Why wisely using data requires insight into potential variables
The system built on one state’s math regarding the No Surprises Act shows what can happen when execs forecast from the wrong population.
Before the No Surprises Act went into effect in 2022, federal regulators needed to estimate how many billing disputes its new arbitration system would need to handle each year. They landed on roughly 17,000 disputes annually.
By the end of 2025, the independent dispute resolution portal had received 4.8 million disputes since launch, roughly 280 times the original estimate.
The No Surprises Act is working. Surveys and federal data both show it has shielded millions of patients from unexpected out-of-network bills. The failure in setting an estimate lies one layer down, in the infrastructure built to support it, which was sized using a forecasting model built on data from the wrong population.
Getting a number 280 times too small
The 17,000 figure was not arbitrary. Federal regulators built their estimate around the arbitration experience of the states that already had a state-level dispute process, most notably New York, which pioneered its own independent dispute resolution system for surprise bills in 2014 and had handled about 1,000 disputes a year under it.
The model had at least two structural flaws. It did not separate New York residents covered by state-regulated plans, who were already subject to the state's balance billing law, from those covered by federally regulated plans who were not previously protected at all. And it assumed the rest of the country's dispute behavior would mirror that of one state, when the federal law's scope, financial incentives and provider participation patterns differed substantially from that of New York.
A denominator error at the design stage became a capacity crisis at the operational stage. That progression of a small data assumption becoming a large downstream failure, is a pattern healthcare executives should recognize immediately. It shows up constantly in far less visible enterprise forecasting decisions, ranging from staffing models to capital budgets.
The gap is more revealing when measured relative to the original plan rather than in raw dispute counts. Volume ran at roughly 22 times the projected pace in the program's first two years, climbed to 86 times the pace in 2024, and reached 153 times the pace in 2025. The system was not simply undersized once. It fell further behind its own design target every single year.
What the undersized system cost
The consequences of the miscalculation were not abstract. The IDR process has generated at least $5 billion in costs since 2022, with administrative expenses alone accounting for more than half of that total.
A system sized for 17,000 disputes a year could not process millions without severe delays. As of mid-2025, about two-thirds of determinations were taking longer than the 30-day window required by law, and 430,000 disputes remained backlogged.
Since then, regulators have made real progress. Automated eligibility checks, additional certified arbitration entities and streamlined processes helped IDR entities close more disputes than were filed for much of 2025 and early 2026, substantially reducing the backlog.
That recovery is a genuine data management success story, but it was reactive. It fixed a capacity problem that better upfront population modeling could have anticipated.
The cost of substandard forecasting
The tri-department estimate that produced the estimate of 17,000 annual disputes was a single-variable scaling calculation; it was based on one state's dispute count multiplied by a population share.
Healthcare organizations increasingly use far more sophisticated tools for this kind of problem. Machine learning demand forecasting models are already used to predict hospital bed occupancy, staffing needs and emergency department volume by combining many data sources and accounting for behavioral incentives, rather than assuming uniform behavior across a population.
A federal agency does not need to become a technology company to fix this. It needs to treat a national population estimate as seriously as any other high-stakes forecasting problem. It must pressure-test the sample, model the financial incentives that could drive volume up and validate the model against early real-world data before finalizing a system's capacity.
The same accuracy rules should apply to any healthcare executive sizing a new program, staffing model or value-based contract on a projected population.
Priorities for healthcare executives
The No Surprises Act protected patients exactly as intended. The infrastructure built to support it did not fail because the policy was wrong. It failed because the population estimate behind it was built on a sample of one state, generalized in a way the underlying data could not support.
Across all three examples in this series, the lesson is ultimately the same. Healthcare executives cannot govern what they cannot see, cannot control what they cannot independently measure, and cannot safely forecast from data that does not represent the population they are trying to serve.
Better technology helps. AI will help further. But neither replaces the executive's responsibility to understand where the data came from, what it represents and whether it is strong enough to support the decision being made.
Julia Rehman, DHA, FACHE, FACHDM, is founder and chief operating officer of Kota Kompany LLC.