Health systems are putting AI into production faster than they are fixing the data it runs on.
Moving ahead
91%
are increasing investment in AI readiness and data quality
82%
have an AI governance committee
Left behind
27%
call data integrity a top C-suite priority
18%
formally monitor data quality
18%
are confident in the metadata behind their AI
0%
have formal metadata management
Percent of IT executives surveyed. Source: e4health’s 2026 AI Readiness & Data Governance Survey.
73% of IT executives surveyed say AI is already in production at their organizations. Only 27% say data integrity is a top C-suite priority. The distance between those two numbers is the readiness gap.
Why it matters
AI reads, codes, and bills from the patient record. If the patient identity, terminology, and metadata behind that record are unreliable, AI scales the errors along with the efficiency.
How we know: e4health surveyed IT executives, all CHIME members, on AI readiness and data governance. A Focus Group of healthcare executives, all CHIME members, was held on Oct. 27, 2026. Below is what the numbers said, followed by what we took away.
Who we surveyed
73%CIOs
27%VPs or directors
Organizations
Community hospitals (36%), health systems (18%), ambulatory care organizations (18%), plus academic, mental health, and community health organizations
Size
45% under 100 beds, 27% with 100 to 599, 27% with 600 or more
AI is past the pilot stage, and it is pointed at revenue
The pilot phase is over for most. Nearly three in four have AI in production today. Another 18% are running pilots, and 9% are still exploring use cases.
Where organizations stand on AI adoption
Percent of IT executives surveyed, from least to most deployed.
9%exploring use cases
18%running pilots
64%in limited production
9%deployed enterprise-wide
Where AI is deployed today
Percent of IT executives surveyed who say each use case is currently deployed.
Administrative automation
64%
Clinical documentation
55%
Clinical decision support
27%
Also deployed: predictive analytics and contact center (18% each), population health and cybersecurity (9% each).
73%
name the revenue cycle as their primary focus for AI, led by billing (45%), patient access (18%), and coding (9%).
91%
are increasing investment in AI readiness and data quality over the next 12 months. More than a quarter say significantly.
The stakes: AI is being pointed at the work that gets hospitals paid. Bad data there does not stay a data problem. It turns into denials, rework, and missed revenue.
“AI is only as good as the foundation it’s built upon.”
Rich Amelio, president, e4health
The committee is seated. The foundation under it is not.
82% have an AI governance committee. Far fewer have the everyday capabilities that tell a committee whether its data can be trusted.
Governance capabilities formally established
Percent of IT executives surveyed. 18% reported none of these.
Oversight
AI governance committee
82%
Everyday data work
Master data management
27%
Data quality monitoring
18%
78%
of organizations with an AI governance committee do not formally monitor data quality.
9%
rate their data governance program highly effective. Another 55% call theirs moderately effective.
36%
have no data governance program in place, or one they rate not effective.
Between the lines: A committee can approve a use case. It cannot vouch for data that nobody is measuring.
Confidence drops the closer you get to the foundation
Leaders trust the record more than what sits behind it. 64% are confident in their EHR data. Just 36% are confident in the MPI that tells AI which patient a record belongs to, and 18% are confident in their metadata.
How confident leaders are, layer by layer
Percent of IT executives surveyed. The number shows those who are confident or extremely confident.
Confident or extremely confident
Neutral
Minimally or not at all confident
PHI is adequately governed for AI use
64%
AI outputs can be traced back to source data
55%
MPI data can support AI
36%
Controls over third-party AI vendors are appropriate
27%
Metadata is sufficient to understand AI data sources
18%
Yes, but: 55% are confident they can trace an AI output back to its source data. Only a third of that group is also confident it has the metadata to understand those sources.
The data problems are basic, and leaders know it
Nothing on the list is exotic. Asked which data quality problems hurt AI initiatives most, leaders picked the unglamorous ones.
Data quality challenges that most affect AI
Percent of IT executives surveyed who ranked each challenge in their top three.
Inconsistent terminology
64%
Also named: duplicate records and limited data lineage (18% each).
45%
rank data quality first among the governance capabilities that most need improvement, more than any other capability.
One question, one theme
“If you could solve one data challenge before expanding AI, what would it be?”
Improve quality and consistency
Solving for mismatching terminology
Clean data
Data quality and consistency
Data quality
Data cleanliness
Cleanup
Central data
Governance
78% of those who answered said some version of the same thing: clean it up. Responses are lightly edited for spelling and capitalization.
What is in the way is not the technology
It is attention, people, and money. Most leaders say data integrity has not reached the C-suite agenda, and the barriers they name are about capacity.
Is data integrity a top C-suite priority?
Percent of IT executives surveyed. Integrity includes completeness, accuracy, and cleanliness.
27%say yes
55%say no
18%are unsure
Biggest barriers to effective AI governance
Percent of IT executives surveyed who ranked each barrier in their top three.
Organizational culture
45%
Also named: regulatory uncertainty, privacy concerns, and technology limitations (27% each); executive alignment and vendor transparency (18% each).
60%
of those who named a single biggest obstacle to broader AI adoption pointed to time, staff, budget, or competing projects.
“Cultural change and balancing between those who think it solves everything and those who think it will end the world.”
Survey respondent, on the single biggest obstacle to broader AI adoption
Beyond the numbers: six things we took away
We have talked these results through with health system leaders. Six themes stood out.
The tools arrived before the strategy
Many organizations bought AI in pieces and are only now deciding what to keep. When several tools read the same chart, each one needs the same source, pulled the same way, and a correction in one place has to reach the others.
Governance covers the model better than the data
Executive review, privacy rules, bias checks, and human sign-off are becoming standard. Asking whether the data underneath is good enough to deploy on is not there yet.
Bad data starts where it is entered
Notes finished hours after the visit, or documentation that captures a fraction of what describes the patient, flow straight into reports and models. Leaders increasingly see data quality as an organizational habit, not an IT project.
Identity cleanup stalls after go-live
Duplicate rates get driven to an acceptable level for an EHR conversion, and then the push stops. A 3% threshold was meant as a starting point, not the standard. Best practice is under 1%.
Revenue, not fear, gets it funded
Headlines and worst-case scenarios no longer move budget. Showing that cleaner data means more revenue captured and more patients served does.
A blank check would not buy a bigger model
Asked where unlimited money would go, leaders point to cleaner capture at the front end, real-time data they control, and a workforce that knows how to use AI.
Almost no one plans to do this alone
Most leaders would bring in help. The two most-named jobs are designing data governance and automating document processing.
Where leaders would consider an external partner
Percent of IT executives surveyed who would consider a partner for each activity. They could choose more than one.
Data governance design
55%
Document processing and automation (IDP)
55%
Data quality remediation
27%
82%
would consider an outside partner for at least one AI-readiness activity. The other 18% selected none.
Our take: three moves before the next deployment
Give the committee something to measure
Most organizations already have an AI governance committee. Put data quality on its agenda and HIM at its table: who owns the data, how it is monitored, and what “good enough to deploy on” means.
Start with identity and terminology
They are the least exciting line items and the first things a model trips over. Manage duplicate records toward 1%, not 3%, and measure your data against a standard such as USCDI.
Do not let cleanup end at go-live
Staffing and budget lead the barriers, so the work tends to stop when the conversion does. Start with a quick audit of a sample of records, borrow capacity for the heavy lift, and keep monitoring so the data stays clean.
e4health helps health systems protect the integrity of patient data with Health IT, HIM, coding, CDI, data integrity, and intelligent automation services. That includes eMPI cleanup, clinical data abstraction, and intelligent document processing through
e4core, our AI-enabled HIM platform.
How ready is your data for AI?
Bring us the messy part: duplicate records, mismatched terminology, legacy data. We will help you find where to start.
Talk with e4health
About this research. Percentages are shares of the 11 CIOs and IT leaders, all CHIME members, who answered e4health’s 2026 AI Readiness & Data Governance Survey, rounded to the nearest whole number. The findings are directional, not statistical. A Focus Group of healthcare executives, all CHIME members, was held on Oct. 27, 2026. The takeaways in this post are e4health’s own.