The Q2 Governance Problem: When AI Goes Live Without a Safety Net

Most health systems that deployed clinical AI in Q1 are now six to eight weeks post-launch and starting to see the governance gap. Dr. Sarah Matt explains what the gap looks like in production, why retrofitting governance is more expensive than building it before go-live, and the four-step rapid governance build for organizations that are already live and need to catch up.

—

Resources & Links:

📖 Get the Book: "The Borderless Healthcare Revolution" is available now on Amazon and major retailers.

💼 Work with Dr. Matt:
Looking for a keynote speaker or strategic advisor?
Visit: drsarahmatt.com

🔗 Connect on Social:
LinkedIn: https://www.linkedin.com/in/sarahmattmd/
YouTube: https://www.youtube.com/@DrSarahMatt-ClinicalRealist

📧 Subscribe to The Briefing: drsarahmatt.com/newsletter-signup
 
—
Disclaimer:
The views expressed on this podcast are those of Dr. Sarah Matt and her guests. They do not necessarily reflect the official policy or position of any affiliated institutions. This content is for informational and educational purposes only and does not constitute medical advice or a professional consulting relationship.

Transcript

SPEAKER_00 0:00

Welcome back to the Clinical Realist. I'm Dr. Sarah Matt. It's late April. And right about now, if your health system deployed a clinical AI tool in Q1, you're starting to notice something. The pilot worked. The go live went pretty reasonably well. And the vendor support team has been very responsive. Clinicians are using the tool less. But something's off. Not catastrophically off, just off enough that you are spending more time and conversations about the tool than you expected at this point. Clinicians are asking questions you don't have clean answers to, and the data is coming in, but nobody is sure what to do with it. The vendors, of course, still engage, but you're starting to realize that their support model was built for GoLive and not for the six-month steady state. If that description sounds familiar, what your experience is a governance gap, not a technology gap, a governance gap. And the good news is it's fixable. The bad news is that the longer you wait, the more expensive it gets. Clinical AI implementations have a predictable pattern after GoLive. The first four weeks are usually the honeymoon phase. Things are new, people are engaged, the vendor is attentive. Issues get resolved fast, and the data looks better than expected because the early adopters are doing the heavy lifting. Then weeks four through eight happen. The early adopters are no longer carrying the teen. The full clinical population is using the tool now, or supposed to be. The vendor support team has deprioritized your account because GoLive is behind them. The implementation team has moved to the next project. And the first real data is coming in, not pilot data, not the first week's novelty data, actual production data from actual clinical operations. And it's raising questions. Why is the tool flagging this class of patients at a different rate than the pilot suggested? Why are these three clinicians using the tool in a way that was not in the training protocol? And why is the recommendation workflow generating pushback from the afternoon shift that the morning shift isn't experiencing? These are not technical questions, they're governance questions. And they require someone in the organization who has the absolute authority to look at them and look at the data, make a judgment about what it means, and decide what needs to be changed. If you built a governance structure before Go Live, you should have that person already, which is great. And that person has been looking at the production data since day one. They have a 30-day review built into the calendar, and they know exactly what they're seeing and what it means. If you didn't build that structure, you're probably improvising. And the questions are real. The data's real. But the person who's supposed to answer them doesn't have a clear mandate. The data review meaning is not on the calendar. And the question of why this is not working looks like it has never been formally defined. The six to eight week window is when that improvisation becomes visible. Not because the tool failed, because the absence of governance becomes apparent in the data. Let me describe what a governance gap actually looks like in a live clinical environment, because it does not announce itself dramatically. It looks like this. And the clinical champion sends an email to the physician population asking, is anyone having issues with the tool? So then three responses come back. Two are vague, and one is from the department chief who says, We talked about this in last month's meeting. So no one knows what meeting that was, and nobody is following up. So one month later, the utilization number is lower, and the same loop repeats. This isn't negligence. Everyone in this loop is a competent professional doing their job the best they can. The problem is structural. There's no designated owner, there's no decision framework, and there's no defined threshold that triggers an actual formal review. There's no meeting with authority to even make a change. The governance gap is not a gap in talent, it's a gap in structure. Now let's contrast this with what a health system with pre-built governance looks like at the same moment. The monthly data report comes in and the physician steward reviews it against the predefined performance benchmarks established at GoLife. Utilization is below benchmark. The steward calls a 45-minute governance review with three prenamed participants. And at that meeting, they use a standard protocol to determine whether the utilization gap is adoption-related, workflow-related, or tool related. They assign an action to the responsible party within a two-week resolution timeline. And the governance record is updated. Same data, completely different response. Because the structure was built before the tool went live. Now here's the economics question I get asked most often in this conversation. Is it worth building the governance structure after the fact? We're already live. Can we retrofit? Yes. Yes, yes, yes, you can retrofit. But retrofitting governance onto a live clinical AI tool, it's much harder than building it before go live. And it's a lot more expensive. And here's why. When you build governance before go live, you're building from a blank slate. You define what looks good before you have data. You name the steward before there's a crisis. And you establish the review cadence before there's a backlog. So the structure's clean. Now, you can retrofit, but when you retrofit, you're building governance while the tool is alive, which means every structural decision you make immediately generates a question about the existing data. So what do we do with the last six weeks of utilization data that nobody reviewed? Do we apply the new governance framework retroactively? And who owns the backlog of unresolved clinical questions? You're building the safety net while people are already falling. It's still worth doing. And if you're, you know, in this position right now, the answer is not to wait until Q3 and start fresh. No, don't do that. The answer is to build the structure as fast as possible and apply it from today forward. But the organizations that are not in this position spend a fraction of the retrofit cost building governance before go live. That's the strategic investment that looks expensive before implementation, but very obvious afterwards. So if you're listening to this episode and your organization is currently in the governance gap, here's the fastest way forward. Step one, name the steward today, not tomorrow, today. Not a committee, one person, a clinician who has the clinical credibility, the organizational standing, and the time to function as the governance owner for this tool. A nurse, a doctor, a pharmacist, I don't care who it is, one person. Then give them the title and the authority in the same conversation. Step two, schedule the darn governance review, the first one this week, and use the production data you have to find the review questions in advance. What are we seeing in utilization? What are the edge cases that have come up in the last six weeks? What are the open clinical questions that nobody has actually answered? One meeting, two hours max, document the outputs. And step three, build the forward governance calendar. 30-day reviews, quarterly deep dives, an escalation protocol for the scenarios that trigger a pause. Put it on the calendar today. Not tomorrow, today. Not when you have time to do it. I mean, you gotta do it now. Like do it now. Do it yesterday. And then step four, define this is not working and what that looks like. This is the hardest conversation and probably the most important one. What performance threshold triggers an actual formal pause for assessment? What clinical outcome triggers immediate escalation? And what utilization floor represents a truly failed implementation? Define these thresholds before you get to them. It's much harder to find them after. Four steps. You can have a functional governance structure running within two weeks from today. That's right. Two weeks from today. The Q2 governance problem is predictable. It happens to most organizations that deployed clinical AI and Q1 without a pre-built governance structure. It's not a reflection of bad intent or bad tech. It's a structural gap that becomes visible at a specific moment in the implementation lifecycle. If your organization is in that gap right now, the fastest path out is a direct conversation about what the governance structure needs to look like for your specific situation. That's the work I do. Discovering Clarity Sessions at the link below. I'm Dr. Sarah Matt. This is the Clinical Realist, and I'll see you next week.