Estimated reading time: 6 minutes · Last updated:
Healthcare AI returns are arriving more slowly than hospital leaders hoped. Bain's US Front Line of Healthcare survey, fielded in June 2026, found that more than 60% of hospital administrators put between 5% and 25% of their IT budgets into AI. Only 40% can clearly measure what that money returns. About 15% report cost savings of 10% to 20%, and most have seen less. The consultancy's authors trace the shortfall to three causes: tools added onto old processes, weak data foundations, and unresolved problems of trust, accountability and clinician involvement. The gap between ambition and results is wide, because nearly two-thirds of administrators still expect meaningful, quantifiable returns within three years.
Key takeaways
- Spending outpaces measurement: More than 60% of hospital administrators in Bain's June 2026 survey devote 5% to 25% of IT spend to AI, but only 40% can clearly measure the return.
- Savings are concentrated: Roughly 15% of administrators report AI cost savings of 10% to 20%, while most report gains under 10%.
- Redesign is rare: Just 10% of provider organisations have rebuilt core workflows end-to-end around AI, while more than half are adding tools for isolated problems.
- Data and scaling hold systems back: Only 15% of clinicians say they can access the data needed for optimal care, and nearly half of organisations scaling AI find it slower and harder than expected.
Table of contents
- Key takeaways
- Hospitals Are Spending Heavily on AI but Few Can Prove the Payoff
- Why Add-On Tools Deliver Less Than Workflow Redesign
- How Data Gaps Undercut AI Before It Reaches the Bedside
- Trust and Accountability Decide Whether Pilots Can Scale
- Will hospital AI start paying off?
- What to be careful about
- Frequently asked questions
Hospitals Are Spending Heavily on AI but Few Can Prove the Payoff
Bain's survey of 100 hospital administrators, conducted in June 2026, shows a serious financial commitment. More than 60% put between 5% and 25% of their IT budget into AI. Expectations are high too: nearly two-thirds believe their AI programmes will produce meaningful, quantifiable returns within three years.
Measurement is where the confidence thins. Just 40% of administrators say they can clearly measure the return on that spending. Those who track it rely mainly on three yardsticks: cost, quality and productivity.
The results that have been measured are uneven. Approximately 15% of administrators report cost savings of between 10% and 20% from AI work, while the majority report gains below 10%. Bain's chart of these figures excludes the 4% of respondents who said they did not know. A small group is seeing double-digit savings, and most systems are still waiting for the payoff they budgeted for.
Why Add-On Tools Deliver Less Than Workflow Redesign
Bain's authors identify the first obstacle as a habit of adoption. More than half of administrators say their organisations deploy individual AI tools aimed at isolated problems. Only 10% have rebuilt core workflows from start to finish with AI built into them.
Fewer still have tied those workflow changes to workforce planning, the decisions about how many staff a system needs and what each person does. One plausible reading of the numbers is that a tool which saves a clinician a few minutes changes little on the balance sheet unless staffing and capacity are planned around it.
The authors argue that leading health systems will sort work into three groups in an AI-native setting: tasks that need human judgment, tasks that improve when people and AI work together, and tasks that AI should lead. They acknowledge that some roles will be reshaped or will no longer be needed, and they say staff will need transparency, open communication and ongoing support through the shift.
How Data Gaps Undercut AI Before It Reaches the Bedside
The second constraint is data. Among the 400 physicians and 100 nurses Bain surveyed, only 15% say they can reach the data they need to deliver optimal care. The remaining majority, about 85%, report that a range of data gaps limits their ability to do so.
Closing those gaps is not yet a funded priority in most organisations. Fewer than one-third have a technology roadmap for the problem that is both clearly defined and backed by resources. The authors conclude that most new AI capabilities are being placed on top of unstable data foundations.
That matters because an AI system can only work with the records it is given. Software that summarises a patient history or flags a change in condition depends on complete, reachable information. Where clinicians themselves cannot find what they need, tools built on the same systems inherit the same blind spots.
Trust and Accountability Decide Whether Pilots Can Scale
Clinicians asked what most stops AI from delivering value on the front line name three problems. They doubt whether AI is accurate and reliable, they are unsure who answers for it when something goes wrong, and they say clinicians have too little say in decisions about AI. Bain's authors rank governance and clinician engagement alongside workflow redesign and clean data as conditions for success.
These barriers surface when organisations try to move beyond experiments. Nearly half of those actively scaling AI say the process has been slower and harder than they expected. The authors argue that each barrier must be dealt with before pilots and proofs of concept can become system-wide deployments.
The sequence explains the difficulty. A pilot can succeed with a small, motivated team that already trusts a tool. Scaling asks far more clinicians to depend on it every day, and that is precisely where open questions about accuracy and accountability start to block progress.
| Finding | Share | Respondents |
|---|---|---|
| Devote 5%-25% of IT spend to AI | More than 60% | 100 administrators |
| Can clearly measure AI ROI | 40% | 100 administrators |
| Report 10%-20% cost savings | About 15% | 100 administrators |
| Expect quantifiable returns within three years | Nearly two-thirds | 100 administrators |
| Redesigned core workflows end-to-end | 10% | 100 administrators |
| Have access to data for optimal care | 15% | 400 physicians, 100 nurses |
Will hospital AI start paying off?
The case for
- About 15% of administrators already report savings of 10% to 20%, showing that double-digit returns are achievable.
- Nearly two-thirds of administrators expect quantifiable returns within three years, which signals continued budget commitment.
- Organisations already track cost, quality and productivity, giving them a basis to extend measurement beyond the current 40%.
The case against
- Only 10% have redesigned core workflows, so most deployments remain add-ons that Bain's authors link to thin returns.
- Fewer than one-third of organisations have a resourced roadmap to fix the data gaps 85% of clinicians report.
- Nearly half of organisations scaling AI find it slower and harder than expected, and clinician distrust of accuracy and accountability remains unresolved.
What to be careful about
- The findings rest on modest samples of 100 administrators and 500 clinicians, so the percentages carry meaningful uncertainty.
- Cost savings are self-reported by administrators, and only 40% say they can clearly measure returns, which limits how far the savings figures can be trusted.
- Bain sells advisory services on the kind of workflow and organisational redesign its authors recommend, which readers should weigh when assessing the prescriptions.
The bottom line
The survey's message is less that AI fails in hospitals than that most health systems have not yet done the work that would let it pay. Spending is substantial and expectations are high, but only 10% have rebuilt core workflows, fewer than one-third have a funded plan for data gaps, and clinicians still question accuracy and accountability. The near-term test is whether the nearly two-thirds of administrators expecting quantifiable returns within three years begin measuring them, since today only 40% can. Readers should weigh the findings with their origin in mind. Bain advises health systems on exactly this kind of transformation, and its samples of 100 administrators and 500 clinicians are modest.
What to watch
- Watch whether more administrators can clearly measure AI returns than the 40% recorded in June 2026; Bain has not said when it will next field the survey.
- Watch for health systems publishing end-to-end workflow redesigns that include workforce planning, the step only 10% have taken; no date has been set.
- Watch whether organisations fund roadmaps to close clinical data gaps, which fewer than one-third had in place; no deadline has been announced.
Frequently asked questions
How much are hospitals spending on AI?
More than 60% of the 100 hospital administrators Bain surveyed in June 2026 allocate between 5% and 25% of their IT spend to AI. Bain did not publish dollar totals for that spending.
Why are hospitals struggling to measure AI ROI?
Only 40% of administrators say they can clearly measure returns, and most organisations use AI tools for isolated problems rather than redesigned workflows. Where AI is not tied to workflow and workforce planning, savings are harder to isolate and track against cost, quality and productivity.
What stops clinicians from trusting AI?
Clinicians in Bain's survey of 400 physicians and 100 nurses cite doubts about AI accuracy and reliability, unclear accountability when errors occur, and limited clinician involvement in AI decisions. Bain's authors treat fixing governance and engagement as necessary for moving beyond pilots.
Related reading