About 34% of medical groups have used AI to build an app or workflow tool in-house, according to MGMA Stat's September 22, 2026 poll of 228 practice leaders. Another 8% tried and did not launch, and 54% have built nothing.
The MGMA poll results
MGMA Stat asked leaders whether their group had used AI to build its own app or workflow tool. The poll drew 228 applicable responses and MGMA published it on September 23, 2026. Here is how the answers split.
| Answer | Share of 228 respondents |
|---|---|
| Yes, built a tool | 34% |
| Tried, did not launch | 8% |
| No | 54% |
| Unsure | 10% |
Source: MGMA Stat, "Would you DIY an EHR with AI?", September 23, 2026. Rounded figures as published, so the column sums to more than 100%.
Add the first two rows and 42% of groups have made an attempt. By that count, 19.0% of the groups that tried (8% of 42%) did not get a tool into use. That ratio is our own arithmetic on MGMA's rounded figures.
What groups built
MGMA sorted the tools that respondents described into two piles. The lower-risk pile covers scheduling and appointment management, internal dashboards, inventory management, policy and workflow documentation, call handling and recalls, and staff tracking.
The higher-risk pile is charting support, biologic-ordering workflows, coding and audit guidance, and patient-facing tools. These touch clinical records or patient contact, so a mistake reaches a patient or a claim.
MGMA did not report how many of the 34% built something in the second pile. The poll shows what people built, not how often. A group that wants to copy a peer should start with the first pile, where a wrong output shows up on an internal report and not in a chart.
Most groups have no written AI rules
An in-house tool is still an AI tool, and it needs a policy. MGMA Stat's January 20, 2026 poll of 328 leaders found that 20% of groups had an AI governance policy and 22% were developing one. Some 56% had none, and 2% were unsure.
The gap was wider a year earlier. MGMA and Humana research from fall and winter 2024 found that 73% of surveyed organizations had no formal governance structure for AI use.
Put the polls side by side. In the September 2026 poll, 34% of groups built a tool. In January 2026, 20% had a policy. In August 2025, 71% already used AI in patient visits. The samples differ, so the numbers do not describe the same groups. They do show that use and building run ahead of rule writing.
How much AI is already in daily use
MGMA Stat's August 5, 2025 poll of 244 leaders found that 71% used AI in some way during patient visits, and 29% did not. Measured against all 244 respondents, 47% used AI in 25% of visits or fewer. Another 10% used it in 26% to 50% of visits, 7% in 51% to 75%, and 7% in 76% or more. Adding the last three rows, 24% of all respondents used AI in more than a quarter of their visits.
The same poll asked whether AI reduced staff workload. Of the groups using AI, 44% saw no reduction, 39% saw one, and 17% were unsure. MGMA reported that groups using AI in more than 25% of visits were more likely to report a workload benefit.
Light use gave light results. Nearly half of all groups sat in the lowest band, and 44% of AI users saw no workload change. A home-built tool that touches a small share of visits has little room to change the workload figure.
What a data breach costs when something goes wrong
Any tool that reads patient data carries breach exposure. IBM's 2025 Cost of a Data Breach report, as summarized by the HIPAA Journal, put the average healthcare breach at $7,420,000. That was down from $9,770,000 in 2024, a drop of $2,350,000, and healthcare still had the highest cost of any industry.
Healthcare breaches also took the longest to find and contain. IBM measured 279 days for healthcare against a global average of 241 days, which is 38 days longer. IBM also reported that 13% of organizations had a security incident involving an AI model or application.
A small practice will not see a $7,420,000 bill, because that figure is an average across large and small organizations. It does show the scale of the risk that a policy and an access review are meant to reduce.
Numbers to check before building
Start by placing the tool in the lower-risk or higher-risk pile from the MGMA list. Then confirm the group has a written AI policy, since 56% of leaders in the January 2026 poll had none. Last, decide how usage will be counted. The 2025 poll linked a workload benefit to use in more than 25% of visits, and only 24% of respondents were above that line.
The September poll had 228 responses, so one respondent moves a result by about 0.4 points. The January poll had 328 responses and the August 2025 poll had 244. All three are MGMA member polls, not random samples of every practice, so treat them as a read on engaged practice leaders.
Sources
- MGMA Stat, Would you DIY an EHR with AI?, September 23, 2026
- MGMA Stat, AI governance in medical group practices, January 21, 2026
- MGMA Stat, Most practices use some form of AI, but is it reducing staff workloads, August 6, 2025
- HIPAA Journal, Average Cost of a Healthcare Data Breach Falls to $7.42 Million, 2025
Talk to us
Fifteen minutes about your practice and the systems you run.