AI-facilitated constraint resolution for healthcare systems
How AI can dissolve complex multi-stakeholder conflicts by surfacing hidden assumptions, not by replacing human judgment.
The Familiar Impasse
Every health system knows this meeting. Three service lines present growth plans. Each one says the same thing: “We need more space.” The board sees a capital problem that will take years and tens of millions of dollars to solve. The conversation stalls. Someone proposes a feasibility study. The meeting ends where it began.
But imagine someone asks a different question: “What’s happening in those rooms at 2 AM?”
The answer is nothing. The rooms are empty.
This is not a space problem. It is a systems problem. And systems problems have systems solutions.
Why Smart People Get Stuck
The impasse is not caused by lack of intelligence. It is caused by the structure of the problem itself.
A typical capacity question involves at least four stakeholder groups, each holding legitimate requirements that appear to conflict with one another.
Governance demands fiscal responsibility and demonstrated utilization before committing capital. Executives need growth now, not in five years when a new wing opens. Medical staff need safe conditions and sustainable schedules that do not burn through their workforce. Patients and the communities they represent need timely access to care.
A skilled human facilitator can hold three or four of these tensions simultaneously and make progress. But the real problem has dozens of interacting constraints, not four. No one can see the whole system at once. So the conversation collapses into positional negotiation, where each group advocates for its own priorities, or it defaults to the most expensive option: build.
Each line represents a tension between stakeholder requirements. No single person can see all of these at once.
Conflicts Are Made of Assumptions
Eli Goldratt, creator of the Theory of Constraints, demonstrated that most organizational conflicts are not genuine dilemmas. They are artifacts of unstated assumptions.
Consider a statement that sounds like a hard fact: “We cannot extend procedural hours because we cannot staff them.” This single sentence encodes at least three assumptions. Extended hours means overnight shifts. Staffing must come from current employees working additional hours. Every shift must be financially self-sustaining in isolation.
Challenge any one of those assumptions and new options appear. Perhaps extended hours means 6 AM starts, not midnight operations. Perhaps staffing comes from a regional float pool or a partnership with a nearby system. Perhaps the shift pays for itself across the service line rather than in isolation.
The difficulty is not that these assumptions are wrong. It is that they are invisible. They hide inside dozens of interacting requirements, in the spaces between departments, between policy documents, between what people say in meetings and what they take for granted.
A Different Kind of AI Application
The healthcare industry’s current conversation about AI focuses on workflow automation. AI scribes, documentation assistants, scheduling optimizers. These are valuable, but they are incremental. They make existing processes faster.
A more fundamental application is using AI to solve complex constraint satisfaction problems that no human team can hold in working memory.
Given structured requirements from each stakeholder group, an AI system can identify every conflict between requirements across all groups simultaneously. It can surface the hidden assumptions creating each apparent conflict. It can generate solutions that dissolve conflicts rather than compromise between them. And it can trace how any proposed change ripples across all groups, revealing second-order effects that would take a human team weeks to map.
This is not AI replacing human judgment. It is AI making human judgment tractable. The humans still decide what matters, what trade-offs are acceptable, and what risks they are willing to carry. The AI holds the full complexity so the humans can focus on the decisions that require wisdom, not just working memory.
How the Process Works
The process unfolds across five phases, each building on the last.
Phase 1: Requirements Gathering. Each stakeholder group independently develops their requirements. These are specific, concise statements with underlying rationale. Not wish lists. Not strategy decks. Concrete needs with the reasoning behind them. This phase is deliberately separated by group so that each can articulate their needs without immediately negotiating against others.
Phase 2: Assembly and Conflict Mapping. All requirements are assembled into a single structure. AI maps every conflict, identifies clusters of tension where multiple requirements interact, and surfaces the embedded assumptions holding each conflict in place. The output is a complete map of the problem space, something no human team could build by hand.
Phase 3: Assumption Challenging. This is where breakthroughs happen. Working sessions examine each AI-identified assumption. Which constraints are truly fixed by regulation, physics, or contract? Which ones feel fixed but are actually organizational choices? Which ones were true five years ago but are no longer? A surprising number of “hard” constraints turn out to be soft when examined carefully.
Phase 4: Solution Generation and Iteration. AI generates solution directions that satisfy the maximum number of requirements simultaneously. The group reacts. They push back, add constraints the model missed, flag political realities. AI revises. Each round narrows toward solutions that all parties can genuinely support, not because they compromised, but because the underlying conflicts were dissolved.
Phase 5: Validation. Stakeholders stress-test the proposals through interactive AI sessions. What happens if hiring takes six months longer than planned? What if patient demand grows 20% faster than projected? What if a key regulation changes? The AI models these scenarios, showing where solutions hold firm and where they need reinforcement. The result is not a fragile plan but a robust one with known boundaries.
The output is not a consultant’s recommendation handed down from outside. It is a solution the stakeholders built together, with AI serving as the tool that made the full complexity visible and workable.
What This Makes Possible
This approach changes the fundamental character of the conversation. Instead of “your needs versus mine,” the framing becomes “here are all our needs, and here is a way to satisfy them that none of us would have found alone.”
Problems that would take months of committee work can be structured in weeks. Solutions that seemed impossible become achievable once the assumptions propping up the impossibility are examined. Capital expenditures that appeared inevitable may prove unnecessary when the underlying systems problem is solved differently. And the alignment that emerges is durable, because it comes from genuine satisfaction of each group’s needs rather than from grudging compromise.
This is not a theoretical possibility. The Theory of Constraints has been producing these results in manufacturing and operations for decades. What has changed is that AI now makes it possible to apply the same rigorous logic to problems involving dozens of stakeholder requirements interacting across multiple dimensions simultaneously.
Next Step
If your organization is facing a complex operational challenge with competing stakeholder needs, AI-facilitated constraint resolution may offer a path forward. We would welcome a conversation about whether this approach fits your situation.