Key Takeaways
- Published EMR-plus-AI cases center on drafting documents such as discharge summaries and creating records from conversation (voice charting). In every case the AI drafts and a physician or staff member reviews.
- Nagoya Medical Center reported that in a pilot the average time to write one discharge summary fell from 28 to 8 minutes (Fujitsu Japan, November 19, 2025). The annual saving of 50 million yen or more is an estimate that assumes a physician's hourly wage of 10,000 yen.
- The figures mix measured values, estimates and projections. The clinic voice-charting figure of 6 minutes per visit is an estimate derived from recording length, not a measured value.
- To adopt this at your own clinic, measure current recording time first, then compare after a few weeks of trial use, including the time spent reviewing.
Before adopting AI in an electronic medical record (EMR), most clinic directors and administrators want numbers from institutions already using it more than a feature list. This article takes three cases announced officially by vendors and medical institutions and sorts out what each number actually measured. It then explains how a clinic can use them in its own decision.
This article is general information based on publicly available information as of November 1, 2026. The figures are values announced by each company and institution and do not guarantee results.
The Big Picture
The cases we confirmed fall into two kinds of AI use.
| Case | Announced by / date | Setting | How AI is used | Main published figure | Nature of the figure |
|---|---|---|---|---|---|
| Nagoya Medical Center | Fujitsu Japan, Nov 19, 2025 | National Hospital Organization hospital (full operation in all departments) | Drafts discharge summaries from clinical records | Time per discharge summary: average 28 to 8 minutes | Pilot result; cost saving is an estimate |
| Seikokai Hospital (56 beds) | Henry, Jun 29, 2026 | 56-bed hospital (trial Apr 20 to Jun 5, 2026) | Voice input plus generation of records and documents | About 386 hours a month saved across all roles | Estimate (pre-AI from EMR logs and interviews; post-AI measured during the trial) |
| Clinic voice charting | Atomic Software, Aug 3, 2026 | Participating clinics (early access from May 2026) | Records the consultation and creates the chart entry | 6 minutes per visit; 24–36 hours a month equivalent | Estimated from recording length versus manual recording time |
What they share is that the AI does not finalize the record. It drafts, and a person reviews.
Case 1: Drafting Discharge Summaries with Generative AI (Nagoya Medical Center)
On November 19, 2025, Fujitsu Japan announced that a generative-AI document drafting service had gone into full operation for discharge summaries at the National Hospital Organization Nagoya Medical Center. Operation began on October 31, 2025, across all departments.
- Background: The center writes about 16,000 discharge summaries a year, selecting and transcribing information from the EMR, which was a heavy burden on physicians.
- Result: In a pilot in December 2024 in several departments including orthopedics, the average time per patient fell from 28 to 8 minutes, a reduction of more than 70 percent (20 minutes saved, about 71 percent).
- Cost: The release expects annual savings of about 50 million yen or more, but this is an estimate assuming a physician's hourly wage of 10,000 yen.
- Safety: Because the range of clinical information given to the AI can be selected individually, the course of treatment in other departments can also be grasped, which the center confirmed was useful for medical safety. The service is cloud-based but uses a closed network over a dedicated line; the company states that clinical data is not used to train the generative AI and is not stored in the cloud. It also supports HL7 FHIR.
Discharge summaries belong to institutions with inpatient beds, but the idea of using chart contents to draft a purpose-built document is the same as drafting referral letters or certificates at a clinic.
Case 2: Redesigning Record Work with Voice Input and Document AI (Seikokai Hospital)
On June 29, 2026, Henry announced a trial with Seikokai Hospital (56 beds, Kadoma City, Osaka) that built generative AI into a cloud-based EMR to rework record keeping. The trial ran from April 20 to June 5, 2026.
- Workflow: Staff dictate within the chart screen, and the AI creates documents from the chart data and audio. Staff review and approve every AI-generated record before it is finalized.
- Ward records: The old workflow had five steps: observing the patient, paper notes, paper notes of vitals, returning to the nurses' station, and transcribing into the chart. The new one has two: record the bedside conversation, then enter vitals while transcription runs. A record takes about 1 to 2 minutes, and the paper notes and transcription steps disappear.
- Projected savings: About 386 hours a month across all roles (about 188 hours from voice-input records and about 198 hours from document AI such as summaries).
- Basis of the figures: The monthly savings are estimates. Pre-AI times come from EMR logs and staff interviews (April 2026); post-AI times were measured during the trial and include time for confirmation and minor correction. The release notes that results may differ depending on facility size, specialty and staffing.
Case 3: Voice Charting at Clinics (Atomic Software)
For clinics, Atomic Software announced its "AI Voice Chart" on August 3, 2026. Early access began in May 2026, and the number of participating clinics is growing.
- Published figure: For clinics that use it actively, a reduction of 24–36 hours a month in chart-recording work, or 6 minutes per visit.
- How it was calculated: From the system, 1,115 recordings by participating clinics between July 1 and July 20, 2026 were taken, and the time a person would have needed to write the chart by hand was estimated from recording length. It is therefore an estimate, not a measurement. (Multiplying 6 minutes by 240–360 visits a month gives 24–36 hours; this is our own arithmetic.)
- Voices from the field: The release includes comments from physicians at several clinics. A dermatology clinic physician said it is now enough to "check the content and make corrections," greatly reducing the burden on staff. Review of the AI output by a person is again the premise.
Reading the Numbers: Five Checks
These figures are not your clinic's results. Before relying on them, check the following.
| Check | What to look at | Example from these cases |
|---|---|---|
| 1. What was measured | Writing time, consultation time or overtime | All three center on recording and document-creation time, not consultation time |
| 2. Measured or estimated | A recorded value or a converted one | The clinic's 6 minutes is estimated from recording length; Seikokai's 386 hours is an estimate |
| 3. Is review time included | Whether time to correct AI output is counted | Seikokai states that confirmation and minor correction are included |
| 4. Scale and specialty | Hospital or clinic; which specialty | Discharge summaries are for hospitals with beds |
| 5. Who announced it | Third-party verification or the vendor's release | For these three, no third-party verification is stated in the announcements |
How a Clinic Can Test This at Its Own Practice
- Measure current time: For one to two weeks before adoption, record how long charting and document creation take. Use EMR logs if available.
- Limit the scope: Start with one or two uses, such as visit records, referral letters or certificate drafts.
- Compare including review time: Compare before and after, counting the time spent reading and correcting the AI output.
- Decide who reviews: Fix in advance who reviews and finalizes. The AI drafts; responsibility for the record stays with the medical institution.
- Check data handling: Confirm in the contract and specifications how clinical data is passed to external generative AI and whether it is used for training or storage. See Security Checkpoints for AI Voice Charting and Using Generative AI at Medical Institutions for detailed checklists.
FAQ
Q. Does consultation time get shorter too? What these three announcements publish is recording and document-creation time. None of the announcements covered here gives a figure for shorter consultations themselves.
Q. Is this irrelevant for a clinic without discharge summaries? No. Even without inpatients, there are many documents built from chart contents, such as referral letters, certificates and information letters. The idea of the AI drafting is the same.
Conclusion
- Published EMR-plus-AI cases center on drafting documents and creating records from conversation.
- Nagoya Medical Center reports an average 28-to-8-minute drop per discharge summary, Seikokai Hospital projects about 386 hours a month saved across all roles, and a clinic voice-charting vendor estimates about 6 minutes saved per visit.
- The figures mix measurements, estimates and projections. Always check how they were measured and whether review time is included.
- At your own clinic, measure current time and compare after a few weeks of trial use, including review time.
For an overview of AI functions, see What Is AI in Electronic Medical Records? and AI Features of an AI-Native EMR; for comparing products, see Comparing AI EMRs for Clinics and Comparing Voice-Input EMRs; for how summarization works, see What Is AI Summarization?.
Talk to Pottech
Pottech provides "AI Karte," an EMR for clinics designed around AI. If you are considering how to bring AI into your clinic's record work, including how to run a trial and how to confirm data handling, please contact us.
Frequently Asked Questions
- Q. How much time can AI in an EMR actually save?
- A. In the published cases, discharge summary writing at Nagoya Medical Center fell from an average of 28 to 8 minutes (pilot), and clinic voice charting saved about 6 minutes per visit (estimated from recording length). Because the measurement methods and settings differ, they cannot be applied to your clinic as they stand. A trial at your own clinic is the reliable way to find out.
- Q. Can the AI-generated record be used as it is?
- A. In every published case the AI drafts and a physician or staff member reviews and corrects. In Seikokai Hospital's announcement, the post-AI times include confirmation and minor correction, and staff review and approve every record before it is finalized.
- Q. Have third parties verified the numbers in these cases?
- A. Within the three official announcements covered here, no third-party verification is stated. Read the figures as values announced by the vendors and medical institutions.
- Q. Can a clinic use AI in the same way as for discharge summaries?
- A. Discharge summaries are a document for hospitals with inpatient beds. For a clinic without beds, the main uses are voice charting that creates records from the consultation, drafts of referral letters and certificates, and summaries of intake questionnaires.
References
- Fujitsu Japan, press release "Fujitsu Japan introduces generative-AI medical document drafting service at Nagoya Medical Center and begins full operation for discharge summaries" (November 19, 2025)
- Henry, Inc., press release on PR TIMES, "Seikokai Hospital (56 beds) redesigns record work to be AI-first with Henry AI" (June 29, 2026)
- Atomic Software, Inc., press release on PR TIMES, "AI Voice Chart under the medicalforce intelligence AI suite cuts up to 36 hours of work a month; two new AI features also announced" (August 3, 2026)