For enterprise IT readers, the useful question is what happens between enabling an AI feature and making it dependable. Microsoft’s support documentation supplies an important qualification: automatic impression generation does not keep updating as a radiologist edits the report. That makes template design and review practices part of the implementation, not administrative housekeeping.
A phased reporting migration
Intermountain expects better login experiences, cloud speech recognition, expanded medical dictionaries, and streamlined reporting. Microsoft also describes Steinberg Diagnostic Medical Imaging, a recently acquired provider with 12 Southern Nevada locations, moving from PowerScribe 360 to PowerScribe One. Its IT director, Josh Staples, reports participation in Microsoft preview programs. Those experiences should not be treated as systemwide results.
Microsoft Learn identifies PowerScribe One as a front-end speech-recognition platform for hospitals, academic centers, and imaging centers. Its documented architecture includes Azure-based infrastructure supporting remote reporting and a PowerScribe-specific Dragon Medical One speech pod on Azure. These are platform characteristics, not a detailed description of Intermountain’s deployment.
The distinction matters: choosing a common platform establishes a direction, but does not by itself demonstrate consistent performance across every site.
Smart Impression adoption is encouraging—not a benchmark
In Microsoft’s account, Kenneth Nowers, MD, Intermountain’s medical director of imaging operations, reports approximately 75 percent active Smart Impression use after most of the radiology enterprise had access for roughly six months. He attributes reduced cognitive burden and improved report quality to the feature. The account provides no defined usage denominator, controlled quality comparison, or quantified time savings.
That is an encouraging adoption signal, but adoption and clinical effectiveness answer different questions. A useful evaluation would examine both whether clinicians use the feature and how much review or correction its drafts require. This is an assessment framework, not a claim about results Intermountain has measured.
The workflow detail administrators should not miss
Microsoft Support describes Smart Impression as generating draft impressions and recommendations from clinical content and dictation style. Radiologists must review the output for accuracy. Automatic mode generates once; subsequent changes to the report body require manual regeneration. Generated content can also replace existing text in recognized Impression and Recommendation fields.
For PowerScribe One environments using this feature, Microsoft documents the following practical sequence:
- Keep findings before the Impression section, with findings and impressions separated.
- Place any Recommendation field after the Impression field.
- Generate the draft using the Impression(M) button, the report editor’s Generate Impression context-menu option, the voice command “Generate Impression,” or Ctrl+Shift+1.
- Review and edit the draft; manually regenerate after relevant body changes when using automatic mode.
Microsoft warns that combined findings-and-impression sections are unsupported. A Recommendation field appearing before the Impression field can cause generation to fail. Successful generation therefore means a draft appears for review—not that the report is ready to finalize.
The implementation lesson is straightforward: test representative templates and late-edit scenarios before expanding use. A polished draft should never become a shortcut around checking whether it reflects the final findings.
Dragon Copilot remains a future direction here
Intermountain’s stated priorities for future Dragon Copilot radiology capabilities include prior-report summarization, contextual clinical chat, and report optimization. Microsoft’s account presents these as anticipated capabilities at Intermountain, not an already deployed radiology workflow or a dated rollout commitment.
The strongest takeaway is therefore narrower than an AI transformation victory lap: Intermountain has chosen a reporting platform, reports substantial use of one embedded drafting feature, and continues its migration. For IT teams considering similar work, the most actionable contribution is the documented operational boundary—templates must be compatible, drafts need review, and automatic generation does not mean automatic synchronization.