№ 01
Cash-flow forecast
Quantile forecasts of realized revenue (p10 / p50 / p90) per period. The lower tail is load-bearing — bonus pools size against p10, not the mean.
A research instrument for primary-care RCM
Forecast the cash
that actually arrives.
Clinics in Harris County bill millions and collect a fraction of it — Texas Medicaid pays primary care at 52¢ on the Medicare dollar. Sizing physician bonuses against billed revenue is a working-capital trap. Cost Predictor sizes them against revenue that will land.
What it does
Three answers, one model.
№ 02
Payer-mix sensitivity
Right-censored lag curves and Beta gross-collection priors per payer (Medicare FFS, MA, Medicaid FFS/MCO, commercial, self-pay) decompose the forecast into the levers that move it.
№ 03
Bonus-pool sizing
One worked answer: a $5M-monthly clinic sizing 5% of p10 leaves roughly $463K on the table per 60-day window versus a $500K perfect-forecast counterfactual — and zero over-allocation risk.
In one paragraph
From a billed series to realized cash, by way of survival analysis.
The pipeline ingests aggregated billing (currently Synthea-generated; partner-clinic 835 ERA next), forecasts gross billed with a quantile time-series model, splits each forecast period by payer mix, and lag-shifts each payer's dollars through a hybrid Kaplan–Meier + LogNormal curve fit on right-censored payment-delay observations. The result is a per-period distribution of realized dollars — the number a clinic admin can write a bonus check against.
Approach A — a single pooled time-series forecast of realized revenue — looks fine in aggregate (0.0998 pooled row breach) but still flags payer-specific row breaches. Approach B passes every payer and every summed bonus window. Why →
What this is — and isn't
A research POC, calibrated to public sources.
Cost Predictor is a Python research artifact, not a commercial product. It runs on synthetic patient data from Synthea v4.0.0 (Harris-County demographics, Texas Medicaid MCO payer encoding) and calibrates per-payer denial / lag / paid-claim realization priors against published sources:
- KFF · Medicaid-to-Medicare fee index, 2024 (Texas primary-care 0.52)
- HRSA UDS Texas Table 4 — per-FQHC payer mix
- Premier Inc. 2024 hospital denial survey (~15% commercial / Medicaid initial denial)
- Texas Department of Insurance prompt-pay statutes (30/45-day floors)
- HFMA MAP Award 2024 — self-pay collection benchmarks
Stated honestly
No real 835 ERA data has been ingested yet. Accuracy is bounded to "directionally useful for cash-flow risk management against the configured payer mix" — not a substitute for clinic-specific historical adjudication data. The single highest-leverage upgrade is a partner-clinic 835 feed under BAA.
Three ways in.
Spend ninety seconds in the live Monte-Carlo demo, read the methodology end-to-end, or install locally and run the experiment runner against the Synthea fixture.