have been conditioned over careers to bring operational execution to boards, not strategic risk framing.
• 11:15
  SPEAKER_00: Telling a board that their AI governance structure is adequate feels like overstepping.
• 11:20
  SPEAKER_00: I'm telling you it's not.
• 11:21
  SPEAKER_00: It's the job.
• 11:22
  SPEAKER_00: The framing that works is fiduciary risk language, not clinical objection language.
• 11:28
  SPEAKER_00: So when you frame physician governance concerns as fiduciary risk, institutional liability exposure, regulatory risk, the accountability gap that exists between approved spend and name clinical oversight.
• 11:40
  SPEAKER_00: Boards understand it because that's their vocabulary.
• 11:43
  SPEAKER_00: I am concerned about clinical outcomes.
• 11:46
  SPEAKER_00: That reads as physician caution.
• 11:48
  SPEAKER_00: I need the board to understand that we are approving a capital expenditure without an approved governance structure, and that gap creates institutional liability.
• 11:56
  SPEAKER_00: That reads as board level input.
• 11:58
  SPEAKER_00: Same concern, very different reception.
• 12:01
  SPEAKER_00: So the specific ask is three things.
• 12:04
  SPEAKER_00: First, a board-approved governance framework that specifies named accountability before each deployment.
• 12:09
  SPEAKER_00: Not after, before.
• 12:12
  SPEAKER_00: Second, explicit organizational authority for the named clinical executive, physician, nurse, pharmacist, you name it, to pause a deployment when a safety signal actually appears without needing to call a committee meeting.
• 12:25
  SPEAKER_00: Third, a performance reporting cadence that reaches the board, not just the executive team.
• 12:31
  SPEAKER_00: And here's what I have seen happen when physician executives bring this framing to the board.
• 12:36
  SPEAKER_00: I worked with a CMIO at a regional system who had watched two AI deployments underperform significantly in an 18-month time frame.
• 12:46
  SPEAKER_00: Neither deployment had a named physician accountability owner at the outset.
• 12:50
  SPEAKER_00: Neither had documented stopping criteria.
• 12:52
  SPEAKER_00: Both had governance committees that met quarterly.
• 12:55
  SPEAKER_00: So when a third major AI investment came before the board, she asked for 15 minutes before the vote.
• 13:01
  SPEAKER_00: She presented one page.
• 13:02
  SPEAKER_00: On the left side, what the proposed deployment was approving.
• 13:06
  SPEAKER_00: On the right side, the governance structure, the deployment required to be viable.
• 13:10
  SPEAKER_00: She did not argue against the investment.
• 13:12
  SPEAKER_00: She argued for the governance framework that would make the investment succeed.
• 13:16
  SPEAKER_00: The board voted to approve both the investment and the governance framework, which they had never actually been asked to approve before.
• 13:22
  SPEAKER_00: And that framework took four weeks to build and three years to become normal operating procedure.
• 13:26
  SPEAKER_00: The CMIO told me that the board presentation was literally the hardest professional conversation she'd ever initiated in her entire career.
• 13:33
  SPEAKER_00: It was also perhaps the most consequential.
• 13:36
  SPEAKER_00: So most boards, when they understand the liability exposure that the current governance gap creates, they're going to want to close it.
• 13:42
  SPEAKER_00: And they're not indifferent to the problem.
• 13:44
  SPEAKER_00: They have not been given the programmatic problem in language they can actually act on.
• 13:51
  SPEAKER_00: Clinical executives who bring it in that language, they're going to get the governance structure they need.
• 13:57
  SPEAKER_00: So health system boards are not the obstacle to good AI governance.
• 14:02
  SPEAKER_00: The absence of a clear governance framework, one that assigns accountability before deployment, not after, that's the obstacle.
• 14:09
  SPEAKER_00: And fixing that requires doctors, nurses, pharmacists, all those clinical executives who are willing to bring the conversation to the board table in the duchier risk language, not clinical objection language.
• 14:22
  SPEAKER_00: The investment is already being approved.
• 14:25
  SPEAKER_00: The question is whether the governance structure that makes that investment actually viable is going to be approved alongside it.
• 14:31
  SPEAKER_00: Right now in most health systems, that's not the case.
• 14:34
  SPEAKER_00: So that's a solvable problem, but it does not solve itself.
• 14:37
  SPEAKER_00: So I'm Dr.
• 14:38
  SPEAKER_00: Sarah Matt, new episodes every Wednesday.
• 14:40
  SPEAKER_00: If you found this useful, share it with a clinical executive who has a board presentation coming up or send it to the board member who just asked why your health systems AI deployment isn't performing as projected.
• 14:50
  SPEAKER_00: Start the conversation.