Allocation Under Scarcity

Budget Allocation Is an Optimization Problem, Not a Negotiation

Every healthcare organization allocates scarce resources. Grant-funded transformation programs allocate dollars across competing investments. State Medicaid agencies allocate waiver funds across counties. Health systems allocate staffing hours across clinics. The question is not whether allocation happens, but whether it happens well.

In practice, most healthcare allocation decisions are made through negotiation. Department heads argue for their budgets. Counties lobby for their share. Whoever builds the most compelling narrative — or has the most political capital — gets funded. The result is a portfolio of investments that reflects bargaining power, not expected impact. This is not a failure of character. It is a failure of method.

Constrained optimization offers a different method. It starts with an explicit objective (what are we trying to maximize?), an explicit budget (what do we have?), and explicit alternatives (what could we fund?). It then selects the combination of investments that produces the highest total value within the constraint. The output is not one person’s judgment. It is the mathematically best portfolio given the stated inputs.

The gap between negotiated allocation and optimized allocation is not theoretical. It is measurable, it is usually large, and it represents real access, real outcomes, and real lives left on the table.

Why Allocation-by-Negotiation Fails

Negotiated allocation has four structural defects that no amount of good faith can fix.

Loudest voice wins. In budget negotiations, the department with the most articulate leader, the most dramatic patient story, or the longest tenure gets disproportionate resources. This is not corruption — it is human decision-making operating without a framework for comparing unlike alternatives. A telehealth expansion and a provider recruitment initiative cannot be compared by narrative alone, because they produce different kinds of value on different timescales. Without a common metric, persuasion substitutes for analysis.

Historical inertia dominates. The single strongest predictor of next year’s allocation is this year’s allocation. Programs that were funded continue to be funded. Programs that were not funded remain unfunded. This anchoring effect — well-documented in behavioral economics by Tversky and Kahneman — means that the allocation portfolio drifts further from optimality each year, because the environment changes but the budget does not. A community health worker program that would produce twice the access gain per dollar as an underperforming IT modernization project will never displace the IT project if “we’ve always funded IT modernization” carries the day.

Tradeoffs remain implicit. When a committee approves a $400,000 investment in care coordination, the question that is almost never asked is: what did we just decide not to fund? Negotiated allocation obscures opportunity cost because each investment is evaluated in isolation. The committee asks “is this worth doing?” (almost always yes) rather than “is this the highest-value use of these dollars?” (requires comparison). The result is a portfolio where every individual investment is defensible but the total portfolio is suboptimal.

No mechanism discovers opportunity cost. This is the deepest failure. Negotiation has no built-in process for surfacing what was sacrificed. The opportunity cost of funding Program A is the value that Program B would have generated — but in a negotiation, Program B’s value was never quantified in comparable terms, so the cost of choosing A over B is invisible. Optimization makes opportunity cost explicit by computing the value of every feasible portfolio, not just the one that was funded.

The Knapsack Problem: A Precise Formulation

The mathematical structure underlying healthcare budget allocation is the knapsack problem, one of the most studied problems in combinatorial optimization (Dantzig, 1957; Kellerer, Pferschy, and Pisinger, 2004). The formulation is direct:

You have a fixed budget (the knapsack’s weight capacity). You have a set of candidate investments, each with a cost (weight) and an expected value (benefit). You must select a subset of investments whose total cost does not exceed the budget and whose total value is maximized.

For indivisible investments — you either fund the telehealth expansion or you do not — this is the 0-1 knapsack problem, which is NP-hard in general but solvable for practical problem sizes using dynamic programming or branch-and-bound methods. For divisible investments — you can fund 60% of a program — the problem reduces to the fractional knapsack, solvable by the greedy algorithm: rank investments by value-per-dollar, fund them in order until the budget is exhausted.

The key insight is not the algorithm. It is the formulation. The act of listing all candidate investments, estimating their costs and expected impacts in comparable units, and computing the optimal portfolio forces a rigor that negotiation never achieves. Even if the estimates are rough, the structure reveals the shape of the tradeoff.

Opportunity Cost Made Operational

Opportunity cost is an introductory economics concept that most healthcare leaders can define but few operationalize. Optimization makes it concrete.

When you solve a knapsack problem, you do not just get the optimal portfolio. You get the value of the next-best alternative — the investment that was excluded and the impact it would have generated. If the optimal portfolio produces 8,200 additional patient visits per year and the second-best portfolio (with one substitution) produces 7,400, then the opportunity cost of choosing the second portfolio is 800 patient visits. That number — 800 visits — is the price of the suboptimal decision, stated in operational units.

Cost-effectiveness analysis (CEA) formalizes this comparison. The incremental cost-effectiveness ratio (ICER) measures the additional cost per additional unit of outcome when choosing one intervention over another. In health technology assessment, NICE in the UK uses a threshold of 20,000 to 30,000 pounds per quality-adjusted life year (QALY); the US commonly applies $50,000 to $150,000 per QALY. These thresholds represent the system’s implicit willingness to pay — and by extension, the opportunity cost of displacing the marginal intervention that would otherwise be funded.

For operational allocation decisions that are not about pharmaceuticals or devices but about program investments, the relevant ICER analog is impact-per-dollar: patient visits generated per dollar, wait-time days reduced per dollar, or lives covered per dollar. The units depend on the objective. The discipline is the same: rank alternatives by incremental value per incremental cost, and fund from the top.

Healthcare Example 1: Rural Health System Transformation Funding

A rural health system receives $2 million in HRSA transformation funding. The transformation team identifies six candidate investments:

Total cost of all six: $2.13 million. Budget: $2.0 million. Not everything fits. Something must be cut.

The negotiated outcome. In a typical committee process, the CMO argues for provider recruitment (“we’re down two physicians”), the CIO argues for IT modernization (“our EHR is three versions behind”), and behavioral health integration has political momentum from the board. The committee funds all six but scales each down by 6%, producing diluted programs that underperform their projections. Or they cut community health workers because CHWs have the least political constituency, keeping the other five at $1.85 million and using the remaining $150K for “contingency.”

The optimized outcome. Rank by access-per-dollar: community health workers (857), telehealth (800), care coordination (700), BH integration (640), provider recruitment (400), IT modernization (160). Fill the knapsack greedily: CHWs ($280K, cumulative $280K), telehealth ($350K, cumulative $630K), care coordination ($300K, cumulative $930K), BH integration ($500K, cumulative $1.43M), provider recruitment ($450K, cumulative $1.88M). Remaining budget: $120K — not enough for IT modernization at $250K. Optimal portfolio: five investments totaling $1.88M, generating an estimated 12,300 patient-visits. The $120K remainder either funds a partial initiative or rolls into contingency.

The negotiated portfolio that cuts CHWs and funds IT modernization instead generates an estimated 10,300 patient-visits — 2,000 fewer than the optimized portfolio. That gap is 2,000 patient-visits per year. It is the cost of allocation-by-committee.

The optimization does not say IT modernization is worthless. It says that at 160 access-gain per $100K, it is the lowest-return investment in the portfolio, and when the budget is binding, it is the one that should yield. If the leadership team decides to fund IT modernization anyway — because the EHR risk is real — the optimization quantifies that decision’s cost: 2,000 patient-visits per year. That is an informed tradeoff, not an invisible one.

Healthcare Example 2: State Medicaid Waiver Allocation Across Counties

A state Medicaid agency has $18 million in 1915(c) waiver funds to allocate across eight counties for home and community-based services. Each county has different need levels (measured by eligible population), different absorptive capacity (existing provider networks, administrative infrastructure), and minimum coverage requirements (no county can receive zero).

This is a transportation/assignment problem: distribute a fixed supply of funding across destinations with heterogeneous demand and constraints. The objective might be to maximize total individuals served, to minimize the maximum unmet need across counties, or to equalize coverage rates — and these objectives produce different allocations.

Equal allocation gives each county $2.25 million. But County A (population 120,000, 4,200 eligible) and County H (population 18,000, 480 eligible) have radically different needs. Equal allocation overfunds County H relative to its eligible population and underfunds County A relative to its need.

Need-proportional allocation distributes by eligible population. County A gets $5.4 million; County H gets $614K. But County H lacks the provider infrastructure to absorb even $614K effectively — it has two HCBS providers and no care management agency. Dollars allocated beyond absorptive capacity produce no services.

Constrained optimization incorporates both need and capacity. Minimize total unmet need subject to: (1) each county receives at least a minimum floor (say $400K) to maintain basic services, (2) no county receives more than its absorptive capacity ceiling, (3) total allocation equals $18 million. The solution may allocate $4.8 million to County A (high need, high capacity), $400K to County H (low need, low capacity), and distribute the remainder across mid-range counties weighted by unmet need.

The optimization also produces shadow prices — the marginal value of relaxing each constraint. If the shadow price on County H’s capacity ceiling is high, that signals: investing in County H’s provider infrastructure would unlock more value than simply sending more waiver dollars. The constraint, not the funding, is binding. This is actionable intelligence that no negotiation process generates.

Multi-Criteria Allocation: When There Is No Single Right Answer

Most real allocation decisions involve multiple objectives that cannot be collapsed into a single metric. A state might want to maximize total access (efficiency) while also ensuring no county falls below a minimum coverage threshold (equity). A health system might want to maximize patient visits (volume) while also improving behavioral health access specifically (strategic priority).

When objectives conflict, optimization does not produce a single answer. It produces a Pareto frontier — the set of all allocations where improving one objective requires worsening another. Every point on the frontier is efficient; no reallocation can improve one metric without degrading another. Points below the frontier are dominated — there exists a feasible allocation that is better on every dimension.

The Pareto frontier is not a decision. It is a decision-support tool. It tells leadership: “Here are your real options. Portfolio A maximizes total visits but leaves two counties below minimum coverage. Portfolio B equalizes coverage rates but produces 15% fewer total visits. Portfolio C is the compromise — 8% fewer visits than A, but all counties above threshold.” The optimization did the analytical work. The decision remains human.

This is the mature use of constrained optimization in allocation: not replacing judgment, but structuring it. Decision-makers who see the Pareto frontier make better tradeoffs than decision-makers who negotiate in the dark, because the frontier makes the cost of every preference explicit.

Political and Organizational Constraints

Real allocation decisions include constraints that are political, not technical. “Every county must receive at least as much as last year.” “The behavioral health line item cannot decrease.” “All regional offices receive equal base funding.” These are not optimization constraints in the mathematical sense — they are policy choices. But they can be encoded as constraints and their cost can be computed.

If the policy requires equal base funding of $1.5 million per county (8 counties, $12 million committed before optimization begins), and only $6 million is available for need-based distribution, the optimization can compute the total access cost of that political constraint. Suppose unconstrained optimization (with only the minimum-floor constraint) produces 14,200 individuals served, but the equal-base-funding constraint reduces that to 11,800. The political constraint costs 2,400 individuals their services. That number should be in the briefing document. The decision to maintain equal base funding may be the right one — political legitimacy has real value — but it should be made with the cost on the table.

This is the fundamental contribution of optimization to political allocation decisions: not overriding politics, but pricing politics. Every political constraint has a shadow price measured in outcomes forgone. Making that price visible transforms allocation from a negotiation (where costs are hidden) into a structured decision (where costs are explicit).

Warning Signs of Suboptimal Allocation

Equal allocation as a default. When every department, county, or program receives the same amount regardless of need, capacity, or expected return, the allocation is optimizing for political simplicity, not impact. Equal allocation is only optimal when all recipients have identical need and identical capacity — a condition that essentially never holds.

Allocation unchanged from prior year. If this year’s budget looks like last year’s budget plus or minus a uniform percentage, no optimization occurred. The environment changed — need shifted, costs changed, new alternatives emerged — but the portfolio did not respond.

No explicit objective function. If the allocation committee cannot state what they are maximizing (access? coverage? equity? some weighted combination?), they cannot evaluate whether the allocation achieves it. An unstated objective function defaults to “minimize conflict,” which is an objective — just not one that serves patients.

Inability to state what was sacrificed. Ask any program manager what was not funded to make their program possible. If they cannot answer, the allocation process never surfaced opportunity cost. Every yes has a corresponding no. If the no is invisible, the yes was not analytically informed.

No sensitivity analysis. If the allocation does not change when the estimates change, either the estimates do not matter (unlikely) or no one tested them. A robust allocation process identifies which assumptions drive the portfolio composition and flags the investments whose inclusion is sensitive to estimation error.

Product Implications

Allocation modeling as a software capability. A grant management or program planning tool should allow operators to enter candidate investments with estimated costs and impacts, set a budget constraint, and compute the optimal portfolio. This is not a research tool. It is a planning tool that replaces spreadsheet-based negotiation with structured optimization.

Pareto frontier visualization. When multiple objectives exist, the tool should compute and display the efficient frontier, allowing decision-makers to explore tradeoffs interactively. “Drag the equity slider and watch the total-access number change” is more powerful than any briefing memo.

Shadow price reporting. Every political or policy constraint should display its cost in outcome units. “The equal-funding constraint reduces projected access by 2,400 individuals” belongs on the allocation dashboard, not buried in an analyst’s workbook.

Scenario comparison. The tool should store and compare multiple allocation scenarios side by side — the optimized portfolio, the negotiated portfolio, last year’s portfolio — with impact projections for each. The gap between scenarios is the actionable metric.

Degradation signal: allocation-to-need correlation. Track the correlation between funding allocated and measured need across recipients over time. A declining correlation means the allocation is drifting from need-responsiveness toward inertia or politics. This is the earliest metric of allocation quality degradation — it moves before outcomes do.

Integration Hooks

Public Finance M6 (Financial Controls and Scenario Planning). Grant budget allocation is the direct application domain for constrained optimization. The knapsack formulation maps precisely onto the grant program manager’s problem: fixed award, competing line items, uncertain returns, reporting requirements as constraints. The scenario planning tools described in Public Finance M6 should embed the optimization logic described here — scenario analysis without optimization is just storytelling with numbers.

Workforce M6 (Workforce Economics and Capacity Planning). Staffing allocation across sites is a constrained optimization problem with workforce-specific constraints that general budget optimization does not capture: licensure requirements (a behavioral health counselor cannot substitute for a psychiatric nurse practitioner), scope-of-practice limitations (which vary by state), relocation and recruitment lead times, and minimum staffing ratios for regulatory compliance. The optimization framework is identical — maximize access subject to constraints — but the constraint set is richer and more tightly coupled to regulatory and labor-market realities.