Expectations for AI were high coming into 2025: In January, MGMA Stat respondents ranked AI tools as the top technology priority this year. That set the bar for us to expect measurable gains in documentation speed, call-center efficiency, coding accuracy, and other areas of practice management. 

This week, we get a sense of that progress: A Sept. 30, 2025, MGMA Stat poll finds that nearly 7 in 10 (68%) medical groups reported adding or expanding use of AI tools in 2025, while 31% did not and another 2% were unsure. The poll had 351 applicable responses. 

Leaders who added or expanded AI in 2025 overwhelmingly targeted clinical work — especially ambient documentation, scribing/dictation, provider note-taking, and charting. The next most common use cases clustered around scheduling and patient communications (reminders, call-center/phone trees, message routing), followed by coding/RCM (risk coding, denials/collections, prior auth) and analytics/reporting.  

Smaller but recurring themes from this week’s respondents included HR screening/compliance, administrative ops/meeting summaries, records retrieval/auto-indexing, virtual care assistants, and quality/HEDIS audits. A handful noted EHR-embedded add-ons and pilot tools; niche mentions touched marketing and other one-offs. 

How AI holdouts are feeling 

Among practices that did not add or expand AI in 2025, many said “no” for now, citing cost, small-practice fit, unclear productivity gains, EHR incompatibility, learning curve, or simply not knowing where to start for that hesitation.  

However, a sizable group among this week’s respondents signaled plans to move ahead in the next year. The most common targets were revenue cycle work (e.g., coding/claims, denials, eligibility/benefits, prior auth), followed by clinical documentation/notes and scribing. Several also mentioned patient communications (reminders/FAQs), clerical support, and some scheduling. 

How we got here 

Around this time last year, most medical groups (53%) reported they had not added or expanded AI tools in 2024, setting the baseline for this year’s activity. Comments clustered around budget limits, planning runways, needing ROI proof, data privacy and security concerns, and staff bandwidth. 

Specifically in the realm of the patient-facing front door, only 19% of groups reported using chatbots or virtual assistants in April, with most deployments handling items such as reminders, FAQ triage, and digital intake — elements frequently embedded in patient portals or phone trees.  

By August 2025, polling revealed that 71% of practices reported using some form of AI in patient visits, though the perceived workload impacts were limited: About 44% of practices using AI said it had not reduced workload yet, and nearly half (47%) said AI was used for only 25% or less of all patient visits, though those reporting heavier utilization also noted stronger gains. 

AI use cases in ambulatory care today 

Clinical documentation and scribing 

Ambient listening tools draft notes in real time, clean up dictation, and slot content into templates, and early gains should show up in provider minutes reclaimed and faster note finalization. Your policies governing this use of AI should spell out attestation steps, PHI handling, and when to revert to conventional documentation. 

Coding and CDI 

AI models have emerged to suggest codes, flag gaps, and highlight audit targets; some estimate denial risk and organize worklists. Still, human-in-the-loop checkpoints remain essential for edits, final code selection, and education feedback loops. Many practice leaders look for specific solutions that are adjacent to or embedded within their existing RCM platforms. 

RCM and prior auth/denials 

The most common use cases seen in ambulatory care are in eligibility checks, auth packet prep, status polling, denial prediction, and prioritized queues. To gauge your progress, keep tracking A/R days, first-pass yield, and cost-to-collect to verify benefit and guide scale decisions. 

Patient communications, scheduling, and access 

Chatbots and virtual assistants handle reminders, scheduling, and FAQs with safe escalation to staff and consistent tone. Scheduling AI solutions have emerged to assist call centers with slot matching, referral routing, and rules-based optimization, targeting no-show reduction and improved time-to-third-next. 

Adoption lags for familiar reasons as those noted by this week’s respondents: EHR integration and routing complexity, uncertain ROI, governance and privacy reviews, and brand-voice control for automated responses. Leaders nonetheless see near-term wins where the value is easiest to measure: after-hours access to capture appointments, no-show reduction via automated reminders and rescheduling prompts, and call-deflection that shortens hold times and raises your first-contact resolution. To start, consider one or two high-volume use cases, pair with clear service-level targets, and track call abandonment, time-to-third-next, no-show rate, and patient satisfaction to validate scaling up. 

Clinical decision support and imaging 

Two tracks dominate here: FDA-cleared narrow algorithms for detection/triage and generative prompts that aid summarization or next-step framing. Define scope of use, documentation requirements, and explainability standards before go-live. 

Analytics/population health management 

Risk panels, targeted outreach, and equity-aware views are improving where data pipelines are stable. Tools that remediate data quality — deduplication, normalization — expand the value of existing dashboards. 

Compliance, privacy and security 

Strong BAAs, HIPAA/SOC 2 verification, and — where applicable — FDA status reviews set the baseline. Add audit logs for prompts and outputs, retention policies, and model/data governance documentation. 

Governance and guardrails 

Stand up a lightweight AI council; formalize use-case intake; require a vendor checklist (interoperability/FHIR, HIPAA/SOC 2, FDA status, security), risk/bias review, and workforce communications/training.  

Keep humans in the loop for note sign-off, coding/audits, and PA justifications. To track progress, keep a close eye on provider time reclaimed/visit, note finalization time, rework rates, denial rate, first-pass yield, call abandonment, scheduling lead times, and patient satisfaction. 

6 practical takeaways for adding or expanding AI: Lessons from #MGMALeaders 2025 

In their session “Evaluating GenAI in Healthcare and Where to Start,” a trio of speakers – Jon Wang, MS, founder and co-CEO of Assort Health; Aamer Hayat, chief administrative officer of Northern California Retina Vitreous Associates; and Dr. Bryce Lokey, chief medical officer and director of operations for Cedar Point Health; explored the realities of embracing new AI technologies today: 

1. Start where the pain is (and measure the healing) 

Begin with one high-volume, repetitive administrative workflow — scheduling/phones, refills, reminders — and baseline the KPIs you’ll move (answer rate, hold time, recovered bookings, staff hours). Run a time-boxed pilot and publish quick wins before scaling. 

2. Insist on try-before-you-buy.  

Pilots change minds. Lokey said physicians went from reluctant to reliant once they could test ambient documentation: “Please don’t take it away from me,” he recalled them saying. “I see two more patients today. I can leave on time. All my notes are done.” Offer hands-on trials and avoid productivity quotas during rollout.  

3. Protect the patient experience with easy opt-outs.  

Defaulting to AI on phones is fine only if a human is one step away. As Wang put it, “if a patient asks for a human, we give them a human.” Maintain simple escape hatches, then use AI where it shines (e.g., triage, rules-based scheduling, and after-hours access).  

4. Redeploy people, don’t replace them.  

Frame AI as teammate, not a threat. Panelists described moving call-center staff into higher-touch work while AI absorbed repetitive tasks; leaders rounded on teams, named the agentic AI (“Iris”), and involved staff in tool design to build buy-in. Stage deployments (e.g., one department, then centralized phones, then satellite offices) and tune triage rules before going system-wide.  

5. Do real vendor due diligence.  

Run head-to-head performance checks, verify uptime, and require live EHR/phone integrations and referenceable outcomes. Lokey’s contracting tip: “watch the fine print in contractual languages” and avoid lock-ins or forced seat escalators. 

What to watch heading into 2026 

There is a broad range of vendors promising AI solutions. As major EHR developers begin to integrate tools into their platforms or turn them into EHR-native features, it may accelerate the pace of vendor consolidation. Vendors who were part of pilots just a few years ago may disappear from the landscape. Keep this in mind as you approach your RFPs. 

Whether you are adopting new solutions or scaling what you’re using today, have a formal plan for governance of tools, understanding of what level of integration is needed for effective deployment, and the user training to make change management easy. Small investments of time here can accelerate your time to realizing the benefits of these new technologies and minimize failed rollouts.