Why Cost-Allocation Accuracy Matters More Than Hours Saved

§A — What cost allocation actually does

A Medicare cost report does two related things. First, it captures the provider's total costs and revenues for the fiscal year. Second, and more consequentially, it allocates those costs across cost centers so that Medicare pays for only the share attributable to Medicare beneficiaries. That second step is cost allocation, and it is the part that determines settlement.

The mechanics, per PRM Pub. 15-1 Chapter 23 and form-specific instructions in PRM Pub. 15-2:

• Cost-center setup (Worksheet A). Trial-balance accounts are reclassified into CMS cost centers (administrative + general, plant operations, nursing, physical therapy, etc.). The mapping must reflect the operational reality of the provider — not a default template.
• Adjustments (Worksheet A-6 + A-8).Reclassifications + adjustments are documented with their rationale + source. Adjustments without supporting documentation are the most common desk-review finding.
• Step-down allocation (Worksheet B + B-1). Overhead cost centers are apportioned to revenue-producing cost centers using statistical bases — square footage, FTE hours, patient days, etc. The bases must reflect actual usage patterns, not round-number defaults.
• Settlement reconciliation (Worksheet D for HHA; Worksheet F for Hospice). Allocated costs are compared to interim Medicare payments to compute the settlement variance — overpayment or underpayment.

Cost-finding accuracy at each of these steps compounds. An adjustment misclassified on Worksheet A propagates through step-down on Worksheet B, alters the cost-per-cost-center on Worksheet C, and lands on Worksheet D/F as a settlement variance against the wrong cost basis. The MAC's desk review checks this chain. The reopening process under 42 CFR § 405.1885 exists precisely because the chain can break.

§B — Why accuracy matters more than hours

The CMS Paperwork Reduction Act estimates put the cost-report filing burden at 195 hours for HHA (OMB 0938-0022) and 188 hours for Hospice (OMB 0938-0758). These are the official figures — not industry estimates, not extrapolations. They represent the hours CMS expects a competent preparer to spend on a single annual cost report.

A tool that cuts those hours in half but introduces a cost-allocation error doesn't save the provider time — it shifts the work from filing to MAC interaction. The replacement cost shows up in:

• Desk-review back-and-forth. The MAC has 30 days under 42 CFR § 413.24(f)(5)(iii) within which the provider must cure any deficiencies; there is no parallel hard deadline for the MAC to render its determination. Each cycle of question-and-response on an opaque allocation is staff time + documentation effort + delay.
• Settlement variance. A wrong allocation produces a wrong settlement figure on the NPR. If Medicare ends up owing the provider less than the actual cost-share allocation would have produced, the provider is paying for Medicare beneficiaries out of pocket. If Medicare ends up owing more, the provider is on the hook for repayment.
• Reopening risk. Errors found later trigger reopening under 42 CFR § 405.1885 — generally within three years of the NPR, longer for fraud or similar fault. A reopened cost report consumes staff time + creates audit attention + signals to MAC pattern-analytics that the provider's reports are unreliable. The downstream cost exceeds the upstream hours saved.
• Documentation adequacy under enforcement. The cost-finding standard at 42 CFR § 413.24 requires the provider to maintain "adequate cost data" such that costs can be accurately determined. When CMS or the OIG examines a report, the standard is the documentation quality, not the filing speed.

§C — The deterministic-engine principle

The right architectural division of labor between AI and a deterministic engine is the difference between a tool that ships hours-saved-as-a-feature and a tool that ships accuracy-as-a-feature.

AI is excellent at classification. Mapping a general-ledger account named "Maintenance — Building" to the CMS cost center "Plant Operation and Maintenance" is a pattern-recognition task. Confidence-scored classification with human review at the boundary is a known good pattern.

AI is poor at arithmetic. Step-down allocation involves ratio computations across many cost centers, propagated across multiple worksheets, with rounding conventions specific to PRM Pub. 15-1 Chapter 23. Asking an LLM to produce the numbers directly invites:

• Non-reproducible outputs. The same inputs can produce different numbers across runs. A cost report that doesn't reproduce deterministically cannot be audited.
• Floating-point drift. Monetary calculations in floats accumulate rounding error. The CMS instructions specify the rounding methodology; AI output bypasses it.
• Untraceable derivation. An AI-generated number has no formula chain. The cost-finding standard at 42 CFR § 413.24 expects each number to be derivable from documented inputs.

The right architecture — the one we build to — is simple: AI classifies, a deterministic engine calculates, no AI output ever lands in a numeric column on the cost report. Every customer-visible number has a corresponding audit-trail entry with its formula, its inputs, and the regulatory citation that authorizes the calculation. When the MAC asks where a number came from, the answer is one query away.

§D — What to evaluate when picking cost-report tooling

A practical evaluation checklist for cost-report tooling, ordered by importance:

1. Audit-trail completeness. For each line on each worksheet, can you retrieve the formula, the inputs, and the regulatory citation that authorized the calculation? If the answer requires more than one click-through, the tool is not documentation-adequate.
2. Decimal arithmetic on money. Are monetary values stored as Decimal types (or as `numeric` with appropriate precision at the database level), or as floats? Floats accumulate rounding error and fail the rounding-convention discipline PRM Pub. 15-1 Chapter 23 expects. This is a question to ask in the demo.
3. Step-down methodology fidelity. Does the tool implement the allocation methodology specific to your form (CMS-1728-20 for HHA, CMS-1984-14 for Hospice), or does it apply a generic step-down? PRM Pub. 15-2 specifies form-specific instructions; the tool should match those, not approximate them.
4. Reproducibility. Re-running with the same inputs should produce identical outputs — every cell, every cent. A re-run during a reopening years later should match the original filing exactly. Ask the vendor to demonstrate this with a synthetic fixture.
5. Cross-worksheet reconciliation. Worksheet totals should reconcile automatically: A → B → C → D/F. The tool should block submission if reconciliation fails, not warn-and-proceed. PRM Pub. 15-2 reconciliation rules are non-negotiable; the tool should treat them the same way.

We're building cost-report preparation tooling around this evaluation framework. Not a CMS-approved vendor yet — pursuing that designation through the Office of Financial Management Part A Cost Report Division. Whether you use a CPA firm, a billing agency, or an independent consultant, the documentation standard under 42 CFR § 413.24 is the same, and your authorized representative signs the cost report regardless of preparer. The accuracy of what they sign is what matters.

§E — Frequently asked questions