What AI could return for your state Medicaid program: the economic framework
This independent analysis gives you a number to present to the board. Here is the methodology behind it, (and how to apply it to your program).
When a state agency or HIE goes to its board with an AI investment recommendation, one question comes before all others: what does this return?
Not what AI can do in theory. What the economic case is, grounded in evidence that holds up to scrutiny. That question has a credible answer. And here is the methodology behind it.
The economic evidence
In February 2026, NZIER, an independent not-for-profit economic consultancy, completed a rapid assessment of what international research shows about AI’s impact on health system costs1. The core finding draws on the National Bureau of Economic Research (NBER 2024): AI adoption in healthcare reduces health system costs by 5 to 10 percent2. Approximately 35 percent of those gains come from reduced administrative expenditure. The rest comes from higher medical productivity.
That estimate comes from peer-reviewed economic research, synthesized by NZIER - an independent not-for-profit consultancy that has been informing public policy in New Zealand for more than 65 years. The underlying source, NBER 2024, is not healthcare vendor research. It is economics research on healthcare productivity.
Applying the model to any state Medicaid program
The methodology applies to any state. Take the annual Medicaid expenditure, apply the 5 to 10 percent range, multiply across ten years.
The NBER research covers the US health system broadly. Medicaid populations - older, more rural, and carrying higher chronic disease burden than the general insured population - sit within the higher end of that range, where administrative inefficiency and avoidable hospitalization costs are greatest.
A rural US state illustrates the math. One program spent $8.49 billion in FY2024, covering 900,000 enrolees — 37 percent of whom live in rural areas, 14 percent carrying three or more chronic conditions. That profile sits squarely within the population the NBER research covers. Applied to that expenditure, the ten-year savings range runs from $4.2 billion to $8.5 billion - equivalent to roughly six months of additional Medicaid services from within existing expenditure3.
The same calculation runs for any state. The inputs are public. The methodology is in the report below.
The micro number: a ground-up check
The macro figure is useful for board conversations. A parallel model, built from the ground up on public data, shows what that looks like at the use-case level.
Orchestral’s analysis, built on CDC prevalence data4, CMS hospitalization costs5, and published adverse event rates6, projects that applying AI-assisted medication safety review to every diabetic patient’s prescribing in a rural US state over one year could prevent 105 hospital admissions for type 2 diabetics, with an estimated saving of $2.3 million in direct hospital costs.
One use case. One year. One state. Verifiable from public data.
The macro and the micro are built independently but point to the same place: the economic case for AI in Medicaid is calculable, from public sources.
The methodology
The macro figure and the medication safety model draw on the same foundation. Orchestral commissioned NZIER to assess what international research shows about AI’s impact on health system costs. Their rapid assessment synthesizes findings from NBER, OECD, and systematic clinical reviews across the WHO’s six health system building blocks1.
The full report is the economic framework. It applies to any state Medicaid program.
The H.R.1 window
For US state agencies and HIEs, the current funding environment adds timing to the economic case. H.R.1 creates a funding window tied to CMS technical scoring factors re-evaluated annually. Five program areas are directly addressable by predictive AI. Workload funding allocation follows demonstrated capability, not stated intent.
The economic case for acting is clear. What is less clear is which vendors can demonstrate measurable outcomes against those CMS scoring factors, year on year, with evidence that holds up to the same level of scrutiny as the analysis above.
That is the right question for your next vendor meeting. Not what predictive AI costs. What it costs to leave the return on the table while the funding window is open.
Download the report: Benefits of AI for the Health System
References
NZIER. Benefits of AI for the Health System: A Rapid Assessment. February 2026.
NBER. The Economics of Artificial Intelligence: Health Care Challenges. 2024.
NZIER report. NBER (2024) 5-10% productivity estimate applied to KFF FY2024 state Medicaid expenditure data.
CDC. National Diabetes Statistics Report. Atlanta, GA: US Department of Health and Human Services; 2026. Accessed May 2026.
CMS. Medicare and Medicaid Statistical Supplement. 2013.
The underlying analysis uses Hakkarainen KM et al. PLoS One. 2012;7(3):e33236. Budnitz DS et al. N Engl J Med. 2011;365(21):2002-2012. Budnitz DS et al. JAMA. 2021;326(13):1299-1309.
Ten-year savings ranges apply NBER (2024) productivity estimate of 5-10 percent cost reduction to state Medicaid expenditure data. State Medicaid expenditure figures sourced from Kaiser Family Foundation (May 2025). Medication safety projections based on CDC prevalence data, CMS hospitalization costs, and published adverse event rates. No proprietary data used.